Report No. 45111-YF Serbia Baseline Survey on Cost and Efficiency in Primary Health Care Centers before Provider Payment Reforms January 26, 2009 Human Development Sector Unit Europe and Central Asia Document of the World Bank SERBIA BaselineSurvey PHC Centers CONTENTS Page ACKN0WLEDGEMENTS ............................................................................................... i EXECUTIVE SUMMARY .............................................................................................. ii Chapter 1. INTRODUCTION ...................................................................................... 1 1.1 Background on Health Sector Reforms................................................................. 1 1.2 Experience with Payment Reforms....................................................................... 4 Chapter2 . DATA AND METHODOLOGY .............................................................. 6 2.1 Data on Primary Health Care ................................................................................ 6 2.2 Outcome Measures ................................................................................................ 7 2.3 Analytical Methods ............................................................................................... 8 Chapter 3. RESULTSPRIMARY HEALTHCARE ............................................... 10 3.1 Descriptive Analysis............................................................................................ 10 3.1.1 General characteristics o f DZs................................................................... 10 3.1.2 Expenditures inDZs ................................................................................. -12 3.1.3 Inputuse ofDZs ......................................................................................... 13 3.1.4 Outputs produced by DZs.......................................................................... 15 3.1.5 Revenue sources o f DZs ............................................................................ 17 3.1.6 Productivity o f DZs................................................................................... -17 3.2 Summary o f Econometric Results....................................................................... 18 3.3 Summary o f Baseline Performance Measures..................................................... 18 Chapter 4. CONCLUSIONS AND RECOMMENDATIONS ................................. 21 REFERENCES ................................................................................................................ 26 ANNEX ............................................................................................................................. 30 Annex 1: Technical Annex: ........................................................................................... 3 0 1) Reviewo f Studieson ProductionFunctions and Cost Functions...................30 2) Econometric Methods Usedinthis Analysis .................................................. 35 Results Production Function ..................................................................................... 39 3) Results from Econometric Analysis ............................................................... 38 Results Cost Function.,.............................................................................................. 41 4) Model BuildingProcess.................................................................................. 43 Annex Table 1: List of PHC Centers Surveyed. January . 2007.................61 December Annex Table 2: Variables inthe PHC Facility Survey. January -December 2007.....64 Annex Table 3: Staffing inPrimary Health Care Centers. January - December2007 -65 Annex Table 4: Questionnaire for Primary Health Care Centers.................................. 66 Figures Figure 1.1: Hospital beds per 100.000 pop.......................................................................... 2 Figure 1.2: Hospital admits per 100pop ............................................................................. 2 Figure 1.3: Average length of stay ...................................................................................... 2 Figure 1.4: THE inYOof GDP ............................................................................................. 2 Figure A.1: Ranking of DZs by Production Efficiency Score........................................... 41 Figure A.2: Ranking of DZs by Cost Efficiency Score..................................................... 43 Figure A.3: Ranking of DZs by Production Efficiency Score across Different Model Specifications.................................................................................................................... -53 Figure A.4: Ranking of DZs by Production Efficiency Score across DifferentUnits of Analysis (Service Point) .................................................................................................... 54 Figure A.5: Ranking of DZs by Cost Efficiency Score across Different Model Specifications.,................................................................................................................... 60 Tables Table 1.1: Public facilities providinghealth services. in2008............................................ 1 Table 2.1: DZ Performance MeasuresAffected by Capitation Payment System ...............8 Table 2.2: Descriptive variables and expectedresults of production analysis.................... 9 Table 2.3: Descriptive variables and expected results incost analysis............................... 9 Table 3.1: Description of DZ Characteristics.................................................................... 11 Table 3.2: DZ Annual Expenditure on Inputs in2007. in% o f total expenditures ...........12 Table 3.3: Distribution o f Staff inDZs. inpercent of staff category ................................ 14 Table 3.4: Large equipment inworking order. in % o f DZ category ................................ 14 Table 3.5: Utilization o f DZ Space, inpercent o f total square meter................................ 15 Table 3.6: Output o f DZs, number o f services .................................................................. 16 Table 3.7: DZ Referrals, number o f referrals and inpercent of total visits....................... 16 Table 3.8: DZ Sources o f Revenue, inpercent o f total revenues...................................... 17 Table 3.9: DZ Productivity, number o f visits inDZs ........................................................ 18 Table 3.10. Baseline Values o f DZ Performance Measures............................................. -19 Table A.1: Estimation o f the Stochastic Frontier Production Function ............................ 40 Table A.2: Determinants o f Inefficiency inDZ Production.............................................. 40 Table A.3: Estimation o f the Stochastic Frontier Cost Function....................................... 42 Table A.4: Determinants o f DZ Cost Inefficiency ............................................................ 42 Table A.5: Summary o f Alternative Models and Specifications Estimated...................... 44 Table A.6. Likelihood Ratio Test Statistics for Alternative Specifications o f the Traditional Production Function........................................................................................ 46 Table A.7. Likelihood Ratio Test Statistics for Alternative Specifications o f the Stochastic ProductionFrontier Model............................................................................... 49 Table A.8. Estimates o f Inefficiency as a Share of Total Variance Across Alternative SF Production Model Specifications ....................................................................................... 52 Table A.9. Likelihood Ratio Test Statistics for Alternative Specifications o fthe Traditional Cost Function.................................................................................................. 55 Table A.10. Likelihood Ratio Test Statistics for Alternative Specifications of the Stochastic Frontier Cost Function .......,.......,...........,....... ......,.....,.................I..............I....57 ACKNOWLEDGEMENTS This report was preparedby ateam headedby Pia Schneider (Senior Economist, ECSHD), and comprising Cheryl Cashin (Consultant), Johannes Koettl (Junior Professional Officer, ECSHD), Ana Djordjevic (Consultant, World Bank Belgrade), and the researchteam working with CESID ledby PredragDjukic and Milos Mojsilovic. Cheryl Cashin was the main author, wrote the methodology, conducted and wrote the descriptive and econometric analysis and conclusions. The CESID researcherswere responsible for data collection, data entry and management and organizing the final workshop. Johannes Koettl provided researchassistance, and Ana Djordjevic and Hermina Vukovic logistical support inBelgrade. Sreypov Tep and Viktoria Lebedeva were responsible for the document processing.Pia Schneider developedthe study design and finalized the report. Detailed comments and suggestions at various stages of preparation were provided by the Minister of Health of Serbia H.E. Tomica Milosavljevid and the Deputy Minister of Health Elizabet Pauvnovic; and by Abdo Yazbeck, Tamar Manuelyan Atinc, Armin Fidler, Julian Schweitzer and MukeshChawla (all World Bank). Thanks go to Adam Wagstaff, Lead Economist (DEC) and Jack Langenbrunner, Lead Economist (EASHD), who provided helpfulcomments on an earlier draft; and to Randy Ellis, Professor o f Economics, Boston University who was the peer reviewer of this final report. This study was sponsoredby the World Bank health sector strategy implementation program administered by MukeshChawla Sector Manager HDHNP. The team gratefully acknowledges the collaboration provided by the Ministry of Health, the Institute of Public Health, and the Health Insurance Fund of the Government of Serbia. The team expresses its special gratitude to the staff working inall Primary Health Care facilities and hospitals who have participated inthis survey and dedicatedtheir time to fill in the questionnaires. Valuable feedback to the preliminary findings of this report was provided by the participants of a workshop inBelgrade on September 25,2008. i EXECUTIVE SUMMARY 1. The Serbian Ministryo f Health (MOH) andthe Health Insurance Fund(HIF)are planning to change the provider payment from currently a line-item budgetto (i) capitation payment inPrimary Health Care (PHC) centers, and (ii) form o f some case-based payment such as diagnosis-related groups (DRGs) inhospitals. With this payment change the Government aims to set incentives to providers that will lead to a more efficient provision o f care and contribute to the sector's financial sustainability. 2. The purpose o fthis study is to conduct a baseline survey on the cost and efficiency in Primary Health Care (PHC) Centers (Dom Zdravlja - DZ) in Serbia before the implementationo f the payment reforms. Data were collected in 147 DZs (see Annex Table 1). Results are used (i)to inform the payment reform and (ii)toestablishabaselineonhealthsectorperformanceincludingutilization, quality, cost and efficiency against which the impact o f the reforms can be assessedina follow-up survey. Recommendations about the technical payment system design and capitation formula are beyond the scope o fthis report and have beenundertaken as a separate activity. This study was conducted with the support o f World Bank health sector strategy funds'. 3. The current line-itembudgetspaid from the HIF to DZs and hospitals are based on the number o f staff who are allocated to health facilities according to their number of beds. This creates an incentive to providers to use more staff and beds, but does not rewardbetter productivity, quality o f care, or health outcomes. Under capitation the DZs will bepaid, inadvance, a pre-determined fixed rate to providea defined set o f services for each individual enrolled with the DZ for a fixed period o f time. Capitation sets incentives to improve efficiency through reduced input use per patient, more output achieved with fewer inputs (e.g. more visits per physician), combining inputsmore effectively (e.g. shifting some expenditures from staff and utilities to medicines and supplies), increasing preventive services, and providing fewer diagnostic services. 4. Under case-basedpayments such as DRGs,hospitals will bepaidthe average cost o f producing a "case" in an average hospital, which may be adjusted to account for regional economic conditions, and include indirect costs such as teaching and capital cost. A shift from line-item budgets to case-basedpayment inhospitals i s expected to lead to more inpatient admissions, shorter average lengtho f stay and higher patient turnover per bed, which may also increase hospital expenditures for the HIF. 5. During the past years, several donors have assisted the M O H and the HIF in provider payment and structural reforms. The World Bank, European Agency for Reconstruction (EAR), and the Red Cross inKraljevo have proposed different 'Healthy Development:The World Bank strategy for HNP results. April 24, 2007. .. 11 capitation formulae for PHC. The MOH has not decided yet on the capitation formula and i s currently investing inthe availability o f patient data, legal changes and institutional reforms, and the population registrationwith their preferredPHC providers. In2007/8, the M O H with World Bank support has started preparatory work for DRG costing in six pilot hospitals to estimate risk-distributions and costs. Future efforts will focus on information technologies, data management and analysis inhospitals and inthe HIF, monitoring, evaluation and fine-tuningo f the case-mix and DRGrate. The MOHhas conducted ahumanresources strategy (MOH Serbia, 2005) and a health sector restructuring strategy (Sanigest, 2007). The human resource strategy guidedthe 9.5% reduction inthe healthworkforce from 2004 until2007. The M O H has closed 1,835 hospital beds since 2004; and based on the restructuring strategy plans to close additional hospital beds by 2010. With the support o fthe EAR andthe World Bank, the MOH has strengthened management inhealth facilities to ensure that directors use their increased responsibility in decentralized health facilities to adjust their input factors such as staff and equipment*. The additional value added through this baseline survey i s to provide a comprehensive analysis o f the performance in PHC centers, against which any future reforms that affect provider behavior can be assessed. 6. The methodology usedinthe baseline survey includes descriptive analysis of key performance measures inPHC centers, as well as an econometric analysis o f the current production and cost functions inPHC centers. The analysis aims to provide insight into the current level o f efficiency as well as the determinants o f the factors that influence efficiency. A technical annex contains the econometric literature review, methodology, analysis, results and describes the model building process. 7. The main finding from this baseline survey is that DZs differ substantially intheir efficiency. Although DZs are generally working with the same level o f staff, medical equipment and space, which are largely dictated by the system, they produce different levels of output such as consultations etc. To some extent the level o f productivity inDZs may be affected by the age/gender structure o f the population, particularly by the number o f children inthe DZ catchment area. There is very little variation inthe cost-efficiency o f DZs, because DZ expenditures are largely pre-determined as prices o f input factors (e.g. wages) are defined on a national level. 8. Additional findings show that expenditures inDZs are dominated by personnel costs (70% of total cost). This i s at the expense of medicines and supplies, which are also needed to improve the scope and quality of DZ services. DZs are currently very constrained by their fixed costs and thus intheir ability to improve cost efficiency, as their personnel costs are determined externally by the system. Ifpersonnel costs areexcludedfrom capitationandthe HIFcontinues to pay for 2In addition, two recent Bank reports on decentralizationand on the experiencewith provider payment reforms in Europe have informed the planned health sector reforms (World Bank, 2007; Schneider, 2007). iii staff based on a line-item budget, then only about 30 percent o f total DZ costs can be managed by DZs under capitation. 9. There are large areas of unusedspace inDZs that are reducing DZ productivity. Half o f DZs have at least some large equipment but relatively few diagnostic tests appear to be performed. Thus, the productivity o f DZ staff, equipment and space can be improvedby usingequipmentmore often or reducing the number o f equipment, and reducing non-clinical space inDZs. 10. DZ mainly produce curative visits, andthere appears to be an excessive number o f laboratory tests and injections; whereas relatively few preventive care services are provided. Referral rates for DZs are generally low which may be relatedto a similar low severity case-mix across DZs, but the current data do not allow determiningwhether patients are being referred more or less thannecessary. DZs could become more productive by either increasingthe number o f consultations and other outputs, or reducingthe number o f staff without reducing the total numberof visits. 11. There is currently unequal allocation ofpublic resources for primary care across DZs. This i s likely to be the result o fthe way the line-item budget is now defined based on the number o f staff and other factors. This inequality can be addressed by the capitation payment system if funding from the HIF, municipalities, and other government sources are pooled at a higher level and distributedto DZs through a unifiedcapitated rate per enrolled individual. 12. Once capitation has been introduced, it may be expected that DZs will provide more preventivecare visits to patients to reduce the need for more expensive diagnostic and curative visits; inaddition, unnecessary laboratory tests and injections may also be reduced. Also, capitation may lead to higher referral rates to hospitals, as DZs have the incentive to reduce their cost, and hospitals paid by DRGshave an incentive to hospitalize more patients. Therefore, additional measures may be needed to make sure that capitation leads to prevent adverse effects on quality and access to care and on hospital spending. 13. The Government, incollaboration with health sector partners, has already started implementing several measures to prepare the sector for the planned provider payment reform, including: (a) The Government i s currently reviewingproposals to adjust the capitation rate by coefficients for age and gender o fthe enrolled population, andto include additional incentives for preventive services, such as a bonus payment to DZs who achieve an agreed level for childhood immunization or maternal care. Also, the Government i s considering geographic adjustment coefficients in the capitation payment to adjust for higher utility costs inDZs inmountainous areas. iv (b) The M O Hhas started a review of clinical guidelinesto ensure that these are compatible with the scope o f services that will be financed by capitation, and provide appropriate guidance to staff on laboratory tests, injections, other procedures, and referrals. (c) The M O H and HIF incollaboration with donors have provided extensive management support to DZ managers such that they will be able to successfully respondto the incentives set by the new per capita payment system. This collaboration is ongoing and includes management training and new accounting systems inall DZs. (d) The Government is undertaking major investmentinthe data systems inDZs and HIF incollaboration with the World Bank loan and the EAR. The HIF has started to improve reporting and analysis o f key data relatedto DZ performance, including population size and demographic structure, services provided, and resource use. 14. Financial incentives set by the capitation payment system may not be enough to trigger a behavioral change among DZsthat leads to more efficient care. Thus, other supporting policy changes and improved data and information systems may be requiredto realize the benefits o fthe newpayment system. Based on the conclusion from this baseline analysis, several additional steps could support the development o f capitation payment inDZs, prevent adverse effects inreaction to the financial incentives set by the payment, and improve the efficiency o fthe sector. These measures could be implementedina phased approach with Phase I focusing on the following seven steps. Payment System Design (a) Pool PHC funds from the HIF and other public sources, and pay all DZs a unified capitation rate with appropriate adjustmentsfor cost variations with the objective o f allocating primary care resource equally across DZ. (b) Include salaries inthe capitation amount and adjust related humanresources policies and laws to give more flexibility to DZsto improve their cost efficiency. (c) Specify inreferral guidelines the scope o f services at the PHC level and appropriate referral pattern to preventunnecessary referrals to higher-cost hospitals. Inaddition, the capitation system should include measures, such as open enrollment, a quality monitoring system, or outcome-based bonuses and penalties for unjustified referrals. (d) While the Serbian health insurance law already sets the legal frame for co- payments, the government may consider developinga revisedcost-sharing policy as part o f overall provider payment reforms. V ManagementIssues (e) Assess the regulations and constraints that may affect the ability o f DZs to manage their resources more efficiently. This will include regulations affecting the scope o f service at different levels o f care, public procurement laws, and the public labor law according to which public sector employees in DZs are still on the HIFpayroll and planned centrally. (0 Conduct anassessmentofessential medicines that are necessary for effective PHC, and consider the potential for limited financing o f essential medicines withinthe context o f capitation. (g) Develop a cost-effective package o fmedical equipment that shouldbe available at the PHC level. Consider reducing the number o f equipments in DZs inthe vicinity of a hospital where patients could be referred. Examine whether more basic equipment is available that may contribute more to DZ productivity (e.g. blood pressure cuffs and scales) 15. During Phase 11,additional attention could be givento the following three measures that arise from the conclusion o f this analysis: ManagementIssues (a) Consider reorganizing space inDZs, movingto smaller buildings, or redirecting excess space to other purposes. DZs could rent out non-clinical space to purposes such as private doctors or dentists, or to day-care centers for individuals who need supervision such as elderly or disabled individuals. (b) Hospital payment reforms such as DRGs stimulate changes inhospital care such as shorter hospital stays that will be felt inother parts o f the health care system. As a result, DZs and community care as well as long-term care departments will have to be ready to provide a greater degree o f follow-up to patients who have been discharged from hospitals. Data Systems (c) Collect information on the disease profiles o f the populations served, outcomes, and overall access to essential medicines inthe country. Develop quality and outcome measures that can be monitored by DZs themselves for internal management purposes, and at the system level by the HIF and MOH. vi CHAPTER 1. INTRODUCTION 1.1 BACKGROUND HEALTH ON SECTOR REFORMS 1. The Serbian Ministry o f Health (MOH) and the Health Insurance Fund (HIF) are planning to change the provider payment from currently a line-item budget to (i) capitation payment inPrimary Health Care (PHC) centers, and (ii) form o f case- some based payment such as diagnosis-related groups (DRGs) in hospitals. The purpose o f this study is to conduct a baseline survey on the cost and efficiency inPrimary Health Care (PHC) Centers in Serbia before the implementation o f the payment reforms. Results can be used to inform the payment reform and to establish a baseline on health sector performance including utilization, quality, cost and efficiency against which the impact o f the reforms can be assessed in a follow-up survey. Recommendations about payment system design and capitation formula are beyond the scope of this report and have been undertaken as a separate activity. This study was conducted with the support o f World Bank health sector strategy funds3. 2. Serbian PHC Centers (Dom Zdravlja DZ) are organized either as a separate entity or - as part of the secondary care hospital - Zdravsteni centri. DZ provide basic primary services4. Patients in need for specialized primary care are referred to one o f the 19 specialized centers (Zavodi) or to one o f the 120 hospitals (Table 1.1). As part o f the decentralizationstrategy, all DZ are becoming independent from hospitals. Table 1.1: Public facilitiesprovidinghealth services, in2008 Facility type Number of facilities Dom Zdravlja (Primary Health Care Center) 159 Zavodi (Specialized health center) 19 General Hospital (Opsta Bolnica) 37 Specialized Hospital (Specijalna Bolnica) 14 Single Specialty Clinic 23 Multi Specialty Institute 38 Clinical Hospital Center (Klinika Bolnica Centar) 5 Clinical Center (Klinika Centar) 3 3 Healthy Development: The World Bank strategy for HNP results. April 24, 2007. Includingpreventivehealth care, emergencycare, general medicine, women and child health, visiting nurse services, laboratory and other diagnostics. When other services are lacking in the area, the DZ also provides:dental care, occupational medicine, physical medicine, rehabilitation, ambulance transportation. When the catchment area for a DZ is more than 20,000 people and it is locatedmore than 20 km from a General Hospital,the DZ also providesspecialist services for: internalmedicine, pneumo-physiology, ophthalmology, otolaryngology,psychiatryand cardiology. 1 3. In 2006, Serbia had relatively more hospital beds (Figure 1,1), lower inpatient admissionrates (Figure 1.2) and longer average lengthsof hospitalstays (ALOS) than neighboringcountries (Figure 1.3). However, the MOHhas taken severalmeasures to reduce bed numbers and shorten ALOS inhospitals. Total health expenditure (THE) of 8% of GDP in20055was inline with neighboringcountries (Figure 1.4). Figure 1.1: Hospital beds per 100,000 pop Figure 1.2: Hospital admits per 100 pop 20 15 10 5 0 1 I ' Source WHO 2006. Source: WHO 2006. http://www.euro.who.int/hfadb http://www.euro.who.intlhfadb Figure 1.3: Average lengthof stay Figure 1.4: THE in YOof GDP ! 1 :t8 12 10 6 68 4 4 2 2 0 0 I I ' J Source: WHO 2006. Source: WHO 2005. http://www.euro.who.int/hfadb http://w .euro,who.int/hfadb 4. The Serbian Government plans to change the provider payment system to set incentives to providers that will lead to a more efficient provision of care. The current line-itembudgets are based on the number of staff who are allocatedto health facilities according to their number of beds. This creates an incentive to providers to use more staff and beds, that define their budget, but does not rewardimprovement in productivity,quality of care, or healthoutcomes. 5. Under capitation the payment to a DZ i s not linked to the inputs used or the volume o f services provided. Rather, the DZ will be paid, in advance, a pre-determinedfixed 5 MOH Serbia: Development ofNational Health Accounts in Serbia- Phase 111.August 31,2007 2 rate to provide a defined set o f services for each individual enrolled with the DZ for a fixed period o f time. Therefore, some financial risk i s shifted from the HIF to the DZ. Ifthe DZ expenditures are greater than the capitation budget, it will be liable for this difference. If there are efficiency gains and costs are lower than the capitation budget, the DZ can retain and reinvest this surplus in the provision o f care which would be expected to lead to better health outcomes. Thus, capitation sets incentives to improve efficiency through reduced input use per patient, more output achieved with fewer inputs (e.g. more visits per physician), combining inputs more effectively (e.g. shifting some expenditures from staff and utilities to medicines and supplies), increasing preventive services, and providing fewer diagnostic services. This change in treatment behavior is expected to lead to a more efficient input mix (i.e. staff, drugs), fewer inputs and increased productivity, more efficient output mix, better quality o f care, and better health (Cashin et al. 2007). 6. Per capita payment may also create incentives for unintended consequences. For example, there may be an incentive for DZs with capitation payment to under-provide services and keep costs low or refer patients to specialists who are paid fee-for- service and hospitals. These negative outcomes may be prevented to some extent, with additional checks andbalances inthe system, such as open enrollment in DZs, a quality monitoring system, or outcome-based bonuses for better quality compliance, or even penalties if DZs skimp care on patients. Therefore, capitation rates are generally adjusted for age and gender o f the population registered with the PHC center, and geographic criteria. Some capitation formulas include quality-based components (e.g. cancer screening rates, immunization rates) to set incentives to providers to improve quality o f care and prevent adverse effects such as under- provision o f care. 7. Under case-based payments such as DRGs, hospitals are paid the average cost o f producing a "case" in an average hospital, which may be adjusted to account for regional economic conditions, and include indirect costs such as teaching and capital cost. A shift from line-item budgetsto case-basedpayment in hospitals i s expected to lead to more inpatient admissions, shorter average length o f stay and higher patient turnover per bed, which may also lead to higher hospital expenditures for the HIF. However, as capitation i s expected to improve access to PHC services, it may be expected that hospitalization rates, particularly for conditions that can be preventedor managed at the DZ, will decline (Kozak et al, 2001). 8. During the past years, several donors have assisted the M O H and the HIF inprovider payment and structural reforms. The World Bank, European Agency for Reconstruction (EAR), and the Red Cross in Kraljevo have proposed different capitation formulae for PHC. The M O H has not decided yet on the capitation formula and i s currently investinginthe availability o f patient data, legal changes and institutional reforms, and the population registration with their preferred PHC providers. In 2007/8, the MOH with World Bank support has started preparatory 3 work for DRG costing in six pilot hospitals to estimate risk-distributions and costs. Future efforts will focus on information technologies, data management and analysis in hospitals and in the HIF, monitoring, evaluation and fine-tuning of the case-mix and DRG rate. The World Bank supported the MOH in conducting a human resources strategy (MOH Serbia, 2005) and a health sector restructuring strategy (Sanigest, 2007). The human resource strategy supported a 9.5% reduction in the health workforce from 2004 until 2007. The MOH has closed 1,835 hospital beds since 2004; and based on the restructuring strategy plans to close additional hospital beds by 2010. With the support of the EAR and the World Bank, the MOH has strengthened management in health facilities to ensure that directors use their increased responsibility in decentralized health facilities to adjust their input factors suchas staff and equipment. Inaddition, two recent Bank reports on decentralization and on the experience with provider payment reforms in Europe have informed the planned health sector reforms (World Bank, 2007; Schneider, 2007). 1.2 EXPERIENCEWITH PAYMENTREFORMS 9. The experience from other countries with introducing capitation payment inPHC and DRGs in hospitals (Schneider, 2007) is being used in this analysis to describe how provider behavioral indicators presented in Table 2.1 may change in Serbia in response to the payment change. Corrective interventions by the MOH will be proposed to prevent that any expected behavioral change by providers in response to the financial incentive set by the payment system, will have a negative effect on the provision and quality of care. 10. Generally, provider payment reform tends to be undertaken in response to structural problems in the health sector that require a major reorientation of overall financing and service delivery. For example, in 2002 New Zealand introduced PHC reforms, which included the formation of new non-profit PHC entities funded through a per capita payment system, in order to address health disparities across socioeconomic groups that arose from the fee-for-service payment system (Hefford, 2005; MOH New Zealand 2001). In Costa Rica the PHC sector was reorganized into autonomous cooperatives paid by capitation in the early 1990s to address declining quality of services, low morale among providers, and long waiting lists for diagnostic and other services (Clark, 2002; Gauri et al, 2004). In the former Soviet Union, the historical neglect of the PHC sector, over-specialized and fragmented care, an unsustainable hospital infrastructure, as well as little emphasis on prevention and health promotion brought about unprecedenteddeclines in health status throughout the region early in the post-Soviet transition period (WHO, 2001). Several post-Soviet countries embarked on comprehensive health financing and service delivery reforms, with the restructuring and strengthening of PHC, supported by capitation payment (Borowitz et al, 1999; ZdravReform Program, 2000). 4 11. Provider payment reforms are rarely implemented in isolation, so it i s difficult or impossible to attribute changes in provider or system performance to the impact of the payment reform. Some studies do show that moving from a fee-for-service (FFS) payment system to capitation can have significant effects on costs and output. For example, when Ireland switched from FFS to a capitation system, physician visits declined by an estimated 20 percent. Moving from a line-itembudget to capitation, as is planned for Serbia, i s likely to have the greatest impact on efjciency through the flexibility that providers will gain to change their mix of inputs. However, increasing provider flexibility will require legal changes. Some evidence from former Soviet Central Asian republics shows that moving from line item budgets to capitation motivated providers to decrease expenditures from staff and utilities and invest in medicines and supplies, enabling PHC providers to increase the scope and quality of their services (Cashin et al, 2007). The increased flexibility also allows PHC providers to change their output mix to emphasize preventive services, which may reduce costs for them by reducing follow-up visits. When a per capita payment system was combined with a quality monitoring system, PHC providers in the Karaganda region of Kazakhstan shifted their services toward preventive services resulting in a 30 percent increase in prevention visits between 2001 and 2004. This shift was accompanied by a decline in the rate of potentially avoidable hospital admissions for ulcers, asthma, and anemia (Cashin et al, 2007). 12. Case-based payments in hospitals such as DRGs generally result in more admissions and a shorter ALOS. During the past 20 years, the number of hospitalizations increased markedly in countries with case-based payment, while it remained on a similar low level in Spain, Canada and the Netherlands, where physicians are paid a monthly salary independentof their workload. Hospitals may also have an incentive to admit a patient who could be treated more efficiently in a DZ or day-surgery setting. In these cases, DRGs may conflict with overall expenditure controls by setting an incentive to increase the number of hospitalized cases, resulting in higher hospital expenditures (Docteur et al. 2003). After the DRG payment was implemented in the U S Medicare system, ALOS in hospitals fell by 15% in the first three years; and fell as much as 24% for some diagnoses (Cashin, et al. 2005). 13. The rest of this report is organized is follows. Chapter 2 presents the data and methodology used in this survey to evaluate the cost and efficiency performance in DZs. Results are presentedand discussed inChapter 3. Based on findings, Chapter 4 concludes and proposes several reform measures to support the effect of the provider payment reform. The Annex contains additional information including a technical annex with an overview on the literature on cost and efficiency analysis, the econometric analysis, and the model building process; a list of health facilities included in the survey, summary results, and the questionnaire used to collect information. 5 CHAPTER 2. DATAAND METHODOLOGY 2.1 DATAON PRIMARY HEALTH CARE 14. The M O H decided to survey all DZs as this would provide a unique data base to establish a new inventory o f DZs following the recent organizational changes caused by the separation o f DZs from hospitals under decentralization. The data for this analysis come from a baseline survey o f revenues, expenditures, staffing patterns, and service provision in 147 o f 159 PHC centers in Serbia (see Annex Table 1). The survey does not include the 9 DZs in Kosovo6 and the 3 DZs who submitted incomplete questionnaires as they were undergoing organizational changes. A follow-up survey is expected to be conducted after the implementation o f the payment reforms inDZs. 15. Questionnaires collected existing routine data from DZs based on monthly patient and facility registers. The questionnaires were developed by representatives from the MOH, CESID and the World Bank team and pilot-tested in February 2008 in three DZs7. The M O H sent questionnaires to all directors of DZs in Serbia, and made reminder phone calls. Data collection took place over a six-week period in May-June 2008 and covers a time period o f twelve months from January - December 2007. Questionnaires were completed by staff in DZs in collaboration with field interviewers working for an independent local research firm CESID in Belgrade. Interviewers were trained, and visited all DZs at least once with follow-up visits to ensure data validity and completeness. CESID entered and cleaned data in Excel, which was transferred into STATA for analysis. Specific variables collected in the baseline survey are listed inAnnex Table 2. The unit o f analysis i s the DZ. 16. Although the research team followed a consistent data collection approach, data quality may still be an issue in this study, and results should be interpreted with caution. The study relies on existing data, which are o f unknown quality, and there i s evidence o f some inconsistencies. For example, 70 percent o f DZs report more than 5 visits per year per person in the catchment population, and 25 percent o f DZs report more than 10 visits per person per year. This clearly raises concerns about inflated output data. In addition, a significant share o f DZs has, what appear to be inflated staff numbers, and for DZs that are still part o f a hospital complex, there was also a problem o f not being able to separate out expenditures and input use by DZs from those o f the parent health facility. These data quality issues raise concerns about the interpretation o f the results o f this study and also highlight the need for better data systems. However, results show that there is sufficient variation in production efficiency levels and convergence on cost efficiency levels to conclude that poor data quality i s not entirely driving the results. Under the UN SecurityCouncil Resolution 1244 (1999), Kosovo is administered by UnitedNationsInterim AdministrationMissionin Kosovo (UNMIK). BaEka Topola, Valjevo and VraEar. 6 2.2 OUTCOME MEASURES 17. There are a variety of methods for assessing the changes in provider performance in response to payment reforms. The experience from other countries shows that capitation payment i s expected to lead to a more efficient health sector. Inthe case o f health care providers, efficiency measurement assesses how well input factors such as labor, space, supplies, machines, and medicines are combined to produce different mixes o f services o f different levels o f quality. For example, if a provider increases quality o f care for the same or fewer resource inputs, it can be said that efficiency has improved. The ultimate output that health care providers can "produce" i s better health outcomes. 18. A set o f performance measures focused on efficiency are analyzed inthis study, both through descriptive analysis and econometric analysis. Efficiency can either be measured from a productiodproductivity perspective, or from a cost perspective. We will focus on measuring efficiency, defined in multiple ways in an attempt to capture these dimensions. These measures and their expected change under the payment reform are summarized in Table 2.1, and will be examined in the follow-up study after the capitation payment system is introduced. In order to accurately compare efficiency, however, the severity o f the patient case-mix treated in DZs and quality o f output (e.g. health care services) should be accounted for, which i s not possible from the data available inthis study. 19. In addition to the descriptive analysis, this study will include a technical annex that uses econometric methods to estimate efficiency inthe production and cost-efficiency inPHC centers based on production and cost frontiers. The production frontier is the maximum output possible with different combinations o f inputs, given the technology and other environmental constraints at the time. Technical efficiency i s achieved when maximum output is produced from a given set of inputs. Suppliers who produce levels o f output below the frontier are not technically efficient. The cost frontier i s the minimumcost that can be achieved to produce a given level o f output at given input prices. A provider is considered cost efficient ifoutput is produced at the minimum achievable cost. Costs may exceed the cost frontier either because the output is below the potential maximum (technical inefficiency) or because the mix o f inputs used is not optimal given current input prices, or a combination of the two (Wagstaff and Barnum 1992). Reasons for technical inefficiency may include poor technology, organizational and managerial inefficiency, or problems with inputs. An organization can also improve efficiency by choosing the output mix that maximizes profits or other desired outcomes (such as health outcomes), or allocative efficiency (Liu, 2000). Knowing whether health care providers are not achieving minimum costs because o f technical inefficiency or a sub-optimal input mix may be important for policy purposes. 7 Table 2.1:DZ PerformanceMeasures Affected by Capitation Payment Category Efficiencyof input use Efficiencyof output mix outputs and reduce inputsper output. Access # of service points per km2 Increase or decrease, dependingon the impact on total input costs. Total # of visits per catchment Providers have both incentive to reduce total population per month outputs, and to increase preventive services. # of preventive visits per catchment Increaseas providers have incentive for population. prevention to avoid more expensive services. Potential Average cost per visit Decrease as inputs become more productive, unintended but may increase as there is an incentive to consequences reducetotal output. % of visits resulting in a referral to a May increase, as providers have incentive to specialist reduce input costs. % of visits resulting in areferral to a May increase, as providers have incentive to hospital reduce input costs. 36 o f visits resulting in a referral to a May increase, as providers have incentive to specialist or hospital reduce input costs 2.3 ANALYTICALMETHODS 20. This baseline study conducts both, descriptive analysis o f key DZ performance measures, as well as an econometric analysis o f the current production and cost 8 functions. The descriptive analysis of DZ production will analyze the following variables describing the level and type of output by DZs, which can be expected to be affected by a change inthe provider payment system (Table 2.2). Variables Expected resulttriggered by capitation payment Total number of visits Total number of visits and services may decrease, since providersare not and/or services compensatedfor increasingoutput Number of visits per The number of visits per enrolled individual may decrease, since registered individual providersare not compensated for increasingoutput. Number of visits per Initially there is an incentivefor fewer overall visits, so the number of physicianper month visits per physicianmay decrease. But capitationalso createsthe incentiveto reduce inputs or shift to lower cost inputs, includingnon- physiciantime. So, the number of visits per physicianper monthmay increase over the longerterm. Percent of visits The mix of outputs may change, shiftingto lower cost outputs. The consultation, share of preventivevisits may increase to reducethe need for more preventive, home visit expensive diagnosticand curativevisits. 21. The descriptive analysis of costs in DZs will analyze the following variables reflecting the level and mix of inputs used by DZs. These variables may be proxy variables for efficient use of inputs and can be expected to be affected by a change in the provider payment system (Table 2.3): Table 2.3: Descriptivevariables and expected resultsin cost analysis Variables IExpected result related after the introduction of capitation % of revenue from patients IfDZs improveefficiency of input use the share of payment from patients may decrease, dependingon the payment definition % of expenditures on Greater flexibility in allocatingexpenditures across inputs may salaries, drugs, other increasethe share of DZ expenditures on drugs and supplies % of staff physician,nurse, Greater flexibility in allocatingexpenditures across inputs may paramedical, admin increase the share of DZ staffexpenditureson nurses and paramedical staff. 22. The econometric analysis in the Technical Annex provides insight into the current level of DZ efficiency relative to the current capability of the entire system, whether levels of DZ efficiency vary, and the determinants of variations inDZ efficiency. 9 CHAPTER 3. RESULTSPRIMARY HEALTH CARE 3.1 DESCRIPTIVE ANALYSIS 23. The descriptive analysis presents findings on the general characteristics of DZs, input use, level of output (services) produced, expenditure, revenue and DZs productivity. The analysis examines differences in inputs (e.g. staff), outputs (e.g. consultations) and productivity across a variety o f characteristics of DZs including being rural or urbanand stand-alone or beingpart of a health center. It may be expected that results in a follow up survey are different if DZs change their treatment behavior, as they respond to the new incentives created by the per capita payment system, or by the related organizational changes. Table 3.10 at the end of this Chapter summarizes the main findings and can be used for comparison with a future survey. 3.1.1 General characteristics of DZs 24. Table 3.1 summarizes the DZ characteristics. The DZs included in this study are evenly split betweenurban and rural (51 and 49 percent, respectively). The majority of DZs are stand-alone clinics (70.8 percent), with the other 29.2 percent beingpart o f health centers. Stand-alone DZs are separate healthcare institutions and are not incorporated in health centers, which generally consist of a hospital and one or more DZs. DZs typically have a central location with a number of outposts. The average number of outposts i s 11.3 per DZ. These service points are staffed by an average o f 8.5 physicians and 25.5 total medical personnel. 25. In terms o f ownership, 68.5 percent have been decentralized and are owned by the municipality, and 28.8 percent are owned by the MOH. At the time o f the baseline, no DZs are privately owned. Most DZs are rural stand-alone clinics owned by the municipality (65.8 percent). 26. The average distance to the nearest hospital i s 27.3 km, with one DZ 200 km from the nearest hospital. The DZs have on average 2.5 outpatient beds and 9.6 inpatient beds, but inpatient beds may have been counted that are actually part o f the DZ's parent hospital. The average DZ catchment area i s 511.7 km2,ranging from 3 to 2,035 km2. The population density averages 381 people per km2,with a reported range of 9 to18,806 inhabitants. Even taking outposts into consideration, some DZs cover a very large territory, which raises concerns about geographic access to PHC in Serbia, particularly in areas where there are no private GPs. For example, 62 DZs (42 percent) have a catchment area greater than 500 km2,and 13 DZs (9 percent) have catchment areas of more than 1,000 km2. The average area per service point i s 84.8 10 km2and exceeds 200 km2for 11 DZs (7.5 percent), with one DZ reporting an area of 1,500 km2per service point. Furthermore, the number of outposts i s not highly correlated with the size o f the catchment area or the catchment population. For example, one DZ with 22 outposts reports a catchment area o f only 7 km2,covering 55,543 people. Another DZ with 20 outposts has a catchment area o f 422 km2, covering 56,011 people. 27. The population served by DZs was reported both as the official catchment area, and as the number o f individuals registered with the DZ. The official catchment area i s more valid at this time, as the process o f registering individuals i s ongoing, and only a small percentage o f individuals had registered as of the time o f the study. The catchment population ranges from 8,228 - 252,131 per DZ. The average population per service point i s 7,248, but the median i s 3,400 individuals and 80 percent o f DZs report a population per service point of less than 6,000. 11 3.1.2 Expenditures in DZs 28. Total annual expenditure of DZs in 2007 averaged 239 million Serbian Dinars (SD), or about U S $4.8 million. This represents a per capita expenditure of 5,576 SD, or US$112. The per capita expenditure varies from 321 to 17,500 SD within DZs catchment area. The average per capita expenditure in the highest spending 25 percent of DZs i s four times that of the lowest spending 25 percent, indicating substantial inequality in the distribution of primary care resources across DZs. Per capita expenditure i s not significantly related to whether the DZ i s urban or rural, or whether it is stand-alone or part of a health center. The average per visit expenditure i s 938 SD, or US$18.8. The average recurrent cost per visit i s 887 SD, or US $17.7, excluding expenditureson investment inequipment or infrastructure. 29. DZs spend 72.6 percent of total expenditures for personnel (Table 3.2). DZs that are part of health centers spend a significantly higher share of their total budget on personnel than stand-alone DZs, 81.4 percent vs. 69.3 percent, respectively.* This difference i s likely because in DZs that are part of a health center, other inputs are provided or subsidized by the health center, and it was not possible to attribute these costs to the DZ. Urban DZs devote a higher share of their budget to staff than rural DZs, but this difference i s not statistically significant. Percent of total Mean per year (range) expenditure on N=143 All DZs Rural Urban Stand-alone In Health Center Personnel 72.6 70.9 74.3 69.3 81.4 (0.27 - 100) (45.4 - 100) (0.27 - 91.5) (0.27 - 100) (59.0 - 91.5) Drugs 10.9 13.1 8.7 13.2 4.7 Supplies 3.2 I 3.1 I 3.3 I 3.3 I 2.8 Utilities 3.5 I 3.4 I 3.6 I 3.3 I 4.0 (0 - 11.5) (0-11.1) (0.21-11.5) (0-11.5) (0.89- 11.1) Transport 2.1 2.4 1.9 2.2 2.0 (0 - 8.4) (0 - 6.3) (0.24 - 8.4) (0 - 8.4) (0.37 - 4.2) Maintenance 3.3 3.1 3.5 3.4 3.1 (0 - 18.3) (0 - 8.8) (0 - 18.3) (0 - 18.3) (0 - 16.1) Investment 4.9 4.6 5.1 5.6 2.9 (0 -5 1S) (0 -22.3) (0 5 1.5) - (0 - 5 1.5) (0 22.3) - Other 4.4 4.1 4.6 5.3 1.9 (0 - 62.6) (0 38.0) - (0 -62.6) (0 - 62.6) (0 - 10.9) *All comparisons of means in the descriptive analysis are made usinga simple two-tailedt-test. 12 30. DZs spend an average of 10.9 percent of their total budgets on drugs, and another 3.2 percent on supplies. Rural DZs allocate 16.1 percent of their expenditures to medicines and supplies, while urban DZs allocate 12.1 percent. This difference is statistically significant at the 10 percent level. Stand-alone DZs allocate 16.5 percent of their expenditures to medicines and supplies, while DZs that are part of a health center allocate only 7.6 percent. Again, this is likely to reflect the difficulty in allocating health center expenditures to the DZs rather than an actual difference in patterns of input use. However, if these DZs benefit from "free drugs" purchased by their parent hospital and they are about to be decentralized to independent DZs (World Bank, 2007), then they should budget for possibly future higher drug expenditures. About 9 percent of DZ expenditures are allocated to other operating costs (utilities, transport, and maintenance), and 4.4 percent to other expenditures. Some DZs report a very high percentage of expenditures as "other," up to 62.6 percent, which indicates that they are not tracking their expenditures carefully. 31. DZs spent 10 million SD on investment per year, or about US$200,000, reflecting 4.9 percentage of total DZ expenditure. Most of the investment (8,275,206 SD) was for equipment, and 1,864,981 SD was for investment in infrastructure. There is no difference in the level of investment between rural and urban DZs, but stand-alone DZs allocate a significantly higher share of total expenditures to investment (5.6 percent) than DZs that are part of a health center (2.9 percent), suggesting that the latter may benefit from cross-subsidies. 3.1.3 Input use of DZs 32. DZs have on average 265 total staff, with 61.8 physicians, 118 nurses, 3.5 paramedical staff, 13.7 administrative staff, and 36.0 technical staff. Per service point, the DZs have on average 35.8 total staff, including 8.5 physicians, 16.6 nurses, 0.43 paramedical staff, 2.7 administrative staff, and 6.6 technical staff. The ratio of population to providers is 195.5 per total staff, 782.4 per physician, and 389.7 per nurse. DZ staff is typically paid based on the public wage regulations through the HIF, but DZs also hire an additional 7 percent of staff from their own funds. Table 3.3 presents a summary of DZ staffing patterns. A detailed description i s in Annex Table 3. 33. There appears to be some imbalance in the distribution of staff across different DZ. On average, 25.4 percent of DZ staff i s physicians, 50.7 percent are nurses, and 1,l percent i s paramedical staff (Table 3.3). The ratio of nurses to physicians i s 2:1. On average, more than 20 percent of the staff is non-medical, with 6.4 percent administrative and 16.5 percent technical. Rural DZs have a lower share of physicians on staff, 24 percent vs. 26.8 percent, which significant. In addition, the ratio of nurses to physicians is higher in rural areas, 2.19 vs. 1.94 in urban areas, which significant. Rural DZs also have a significantly higher share of administrative and technical staff. DZs that are part of health centers have the same share of 13 physicians, but a significantly higher share of nurses than stand-alone DZs, and a significantly higher ratio of nurses to physicians, with an average of 2.20 vs. 2.02 in stand-alone DZs. DZs that are part of a health center have a lower share of administrative and technical staff, as health center staff may carry out many of these functions. staff (0 - 18.9) (0 - 18.9) (0 - 12.4) (2.6 - 18.9) (0 - 11.6) Technical staff 16.5 17.3 15.6 16.8 15.7 (0-35.1) (0 -26.2) (0 - 35.1) (5.5 - 26.2) (0-35.1) 34. Table 3.4 shows more than 44 percent of DZs (64) report that they do not have an ultrasound or an X-ray machine in working order. Only one DZ reported having all of this equipment. Among those that do report having some equipment, the most common machine is an ultrasound, with 52 percent of DZs reporting at least one ultrasound machine, and 30 percent reporting two or more. Nearly half of the DZs have an X-ray machine. Rural DZs have more equipment than urban DZs. Whereas 49.3 percent of urban DZs report having no equipment in working order, only 35.7 percent of rural DZs have no equipment that is in working order. Results on equipment should be interpreted with caution as DZs who are still part of a hospital complex may not have made a distinction between equipment that is physically located inthe hospital and inthe DZ. Equipment in Mean YO working order N=147 All DZs Rural Urban Stand-alone In Health Center YOof DZ that have 43.5 37.5 49.3 42.3 46.5 no ultrasound and no X-ray At least 1 52.4I 58.3 I 46.7I 52.9 1 51.2 ~ IAt least 1 X-ray 46.3 I 48.6I 44.0 I 45.2 I 48.8 machine 14 35. Most o f DZ space i s usedfor non-clinical purposes or consultation rooms (Table 3.5). DZs report an average of 5,797 square meters space, but space per service point averages 913 square meters. All but five DZs have a laboratory, and 66 percent o f DZs have a pharmacy. On average, nearly half of the space in DZs is not used for clinical purposes such as consultation room, laboratory, or pharmacy. Half the DZs devote more than 50 percent of space to non-clinical purposes. Rural DZs have a higher share o f space used for non-clinical purposes than urban DZs; and stand-alone DZshave a higher share than DZs inhealth centers. Space Mean YO N=146 All DZs IRural I Urban IStand-alone 1I n Health Consultation Rooms 43.1 42.8 ~~~ 43.3 41.9 45.8 Laboratory 3.8 3.4 4.1 3.3 4.8 Pharmacy 3.5 4.2 2.9 4.3 1.7 Other (non-clinical) 46.8 49.2 44.5 47.9 44.2 3.1.4 Outputsproducedby DZs 36. The main output produced by a primary health care center i s consultations (Table 3.6). On average, DZs provide 409,194 consultations per year. About 9 percent o f consultations were preventive visits, and DZs report on average 4.4 percent o f total consultations as home visits. The shares o f preventiveand home visits do not vary by DZ characteristics. Overall, emergency services make up only a small share of services provided by DZs, less than 3 percent on average. Stand-alone DZs have a higher share o f emergency visits (3.1 percent) than DZs that are part o f a health center (0.7 percent), but this difference i s not statistically significant. 37. DZs perform a large number o f laboratory tests, with an average o f 1.02 lab tests per consultation. Rural DZs perform fewer lab tests per consultation (0.89) than urban DZs (1.14), but this difference is not statistically significant. DZs that are part of a health center perform significantly more lab tests per consultation (1.45) than stand- alone DZs (0.84). 38. DZs also provide a large number o f injections, with more than half o f all consultations involving an injection. Rural DZs provide more injections per consultation than urban DZs, 0.78 vs. 0.34, which i s weakly significant. Stand-alone DZs have more injections per consultation than DZs that are part of a health center, but this difference is not statistically significant. However, there are very few diagnostic procedures per consultation, with an average of 0.07, which does not differ very significantly by DZ characteristics (Table 3.6). 15 Table 3.6: OutDutof DZs. number of services output AllDZs I Rural Urban I Stand-alone I In Health TOT #coriultations 1 1 1 I I ~ ~~~~~~~409,194 293,946~~ ~ 500,029 390,594 407,502 Total # of Dreventive I 36.545 I 21.793 I 43.887 I 34,341 I 29,979 visits (% o'f total) (9.2%) I (9.0%) I (9.4%) (9.5%) (8.5%) 5,524 1 8,699 I 22,077 II 16,063 II 14,22 1 (% of total) (4.3yo) (4.4%) (42%) (4.8%) (3.0%) Total # of emergency 14,029 5,153 22,549 18,220 3,891 services (2.4) (2.4) (2.4) (3.1) (0.7) (% of total) Total # of laboratory 293,008 183,099 398,521 264,33 1 362,366 tests (1.02) (0.89) (1.14) (0.84) (1.45) (per consultation) Total # of diagnostic 21,360 11,603 30,726 15,467 35,613 procedures (0.07) (0.08) (0.06) (0.06) (0.09) (per consultation) Total # of injections 118,386 117,413 119,306 119,677 115,291 (per consultation) (0.56) (0.78) (0.34) (0.64) (0.38) 39. Referral rates are relatively low among DZs but significantly higher in DZs that are part of a health center (Table 3.7). Overall, 7.1 percent of consultations result in a referral to a specialist, and 5.5 percent result in a referral to a hospital. The total mean referral rate is 12.6 percent, which is reasonable. Rural DZs have a higher rate of referrals to hospitals (6.2 vs. 4.9) and total referrals (13.3 vs. 12.0), although these differences are not statistically significant. DZs that are part of a health center have a significantly higher rate of referrals to specialists than stand-alone DZs (8.9 vs. 6.4), but there is no significant difference in the rate of referrals to hospitals or total referrals. Easy access to specialists in health centers may lead DZ providers that are still part of a hospital complex, to more readily refer their patients. Referral Mean (per year) N=147 All DZs Rural Urban Stand-alone I n Health Center Total # of referrals 19,795 14,664 25,066 17,924 24,3 18 to specialists (7.1) (7.1) (7.1) (6.4) (8.9) (%of total visits) Total # of referrals 17,450 9,5 18 25,066 16,224 20,4 18 to a hospital (5.5) (6.2) (4.9) (5.4) (5.7) (YOoftotal visits) Total # of referrals 37,245 24,181 49,786 34,148 44,735 (YOof total visits) (12.6) (13.3) (12.0) (11.8) (14.5) 16 3.1.5 Revenuesources of DZs 40. The vast majority of DZ revenues from the HIF, 84.2 percent of total revenues on average, with very little variability across characteristics of DZs (Table 3.8). The next most important source of revenue for DZs is from patients, with an average o f 7.6 percent. Only one DZ, however, reports a share of revenue from patients greater than 22 percent. The percentage of revenues from patients is higher for DZs that are part of a health center than for stand-alone DZs (9.1 vs. 7.1). Other sources of revenue include municipalities, the MOH, Ministry of Finance, and donors; though each contributes less than 3 percent to the total. YOof DZ revenue Mean per Year from (range) All DZs IRural I Urban IStand-alone IIn Health Center HIF 85.4 84.2 86.5 84.3 88.4 (0 - 100) (0 - 100) (43.6 - 100) (9.9 - 100) (0 - 97.6) Patients 7.6 7.5 7.7 7.1 9.1 (0 - 100) (0-100) (0-20.1) (0-21.4) (0 - 100) Municipality 2.1 2.6 1.6 2.7 0.53 ' (0 - 87.5) (0 - 87.5) (0 - 14.3) (0 - 87.5) (0 -9.7) Other government 0.95 1.2 0.75 1.3 0.01 sources (0 - 17.9) (0 - 17.9) (0 - 13.0) (0 - 17.9) (0 - 3.3) MOH 0.88 1.5 0.30 1.o 0.60 (0 -44.3) (0 -44.3) (0 - 3.0) (0 -44.3) (0 - 16.4) Donors 0.70 1.1 0.33 0.30 1.8 (0 -42.8) (0 42.8) - (0 - 8.9) (0 - 8.9) (0 -42.8) MOF 0.21 0.20 0.22 0.2 1 0.19 (0 - 9.9) (0 - 9.9) (0 -4.4) (0 -9.9) (0 -4.4) 3.1.6 Productivity of DZs 41. The productivity of DZs is measured in terms of the number of visits per month per physician, per all medical staff, and per individual in the catchment population per year (Table 3.9). Rural and stand-alone DZ report higher productivity. The average number of visits per month per physician is 561.2, and the average number of visits per month per all medical staff is 160.5. The average number of visits per month per catchment population i s 0.70 (8 visits per person per year). This compares to 0.63 (7.6 visits per person per year) in the WHO Europe region. There are significant differences in productivity between rural and urban providers. Rural DZs report 636 visits per month per physician, whereas urban DZs report significantly less, only 488 visits per physicians. There are also more visits per catchment population in rural DZs, 0.81 vs. 0.60 inurban DZs. Stand-alone DZs report an average of 529 visits per physician per month, and DZs that are part of a health center report 488 visits per physician per month. The number of visits per catchment population in stand-alone DZs (0.77) is significantly higher than inDZsthat are part of a health center (0.56). 17 DZ Mean Characteristic (range) # visits/physician/ # visitshedical # visitshegistered month staffhonth person/month (N=122) (N=122) (N=125) All DZs 561.2 I 160.5 1 0.70 (40.8 - 2377.4) (38.5 - 733.8) (0 -4.32) Rural 635.7 201.9 0.81 (40.8 -2377.4) (1 1.9-733.8) (0 -4.32) Urban 487.9 165.6 0.60 (174.0 - 1274.9) (54.5-493.0) (0 -2.72) Stand-alone 582.8 194.6 0.77 (127.0 -2377.4) (1 1.9-733.8) (0 -4.32) Part of a health 488.0 155.4 0.56 center (40.8 - 1551.7) (70.1-453.1) (0 - 2.78) 3.2 SUMMARY OF ECONOMETRIC RESULTS 42. The main finding of the econometric analysis presented in the Technical Annex i s that although DZs appear to be very constrained in their current ability to improve cost efficiency, there i s substantial variation in production efficiency. The efficiency scores of more than half of DZs (65%) are below 70 percent of the frontier or the most efficient level. This shows that there is clearly an opportunity to improve the production efficiency relative to the current capability of the system, and to move all DZs to a level of higher efficiency. The productivity of DZ health staff, equipment and space can be improved by using equipment more often or reducing the number of equipment, and reducing non-clinical space in DZs. Less-efficient DZs would either have to increase the number of consultations and other outputs, or reduce the number of staff and medical equipment and rent out some of the non-used space. There is very little variation in the cost-efficiency of DZs, because their expenditures are largely pre-determined. The quantity of output does not give the full picture, however, and without being able to compare the quality of services or outcomes achieved, it is difficult to determine how much variation in efficiency is actually taking place. 3.3 SUMMARY OF BASELINEPERFORMANCEMEASURES 43 Table 3.10 summarizes the above analysis to provide a baseline profile for the key efficiency performance measures of DZ performance prior to the implementation of per capita payment. Any follow-up analysis should compare results against these baseline values to identify changes in the behavior of providers in response to the payment change. 18 'able3.10: BaselineValues 01 12PerformanceMeasures Category Performance Measure Average Baseline Value (Standard Deviation) Efficiency of input use % of total expenditureon staff 72.6% (14.8) % of total expenditureon medicine and 10.9% supplies (12.5) % of total expenditureon utilities 3.5% (1.8) % of total expenditurenot categorized 4.4% ("other") (6.9) # visits per physiciadmonth 561 (375) # visits per medical stafumonth 161 (123) Populatiodphysician Populatiodnurse 390 (104) Marginal productivity of inputs (% change in output per 1 % change in units of input): Physicians 0.820 (0.26) Nurses 0.264 (0.290) (but not significantly differentfrom 0) Paramedicalstaff 0.001 (0.028) (but not significantly diyerentfrom 0) Space -0.182 (0,124) (but not significantly differentfrom 0) Equipment 0.016 (0.049) (buf nof significantly differentfrom 0) Marginal cost of outputs (change in total expenditureper 1 unit change in output): Consultations 0.003 SD (0.0007) Laboratorytests 0.00 SD (0.008) Other diagnostic tests 0.00 SD (0.003) Efficiency of output % of total visits that are preventive 9.2% mix (8.8) # of lab tests per consultation 1.02 (0.99) # of diagnostic proceduresper consultation 0.07 (0.14) # of injections per consultation 0.56 (1.39) 19 Category Performance Measure Average Baseline Value (Standard Deviation) Access km' per service point 84.8 (163.8) Total # of visits per catchmentpopulation 0.70 per month (0.60) # of preventive visits per catchment 0.61 Potential unintended populationper year (0.54) consequences average cost per visit 887 SD (1 128) YOof visits resultingin areferral to a 7.1% specialist (0.61) % of visits resultingin a referralto a 5.5% hospital (0.90) Total referralrate (YOof visits resultingin a 12.6% referralto a specialistor hospital) (1.1) 20 CHAPTER 4. CONCLUSIONSAND RECOMMENDATIONS 44. The purpose o f this analysis was two-fold. Firstly, results provide a baseline on the PHC performance before the introduction of capitation payment against which a follow up survey can be compared. Secondly, findings highlight opportunities to improve the efficiency of DZs and support the introduction of capitation payment. The main finding is that although DZs are generally working with the same level of staff, medical equipment and space, which are largely dictated by the system, they vary in the level of output (e.g. consultations) they are able to achieve. That is some DZs are substantially more efficient than others. These variations in output cannot be explained by whether the DZs are rural or urban, stand-alone or part of a health center, or by the size of the catchment area, the population density, or the distance to the nearest hospital. 45. The three main conclusions of the baseline analysis can be summarized as follows: (a) There is room for DZsto managetheir resourcesincluding staff, space and equipment, better, which may be stimulated by the financial incentives set through a per capita payment system. However, financial incentives may not be enough to trigger a behavioral change that leads to more efficient care. Thus, other supporting policy changes and improved data and information systems may be requiredto realize the benefits of the new payment system. (b) There is substantial inequality inper capita expenditures across DZsbasedon the current line-itembudget. The new per capita payment system must contribute to redistributing resourcesmore fairly and basedon the number of individuals enrolled, inorder to achieve the goals of improving access to services. (c) A per capita payment systemmay give DZs more flexibility, but only if salaries are includedunder capitation and human resources policies are eased, which may require legal changes. DZs are currently very constrainedintheir ability to improve cost efficiency, as 70% of their costs are for staffing and determined externally by the system, whereas a relatively small portioni s spent on medicines and supplies inDZs. 46. DZs can become more efficient by reducing the number of staff and space, without having to reduce the total number of visits. The finding that there are too many 21 physicians i s supported by a very low population-to-physician ratio o f only 782 which i s much lower than in the WHO'SEurope region with 3,500 people per primary care phy~ician.~ There are large areas o f unused space in DZs and DZ managers may not necessarily be aware about the cost of space, as they do not have to pay rent o f space. DZswith more equipment are not showing a higherproductivity. 47. DZ mainly produce curative visits, and there appears to be a high number o f laboratory tests and injections. Under capitation it may be expected that the share of preventive visits will increase and unnecessary laboratory tests and injections may be reduced. Still, there may be a need to revisit clinical guidelines to specify more clearly when laboratory tests and injections are necessary. Referral rates for DZs are generally low, though they may increase once capitation has been introduced; however, based on the current data it i s not possible to determine whether patients are beingreferredmore or less than necessary. 48. Some DZs report a very high percentage o f expenditures as "other," which indicates that they are not tracking their expenditures carefully, and highlighting the need for better accounting and resource management. 49. The Government has already started implementing several reform measures in collaboration with donors to prepare the health sector for the planned provider paymentreforms. These measures include: (a) The Government i s currently reviewingproposals to adjust the capitation rate by coefficients for age and gender o f the enrolled population, and to include additional incentives for preventive services, such as a bonus payment to DZs who achieve an agreedlevel for childhood immunization or maternal care. Also, the Government i s considering geographic adjustment inthe capitation payment to adjust for higher utility costs inDZs inmountainous areas. (b) The MOHhas started a review o f clinical guidelinesto ensure that they are compatible with the scope o f services that will be financed by capitation, and provide appropriate guidance on laboratory tests, injections, other procedures, and referrals. (c) The M O H and HIF incollaboration with donors have provided extensive management support to DZ managers such that they will be able to successfully respondto the new capitation payment system. This collaboration i s ongoing and includes management training and investment in new accounting systems inall DZs. (d) The World Bank loan and the EARare assisting the Government inmajor investment inthe data systems inDZs and HIF. The HIF has started to 9WHO EuropeanHealth For All Databasehttr,://data.euro.who.int/hfadb/accessedJuly 2008. 22 improve reporting and analysis o f key data related to DZ performance, including population size and demographic structure, services provided, and resource use. 50. Financial incentives set by the capitation payment system may not be enough to trigger a behavioral change among DZs that leads to more efficient care. Thus, other supporting policy changes and improved data and information systems may be requiredto realize the benefits o f the new payment system. Based on the conclusion from this baseline analysis, several additional steps could support the development o f capitation payment in DZs, prevent adverse effects in reaction to the financial incentives set by the payment, and improve the efficiency of the sector. These measures could be implemented in a phased approach with Phase I focusing on the following administrative measures to complement the payment reforms. Payment System Design (a) Improve the equality o f primary care resource allocation by pooling PHC funds from public sources and paying all DZs a unified per capita rate with appropriate adjustments for cost variations. There i s currently unequal allocation o f public resources for primary care across DZs. This i s likely to be the result o fthe way the line-item budget is now defined based on the number of staff and other factors. This inequality can be addressed by the per capita payment system iffunding from the HIF, municipalities, and other government sources are pooled at a higher level and distributed to DZs througha unifiedcapitated rate. (b) Include salaries inthe capitation amount and adjust relatedhumanresources policies and laws to give more flexibility to DZs to improve their cost efficiency. (c) Explore measures to reduce the unintendedconsequences of the per capita payment system. Capitation may create an incentive for DZsto under-provide services or refer patients to specialists and hospitals. To prevent unnecessary referrals to higher-cost providers, referral guidelines should specify the scope o f services at the PHC level and to appropriate referral patterns. These outcomes should be monitored and the payment system should include measures to counteract these incentives, such as open enrollment, a quality monitoring system, or outcome-based bonuses and penalties for unjustified referrals. (d) Develop an appropriate cost-sharing policy. This study shows that DZ revenues from patients are currently 8 percent of total DZ revenue. While the Serbian health insurance law already sets the legal frame for co-payments, the government may consider developing a cost-sharing policy as part o f overall provider payment reform that facilitates access to primary health care, i s 23 linked to essential drug reimbursement, and encourages appropriate use o f services at the different care levels. ManagementIssues 5 1. DZs are currently very constrained by their fixed costs. If the HIF continues to pay for staff based on a line-item budget, then more than 70 percent o f the DZ budgets will continue to be fixed for salaries, and only about 30 percent o f total costs can be managed by DZs. In order to be effective at improving cost efficiency, DZs must have greater flexibility to shift expenditures from salaries and utilities toward more drugs and supplies; and from higher cost staff (physicians) to lower cost staff (nurses and paramedical staff). (e) Assess the regulations and constraints that may affect the ability o f DZs to manage their resources more efficiently, including regulations affecting the scope o f service of different levels o f medical personal, public procurement laws, and the public labor law according to which public sector employees in DZs are still on the HIF payroll and planned centrally. (f) Conduct an assessment o faccessto essential medicines that are necessary for effective PHC,and consider the potential for limitedfinancing o f essential medicines within the context o f capitation. Collect information on the disease profiles o f the populations served, outcomes, and overall access to essential medicines inthe country. (g) Develop a cost-effective package o fmedical equipmentthat should be available at the PHC level. Consider reducing the number o f equipments in DZs inthe vicinity of a hospital to where patients could be referred. Conduct a functional assessment o f the utilization o f equipment, including whether more basic equipment i s available that may contribute more to DZ productivity (e.g. blood pressure cuffs and scales) 52. Phase 2 could give additional attentionto the following three administrative measures to complement the payment reform that also arise from the conclusion o f this analysis: ManagementIssues (a) Consider reorganizing space inDZs, moving to smaller buildings,or redirecting excess space to other purposes. DZs could rent out non-clinical space to purposes such as private doctors or dentists, or to day-care centers for individuals who need supervision such as elderly or disabled individuals. 24 (b) Hospital payment reforms such as DRGsstimulate changes inhospital care such as shorter hospital stays that will be felt inother parts o f the health care system. As a result, DZs and community care as well as long-term care departmentswill have to be ready to provide a greater degree o f follow-up to patients who have beendischarged from hospitals. Data Systems 53. There are some limitations o f the current analysis, which provide some guidance for where data improvements may be necessary to accurately monitor the effects of capitation payment. (c) Collect information on the disease profiles o f the populations served, outcomes, and overall access to essential medicines inthe country. Develop quality and outcome measures that can be monitoredby DZs themselves for internal management purposes, and at the system level by the HIF and MOH. 54. To evaluate the impact o f capitation over time a longitudinal analysis will be completed to assess after the payment reforms whether the reforms lead to an increase inproductivity and a reduction incost inefficiency. The study designwill depend on the approach to implementing the per capita payment system and will, to the extent possible, exploit the phased-in implementation of the new payment system. Better and more differentiated data could produce more reliable estimates o f marginal products and costs, which would allow additional analysis, such as whether the optimal mix o f DZ inputs i s beingused. 55. The follow-up survey will have to isolate the payment effect from the effect of other variables that also may have changed after the payment change was implemented. For example, inthe case o f measuring efficiency gains from implementingcapitation payment, it i s possible that there are also changes in such contextual factors as the availability o f medicines or changes in the supply o f physicians in DZs as a result o f the Government's rightsizing plan, which would also affect efficiency. In addition, some DZs may be selected for participating inthe new payment system earlier in the implementation process which may be related to factors that are also related to efficiency, such as their reputation for historically superior performance, or already having better data collection. 25 REFERENCES Ashenfelter, 0.and Card, D. (1984). Usingthe longitudinal structure of earningsto estimate the effect of training programs. NBER Working Paper Series. Working Paper No. 1489. Battese, G. and Coelli, T. (1995). A model for technical inefficiency effects ina stochastic frontier production function for panel data. Empirical Economics 20: 325- 332. Borowitz M.,O'Dougherty S., Cashin, C., Hafner, G.,Samidjiyski, J., VanDevelde, C., and McEuen,M.(1999). TheKazakhstan Country Program. Abt Associates Inc. USAID-Funded ZdravReform Program. Cashin, O'Dougherty, et al. (2007). The design and implementation of per capita payment systems inthe context of primary health care-centered health systems development inlow- and middle-income countries. Abt Associates Inc. ZdravPlus Program. Almaty, Kazakhstan. Clark, M. 2002. Health sector reform inCosta Rica: reinforcing apublic system. Preparedfor the Woodrow Wilson Center Workshops on the Politics of Education and DeFelice L and Bradford WD (1997). Relative inefficiencies inproduction between solo and group practice physicians Health Economics (6)5: 455-465. ECHSD ConceptNote: Baseline study to evaluate the impact of provider payment reforms on performance inhealth facilities in Serbia. December 18, 2007. Fleetcroft, R. 2006. The relationship betweenprescribing expenditure and quality in primary care: an observational study. British Journal of General Practice 56(529): 613- 619. Gauri, V., Cercone, J., and Briceno, R (2004). Separatingfinancing from provision: evidence from 10 years of partnership with health cooperatives inCosta Rica. Health Policy and Planning 19(5): 292-301. 26 Gaynor, M.and Pauly, M.(1990). Compensation andproductive efficiency in partnerships: evidence from medical group practice. TheJournal of Political Economy 98: 544-573. Gerdtham, U.-G.,Lothgren, M.,Tambour, M.,and Rehnberg, C. 1999. Internal markets and health care efficiency: a multiple-output stochastic frontier analysis. Health Economics 8: 151-164. Giannakas, K.et al. 2003. On the choice of functional form instochastic frontier modeling. Empirical Economics 28: 75-100. Grannemann, T., Brown, R., and Pauly, M. 1986. Estimating hospital costs. Journal of Health Economic 5: 107-127. Greb, S., Delnoij, D.,and Groenewegen, P. (2006). Managing primary care behavior through payment systems and financial incentives. InSaltman, R., Rico, A,, and Boerma, W. Primary Care in the Driver's Seat? Organizational reform in Europeanprimary care. Berkshire, England: Open University Press. pp. 184-200. Gruber, J. (1994). The incidence of mandatedmaternity benefits. American Economic Review 84(3): 622-641, HealthAction International/WHO. 2008. Measuring medicine prices, availability, affordability and price components. Second edition http://www.haiweb.org/medicineprices/. Hefford, M. (2005). Reducing health disparities through primary care reform: the New Zealand experiment. Health Policy 72: 9-23 Hollingsworth, B (2003). Non-parametric and parametric applications measuring efficiency inhealth care. Health Care Management Science 6(4) 203-218. Hollingsworth, B. 2008. The measurementof efficiency andproductivity of health care delivery. Health Economics 17: 1107-1128 Jacobs, R., Smith, P., and Street, A. ( 2006). Measuring Efficiency in Health Care.: Analytic Techniquesand Health Policy. New York: Cambridge University Press. Johnson, S. and Lahiri, K. 1992. A panel data analysis of productive efficiency in freestanding health clinics. Empirical Economics 17: 141-151. Kozak, L.,Hall, M.,and Owings, M. (2001). Trends inavoidable hospitalizations, 1980- 1998. HealthAffairs 20(2): 225-232. 27 Kumar, R. 1999. Research Methodology: A Step-By-StepGuidefor Beginners. London: SAGE Publications. Liu,X. (2000). Improvingthe allocative efficiency ofhealth services. GlobalProgram for Evidence for Health Policy. Geneva: World Health Organization. MOH Serbia and David Cochrane: Serbia Human Resource Strategy for the Health Sector. 2005. Ministry of HealthofNew Zealand. (2001). ThePrimary Health Care Strategy. Wellington, New Zealand. Ministry of Health of the Kyrgyz Republic. 2008. Mid-Term Review Report: Manas Taalimi Health Sector Strategy. Bishkek, Kyrgyzstan. Nordyke, RJ (2002). Determinants of PHC productivity and resource utilization: a comparison of public and private physicians inMacedonia. Health Policy 60(1): 67-96. Puig-Junoy J and Ortun V. 2004. Cost efficiency inprimary care contracting: a stochastic frontier cost function approach. Health Economics (13): 1149-1165 RosenmanR and FriesnerD. 2004. Scope and scale inefficiencies inphysician practices Health Economics (13) 11: 1091-1116. RosenmanR, Siddharthan K, and Ahern M. 1997. Output Efficiency o f Health Maintenance Organizations inFlorida Health Economics (6)3 : 295-302. Ruggiero, J. 1999. Efficiency estimation and error term decomposition inthe stochastic frontier model. European Journal of Operational Research 115(3): 555-563 Sanigest. Health Sector Restructuring Strategy:Network Plan. Serbia. Republicof Serbia. May 2007. Suraratdech, C. and Okunade, A. 2006. Measuring operational efficiency ina health care system: a case study from Thailand. Health Policy 77: 2-23. Schneider, P. (2007). Provider payment reforms: lessons from Europe and America for South EasternEurope. Health, Nutrition and Population Discussion Paper. The World Bank: Washington, D.C. Street A. 2003. How muchconfidence shouldwe place in efficiency estimates? Health Economics (12) 11: 895-907. 28 Wagstaff, A. and Lopez, G. 1996. Hospital costs inCatalonia: a stochastic frontier analysis. Applied Economic Letters 3: 471-474. Wagstaff, A. 1989. Estimating efficiency inthe hospital sector: a comparison of three statistical cost frontier models. Applied Economics 21: 659-672. Wagstaff, A. and Barnum, H. 1992. Hospital cost functions for developing countries. Population, Health andNutrition Working Paper 1044. Washington, D.C.: The World Bank. Weaver, M.and Deolalikar, A. 2004. Economies of scale and scope inVietnamese hospitals. Social Science and Medicine 59: 199-208. Weissman, J. S., Gatsonis, C., and Epstein, A. M. 1992. Rates of Avoidable Hospitalization by Insurance Status inMassachusettsand Maryland. JAMA. 268(17): 2388-2394. World Bank (2007). Serbia Decentralization and Local Service Delivery. Chapter 4. PREM. Europe and Central Asia. World Health Organization (2001). Basic Health Indicators Database. World Health Organization web site: http://www.who.int. Zavras, A,, Tsakos, G., Economou, G., and Kyriopoulos, J. (2002). Using DEA to evaluate efficiency and formulate policy within a Greek national primary health care network. Journal of Medical Systems 26(4): 285-292. ZdravReform Program (2000). Health reform initiatives in Central Asia: ZdravReform Program final report. ZdravReform Program: Almaty, Kazakhsta 29 ANNEX ANNEX1:TECHNICAL ANNEX": 1) Reviewof Studies on ProductionFunctionsandCost Functions 56. The concept o f efficiency measurement is concerned with assessingthe competence with which inputs are converted into outputs (Jacobs et al, 2006). The literature shows a mix o f the production function and cost function approaches to analyzing health provider efficiency, and different units o f analysis (individual physician, provider organization, and purchaser). This review provides a summary of approaches to empirical specification and estimation o f health provider production and cost functions. 57. Different measures o f output produced by health care providers are found in the literature. Most studies focusing on ambulatory care use, some variation o f services (usually visits) per clinic or physician per unit o f time. One study defined output as health, measured as the number o f enrollees in a primary care practice (Rosenman et al, 1997), and one study defined output as the size and composition o f the covered population, number o f visits, and quality o f care as measured by the percentage o f the population for which specific targets were achieved, including vaccination targets and coverage by a healthy child program, and performance o f physicians on a simple knowledge test (Puig-Junoy, 2004). 58. Inputs are measured differently in cost and production functions. In cost function analysis, the use o f inputs i s summarized in a single measure o f the total cost o f the provider, or the total amount spent on all inputs combined. The literature varies substantially in how health care inputs are defined in production functions. One study specifies the number o f visits per physician per week (output) as a function o f the number of hours per week worked by the physician, the number of nursing hours per physician, and the number of administrative hours per physician (DeFelice, 1997). Another study specifies inputs as physician time per visit, indicator variables for the availability o f basic medical equipment, and the number o f nurses per physician (Nordyke 2002). Definition and Studies on Production Functions 59. Health provider production i s typically analyzed (modeled) as a jointly determined process o f workload (output) and input utilization (cost) (Nordyke, 2002). The production function describes the relationship betweenthe total output possible and I OCheryl Cashin PhD, wrote this technical annex 30 different levels and combinations o f inputs, such as labor and capital, as well as external factors that may affect total output (control variables). 60. Adding control variables to the production function recognizes that the production o f health care also embodies behavioral responses by providers, rather than simply a technical relationship between inputs and outputs (Gaynor and Pauly, 1990). These control variables may include physician characteristics, patient case-mix, and market characteristics (e.g., ownership, competition, profit status) and often reflect the policy variables o f interest, such as the provider payment method. Adding these variables allows defining a framework for estimating the effect o f individual providers or clinic characteristics on efficiency, as well as the change o f a policy variable (e.g. new per capita payment method) on provider input and/or output choices (Johnson and Lahiri, 1992). The production function may be described as: Y = Y(L, K, 0,X), where: Y =level of output (e.g. number ofpatients) L =amount of labor input used K =amount of capital inputused 0 =amount ofother inputs used X =other variables that can be expected to affect total output 61 The functional form o fthe production function describes how inputsare expected to relate to output. For example, a Cobb-Douglas functional form i s used when it i s expected that output will change in direct proportion to changes in inputs. The translog production model i s flexible and often used to study how different inputs are substituted for each other inthe production of health care services. Inone study o f the health care system o f Thailand, the authors used a translog production model to examine the production efficiency of the health care system at the macro level (Suraratdech, 2006). The authors examined the productivity o f and substitutions betweenphysicians, pharmacists, nurse, and beds. They found that although nurses and beds make a positive contribution to output (live births per thousand populations), a higher number o f physicians and pharmacists i s negatively associated with improved outcomes. 62 One study in Macedonia estimated a physician-level production function as three jointly determined choices: output as patient visits per physician per week, inputs as time spent/patients and medical equipment/patient (Nordyke, 2002). The control variables included an indicator for whether the provider was public or private to test the impact of facility ownership on efficiency. Results showed that private physicians had higher output (productivity) and chose a different input mix (more eauiDment relative to labor) than Dublic sector Dhvsicians. 1 1 I , 31 Definition and Studies on Cost Functions 63. The cost function describes the relationship between the total input use (cost) and the different levels ofoutput produced. The cost function may be described as: C = C(Y, X), where: C =the total cost Y =level of output X =other variables that can be expected to affect total costs 64. Most studies estimate traditional cost functions that capture the multi-product nature o f health services production. Grannemann, Brown and Pauly (1986) provided the basis for much o f the subsequent work on estimating cost functions for health care providers. The authors specified a cost function for hospitals as a function o f multiple outputs (inpatient days by type, discharges by type, outpatient visits, emergency department visits), input prices, and other factors that may be expected to shift the cost function (e.g. the sources o f revenue o f the hospital, the age structure o f the physician staff, form o f ownership o f the hospital, teaching affiliation, and geographic region). The objective o f their study i s to estimate the marginal cost o f production o f different outputs, recognizing that the level of some inputscannot change easily inthe short run,but other inputs will vary with the level o f output. The marginal cost of each type o f output i s expressed as a fraction of total cost, varying with the level o f output. The authors found that hospital stays in pediatrics, surgery, obstetrics and gynecology, and other specialties are somewhat more costly than cases in general and internal medicine, even controlling for length o f stay. The authors also found that the source o f payment for the case, Medicaid and Medicare, significantly affected cost, with Medicaid inpatient days more costly than Medicare days. 65. Ina study o fhospital costs inVietnam, the authors use a variation of Grannemann's multiple-product cost function. A cost function is specified in which hospital variable costs are a function o f input prices, the number o f hospital beds as a proxy for hospital capital stock, and the level of output o f multiple products, including inpatient days and outpatient visits (Weaver, 2004). Interaction terms betweenthe different products were included to estimate economies of scope, or efficiencies that may be achieved by producing multiple products. The authors found large variations inthe marginal cost o f hospital admissions across categories of hospitals, from US$12 per admission in district hospitals, to US$170 per admission in central general hospitals, and modest economies o f scope. 66. Wagstaff and Barnum (1992) conducted a survey of techniques available for analyzing hospital costs to answer specific policy questions, including (i)are hospitals over-capitalized? (ii)are hospitals inefficient? (iii)should hospitals specialize or provide a broad range of services? and (iv) are there too many hospitals? The authors concluded that although the application o f current methods to hospital cost analysis in developing countries did not lead to clear guidance for 32 answering the policy questions, there were some methodological conclusions. First, the estimation of short-run cost functions is preferred to long-run cost functions, because hospitals may not be employing their long-run equilibrium quantities o f capital. Second, because o f the difficulty o f separating variable from fixed costs, it may be preferable to estimate short-run total cost equations. Third, since hospitals may be technically and allocatively inefficient, frontier models should be used. Fourth, to avoid too many restrictive assumptions about cost structures, flexible functional forms should be specified for cost functions. Finally, because economies o f scale are a long-run phenomenon, the estimation o f short-run cost functions may not fully reflect economies o f scale without calculation o fthe optimal capital stock. Definition and Studies on Stochastic Frontier Analysis 67. A widely accepted method for estimating health care production and cost functions i s stochastic frontier analysis (Puig-Junoy, 2004; Rosenmann, 2004; Street, 2003; Gerdtham, 1999; DeFelice, 1997; Battese 1995; Gaynor, 1990). This method estimates production and cost functions using ordinary least squares (OLS) regression methods. The least squares error term i s decomposed into two components: the random, "white noise" component, and the component attributable to inefficiency. The maximum production or minimum cost "frontier" i s estimated from the observation with the minimum error term, and the inefficiency o f the other observations i s estimated as the deviation from this minimum, excluding the random noise error11,12,13. The SF method generates an overall efficiency value and ranking o f individual units based on how they perform relative to this value (Puig- Junoy and Ortun, 2004). 68. Several studies use the stochastic frontier production approach to analyze provider response to reimbursement and compensation incentives. In one study using the stochastic production frontier approach to analyze provider production, the authors show that output-based reimbursement improved efficiency o f health care providers in Swedish counties (Gerdtham et al, 1999). In another study, the production frontier approach combined with a behavioral production function showed that productivity-based compensation affects the quantity produced by individual physicians but not technical efficiency for U.S. medical groups, because the increased output i s produced by also increasing inputs (Gaynor, 1990). 69. Two approaches to the stochastic frontier cost (SFC) model are found in the literature on health services costs. Inthe first approach, a cost function i s specified as a function o f single or multiple outputs, input prices, with other variables included that may shift the cost function. Average inefficiency is estimated from the decomposed error term (Wagstaff, 1996; DeFelice, 1997; Street, 2003). The Battese, G. and Coelli, T. (1995). A model for technical inefficiency effects in a stochastic frontier production function for panel data. Empirical Economics 20: 325-332 12Wagstaff, A. and Lopez, G. 1996. Hospital costs in Catalonia: A stochastic frontier analysis. Applied Economic Letters 3: 471-474. l3DeFelice, L.and Bradford, W. 1997. Relative inefficiencies in production between solo and group practice physicians Health Economics (6)5: 455-465. 33 other approach i s to specify the cost function only as a function o f output and input prices, with a separate function estimating the inefficiency term that includes the cost shifters as explanatory variables for inefficiency, the parameters o f which are estimated simultaneously with the cost function (Battese, 1995;Puig-Junoy, 2004). 70. In a study o f hospital costs in Catalonia, the authors apply the SFC approach as an extension o f Granneman's cost function model (Wagstaff, 1996). A cost function i s specified with a flexible functional form, with hospital costs a function o f ambulatory visits, emergency cases, and inpatient discharges. Variables are included that may be expected to shift the cost function, including teaching status, a proxy for case-mix complexity, and several variables reflecting the technology o f the hospital. The authors estimate the marginal cost o f producing each o f the three types o f hospital output, which as expected is lowest for ambulatory visits and highest for inpatient cases. The authors also find complementarity between ambulatory visits and emergency cases, which i s reflected in modest economies o f scope. The inefficiency estimate from the model shows that the hospitals were on average operating 58 percent above the minimumcost frontier. One study used the SFC approach to estimate the average inefficiency o f hospitals in the U.K. Department o f Health (Street, 2003). The author found that the average mean level o f efficiency o f 90 percent estimated by the SFC model would have been underestimated by a traditional cost function, which estimated the mean level o f efficiency to be 69 percent. 71. A SFC model also was used to estimate differences in efficiency between solo and group practice PHC physicians in the US (DeFelice 1997). The authors incorporated variables that affect physician effort, including physician experience and non-practice income. The results suggest that whether PHC providers are organized in solo or group practice does not have a significant effect on efficiency. 72. A different study in Catalonia used a SFC approach to measure the efficiency o f contracting out primary health care (Puig-Junoy, 2004). The authors found that there was no efficiency gain from contracting out PHC services instead o f direct public service provision. The cost function showed that the purchaser's average contract with contracted-out PHC providers was 19.3% above the minimum cost frontier for a given level of output, while the average contract with publicly managed PHC providers was only 7.6% above the minimumcost frontier. 73. The stochastic frontier approach i s widely used in efficiency analysis, but it has several limitations. First, the approach requires strong assumptions about the distribution o f the error terms in order to disentangle inefficiency from statistical noise (Wagstaff 1989). The inefficiency error term i s often modeled as a half- normal distribution, which i s done in the present study, but this i s somewhat arbitrary. The SF approach has also been criticized for relying on the proper decomposition o f the error term into inefficiency and noise, which may be 34 problematic (Ruggiero, 1999),14and being very sensitive to the specification of the functional form (Giannakas et al., 2003; Hollingsworth, 2008).'57'6 A further limitation of the stochastic frontier model i s that it can only estimate the degree to which each observation deviates from the best production that i s observed, which itself may also embody inefficiency (DeFelice and Bradford, 1997). 74. Another approach to efficiency analysis i s data envelope analysis (DEA) (Rosenman, et al, 1997). Data envelop analysis i s a linear programming method that derives the efficiency frontier strictly from the data, without the economic theory foundation or statistical approach o f stochastic frontier analysis. The observations with the highest ratios o f output to input are considered efficient, and the efficiency frontier i s constructed by joining these observations in input-output space (Jacobs et al, 2006). The DEA approach has been used extensively to analyze health care provider efficiency in a wide variety o f settings (Hollingsworth, 2003; Zavras et al, 2002). It may be concluded from the literature that stochastic frontier approach i s preferred, however, because it i s a parametric method and therefore accommodates both statistical error as well as error attributable to inefficiency, which DEA does not. That is, the best practice frontier i s not strictly defined by the "highest" outputlinput pairs in the data, as observations are permitted to be somewhat above or below best practices due to measurement error and other random noise (DeFelice, 1997). In addition, SFC analysis can incorporate factors related to behavior (not just input-output mix), such as incentives, which allow an analysis o f provider response to policy changes. 2) EconometricMethodsUsedin this Analysis" 75. The econometric analysis presented in this study will use a SF model to estimate both the production and cost functions. The SF models will estimate the deviation of each DZ from the best performance (maximum output or minimum cost) observed in the sample. This deviation is considered to be caused by both inefficiency and random factors. A one-period production and cost functions will be estimated using the year 2007 as the baseline year. Although output variables are reported by month, and a time-series analysis could be possible, the input variables are reportedfor the entire year o f 2007, so they do not vary inthe data. 14Ruggiero, J. 1999. Efficiency estimation and error term decomposition inthe stochastic frontier model. European Journal of Operational Research 115(3): 555-563. l5Giannakas, K.et al. 2003. On the choice of functionalform in stochastic frontier modeling. Empirical Economics 28: 75-100. 16Hollingsworth,B. 2008. The measurementof efficiency and productivity ofhealth care delivery. Health Economics 17: 1107-1128. 17The complexity and power of the econometric analysis is limited by the content and quality ofthe data. For example, analyzingboththe productionfunction and the cost function can have the added advantage of shedding light on the sources of inefficiency (Wagstaffand Barnum 1996). Ifinput prices are known, it is possibleto measure the excess costs to a provider caused by combininginputs in the wrong proportion usingthe marginalproductestimates and inputprices. Becausethe data do not allow disaggregation of expenditures, and therefore prices, across importantinput categories(e.g. physicians vs. nurses), this level of analysis is not possible. 35 76. The econometric analysis o f the production function for DZs will be completed in two stages. First, a stochastic production frontier function i s estimated to provide estimates of the technical efficiency inthe use o f the inputsby DZs inproducing the main output o f consultations." Variables reflecting the proportion of visits for children, adult women, and adults over 60 are added to control for the case-mix o f types o f visits to DZs. The first stage production function is specified as a Cobb- Douglas production function as follow^:'^ InY, =p,lnL,, +p2lnL,, +p3lnL,, +p, InSpace, +p, InEquipment,+ PJnPropchild, +p,lnPropfem, +p,lnPropsenior[ +(E, ut) + (1) where, lnY, = The natural log o f the total number o f visits per year in DZ iZo 1nLp = The natural log o f the total number physician full-time equivalents in DZ i 1nLNi = The natural log o f the total number nurse full-time equivalents in DZ i 1nLPAi = The natural log o f the total number paramedical full-time equivalents in DZ i InSpace, = The natural log o f the total space in square meters in DZ i InEquipment, = The natural log o f the total ndinber o f machines in DZ i InPropchild, = The natural log o fthe proportiono fvisits by children 0-15 years in DZ i InPropfem, = The natural log o f the proportion o f visits by females 16-59 years in DZ i InPropsenior, = The natural log of the proportion o f visits by adults 60 and above in DZ i El = independent, identically distributedrandomerror for DZ i u, = non-negative unobservable random variable associated with technical inefficiency for DZi that is assumed to follow a half-normal distribution The pjCj=1-5) coefficients generated by the least squares estimation o f equation (1) will provide an estimate o f the productivity o f each o f the inputs including labor, equipment, space etc. The estimated value o f Vi provides information on the level o f production inefficiency o f provider i.The level o f inefficiency may be calculated as the ratio o f frontier maximum output to the observed output for provider i." 77. Second, an equation will be specified to determine the factors that explain differences in inefficiency across DZs. 22323 A variable will be included for the percentage o f revenue the provider receives through capitation, although this will be '' Production functions were estimatedto include two other outputs, laboratory tests and diagnostic tests, but no input variables were found to be statistically significant determinantsof these outputs. 19 A translog specification ofthe production function was tested, becauseit provides a general flexible functional form. A likelihoodratio test failed to reject the hypothesisthat the second order input arameters are equal to zero, so the Cobb-Douglas function was used as the final specification. A Cobb-Douglas specification of the production function will be used, so the empirical estimate will be based on the natural logarithm ofthe output (visits) and inputs (full-time equivalents for physicians, nurses, and paramedicalstaff). 222'Puig-Junoy and Ortun 2004. Batteseand Coelli, 1995. 23 Puig-Junoy and Ortim 2004. 36 zero for all providers in the baseline analysis, and reach a higher value in the follow-up survey. u, = capitation, 4 + $2Private,+ $3 Standalone, +$4 Rural, +$5 Radius, + $6Density, $7Distance, $.$taffRed+m, + + (2) where, Capitation, Percentageofthe revenue of DZ ipaidby capitation Private, An indicatorvariable (011) for whether DZ iis private Standalone, An indicatorvariable (0/1) for whether DZ ipart of a hospital/healthcenter Rural, An indicatorvariable (0/1) for whether DZ iis locatedin a rural area Radius, The radiusofthe catchment area of DZ i Density, The populationdensity of the catchment area of DZ i Distance, The distance from DZ ito the nearest hospital StaffReduct,, % oftotal staffof DZ ireducedbetween 2007 and periodt due to the Ministry human resources policy a randomvariable The coefficients generatedby the least squares estimation of equation (2) will provide $1 an estimate of the impact of eacho f the external factors on DZs' production inefficiency. The production function is estimated using thefrontier function in STATA 9.0. 78. The econometric analysis of DZ costs will use a SF cost function analysis. The total expenditure of the DZ is specified as a function of the level of output across different service categories. Typically input prices are also included when estimating the cost function, but it can be assumed that all DZs face the same input prices. There may be some variation by urbadrural areas, which will be captured by including a variable for rural DZ. Again, variables reflecting the proportion of visits for children, adult women, and adults over 60 are addedto control for the case mix of types o f visits to DZs. The empirical SF cost function is specified as follows:24 where, lnC, naturallog ofthe total annual expendituresof DZ i InY,, natural logofthe annualtotal number of consultationsin DZ i 1nYLt natural logof the annual total number of laboratorytests in DZ i lnYDi natural log ofthe annual total number of diagnostic services in DZ i InPropchiId, The natural logof the proportionof visits by children 0-15 years in DZ i InPropfem, The natural log ofthe proportionof visits by females 16-59years in DZ i InPropsenior, The naturallog of the proportionof visits by adults 60 and above in DZ i "1 independent, identicallydistributedrandom error for provider i 24Again, a more flexible functional form was used initially following that o f Wagstaffand Lopez 1996,but the parameterson secondandthirdorder interactionterms betweeninputswere not foundto differ significantly from zero, so the simpler specification was used inthe final presentationofresults. 37 - non-negative unobservablerandomvariable associatedwith cost inefficiency for provider ithat i s assumed to follow a half-normal distribution The a,Cj=1-3) coefficients generated by the least squares estimation o f equation (3) will provide an estimate o f the average marginal cost o f producing an additional unit of each different type o f output. The estimated value o f u, provides information on the level o f cost inefficiency o f DZ i. 79. A separate function will be specified to model the effects o f independent variables on inefficiency. This step o f the analysis will allow the effect o f the Ministry's human resources policy on input use to be separated from the affect o f the new per capita payment system. v,=0, Capitation, + @,Private, + 6, Standalone, +0, Rural, +e3Radius, + B,Density, +B,Distance, +@,StaffRed,, +P, (4) Where all variables are specified as inequation (2) above. The '? coefficients generated by the least squares estimation o f equation (4) will provide an estimate of the impact o f each o f the external factors on the cost inefficiency o f DZs. The cost function is estimated usingthefrontier function in STATA 9.0 with the cost option. 80. The production and cost functions are used to produce estimates o f marginal productivity of inputs, or the contribution to output o f employing an additional unit o f the input; and the marginal cost o f output, or the contribution to total variable costs o f producingan additional unit of the output. 81. The analysis o f equations (1) through (4) is completed using STATA statistical software. The following specific analysis i s completed. Inthe follow-up study, the impact o f capitation payment on efficiency can be estimated from this basic model. - A comparison o fthe productivity of different inputs - Analysis of the contribution to total costs o f different outputs - Estimate o f the average level o f production and cost inefficiency of DZs, and - ranking o f DZs by efficiency Analysis o f the impact o f external market factors, including the Ministry's restructuring policy, on efficiency 3) Resultsfrom EconometricAnalysis 82. The main objectives of the regression analysis presented in this Section are to: (1) identify the marginal productivity o f key inputs, and the marginal cost o fproduction for the main outputs o f the PHC centers (DZs); and (2) measure and provide a 38 baseline for assessing changes in provider performance, in particular efficiency, after the implementation o f per capita payment. In order to meet both o f these objectives, a stochastic frontier (SF) model was chosen for this analysis over the estimation o f traditional production and cost functions, which do not provide the opportunity to compare the efficiency o f different units. Results on the model buildingprocess are presented inSection4. Results Production Function 83. The DZ production function i s estimated to assess the baseline productivity o f DZ inputs, and the current level o f inefficiency relative to what is achievable in the current system. The estimated coefficients presented in Table A.l reflect the productivity o f different inputs in producing DZ consultation^^^. Only the number o f physicians i s significantly associated with increased output. The related coefficient can be interpretedas a one percent increase in the number o f physicians in DZs leads to a 0.82 percent increase inthe number o f visits, which is significant at the one-percent level (p IzI Dependentvariable ln(tota1# of consultationsat the DZ in2007) # of observations 134 Log likelihood -118.959 Wald chi2 231.84 Prob > chi2 0.000 In (# of physicians) 0.820 3.21 0.001*** In (# of nurses) 0.264 0.91 0.361 In (# of paramedicalstafg 0.001 002 0.984 In(totaI space m2) -0.182 -1.48 0.140 In(tota1# of machines) 0.016 0.33 0.742 In(proportion of visits by -0.031 -2.08 0.038" children 0-15 years) ln(proportion of visits by 0.022 1.65 0.098* females 16-15) Ln(proportion of visits by -0.008 -0.47 0.639 adults 60 and over) Constant 10.13 11.22 0.000"' sigma-v 0.445 sigma-u 0.651 sigma2 0.622 lambda (sigma-u/ sigma-v) 1.461 chi-squared(1) = 4.30 Prob>chi-squared=O.O19 **statistically * significant at the 10% level; ** statistically significant at the 5% level; statistically significant at the 1% level. Table A.2: Determinantsof Inefficiencyin DZ Production Variable Coefficient T Prob > It1 Dependentvariable u(estimatedinefficiency term) # of observations 134 R2 0.0058 F(5, 128) 0.15 Prob > F 0.980 Rural 0.018 0.35 0.728 Stand-alone -0.026 -0.43 0.670 Distance from hospital 0.0006 0.67 0.502 Catchmentarea size 6.70e-06 0.08 0.93 (km2> Population density -9.62e-07 -0.07 0.946 Constant 0.504 6.41 o.ooo*** 40 ***statisticallysignificant at the 10% level; **statistically significant at the 5% level; statistically significant at the 1% level. 87. The stochastic frontier production function estimates can be used to calculate the level of relative production efficiency for each DZ, which i s the ratio o f the total number of consultations to the maximum possible output. The ranking o f DZs by production efficiency score i s presented in Figure A.1. An efficiency score o f 1 indicates that the maximum possible output has been achieved, so scores closer to 1 indicate more efficient DZs. Efficiency scores for DZs range from 0.136 to 0.866, with a mean of 0.641 (median=0.640). Six DZs have efficiency scores below 0.40 raising concerns about their levels o f inefficiency, whereas 14 DZs report rather highlevels o f efficiency with scores above 0.80. Figure A.l: Rankingof DZs by Production Efficiency Score - 0 0.9 S 0.8 .g6 2I: - 2 0.7 S U I 0.6 E 0.5 .o .E90.4 0.3 -g3 E 0.2 0.1 0 1 , , 0 20 40 60 80 100 120 140 DZ Results Cost Function 88. The DZ cost function is estimated using SF analysis to assess the baseline cost efficiency of DZs in producing their main outputs: consultations, laboratory tests, and diagnostic tests. The estimated coefficients reflect the marginal cost o f different DZ outputs (Table A.3). The coefficient on consultations can be interpreted as follows. A one percent increase in the number o f consultations leads to a 0.54 percent increase in DZ cost/expenditure, which i s significant at the one- percent level (p Iz/ Dependentvariable In(thetotal expenditure of the DZ in 2007) # of observations 129 Log likelihood -217.44 Wald chi2 20.02 Prob chi2 0.0012 In (# of consultationsin 0.544 4.35 o.ool*T 2007) In (# of laboratory tests) 0.004 0.12 0.902 In(# ofdiagnostictests) 0.010 0.39 0.699 ln(proportion of visits -0.002 -0.06 0.949 by children 0-15 years) In(proportion of visits -0.009 -0.30 0.768 by.'females 16-15) Ln(proportion of visits 0.031 0.86 0.387 by adults 60 and over) Constant 12.02 2.10 0.036** sigma-v 1.306 sigma-u 0,001 sigma2 1.705 lambda 0.007 chi-squared(1) = 0.00 Probkhi-squared=1.OOO ****statisticallysignificant at the 10% level; ** statistically significant at the 5% level; statistically significant at the 1% level. Table A.4: Determinants of DZ Cost Inefficiency Variable Coefficient T Prob > It/ Dependentvariable u (estimated inefficiency term) # of observations 129 R2 0.145 F(5, 123) 4.17 Prob > F 0.0016 Rural -5.99e'08 -1.33 0.186 Stand-alone -2.15e'08 -0.45 0.653 Distancefrom hospital -5.08e-lo -0.70 0.488 Population density -4.48e-' -3.76 0.ooo"' Constant 0.0007 0.000"' **statistically * significant at the 10% level; ** statistically significant at the 5% level; statistically significant at the 1% level. 42 90. All DZs show a similarly high level of cost efficiency. The mean o f the inefficiency error term is 0.0009, and it does not vary significantly, with a range o f 0.000853 to 0.000855. The analysis cannot reject that an efficiency score o f 1 was achieved by all o f the DZs, again reflecting that there was no variation in cost efficiency. The ranking o f DZs by cost efficiency score i s presented inFigure A.2. Figure A.2: Rankingof DZs by Cost Efficiency Score .-E23E 1.000856 1.000855 U8 0 1.000852 0 20 40 60 80 100 120 DZ 4) ModelBuilding Process 91. This section presents the alternative models shown inTable AS that were estimated in order to arrive at the final econometric specifications using an SF model presented in the previous Section 3. Given the weaknesses o f SF models and the lack of consensus in the health economics literature on the validity o f the SF approach, this section compares the SF models to the estimation o f traditional productiodcost functions in the model-building process. In addition, alternative specifications o f functional form are tested under each approach. Based on the overall comparison o f alternative production model specifications, it i s concluded that the SF production model with a Cobb-Douglas functional form i s preferred. In addition, a plot o f the ranking o f DZs by the estimated production efficiency score shows little variation across alternative functional specifications. Similarly, a comparison o f the results o f the estimations o f the traditional cost functions and the SF cost models indicates that the SF model is preferred. 92. Traditional cost and production functions are estimated with the variables that may affect efficiency included as explanatory "shifter" variables. These models are compared to the stochastic frontier approach, with the shifter variables included in the regression analysis o f the efficiency error term. 43 Approach Functional Form Traditional Stochastic frontier model Production cost Production cost Translog X X X X Cobb-Douglas X X X X Granneman, Brown, and -- X -- X Paulv Cost Function 93. Different functional form specifications for the cost and production functions are examined. The starting point for the production function is a translog specification, which is a general and flexible functional form: where, lnY, The natural logofthe total number ofvisits per year in DZ i27 lnL,i The naturallogofthe total number physicianfull-time equivalentsin DZ i (input 1) The natural logof the total number nurse full-time equivalents in DZ i(input 2) The natural log ofthe total number paramedical full-time equivalents in DZ i (input 3) Inspace, The naturallog of the total space in square meters in DZ i(input4) InEquipment, The naturallog of the total number of machines in DZ i(input 5) lnPropchild, The natural log ofthe proportionof visits by children 0-15 years in DZ i InPropfem, The natural log ofthe proportionof visits by females 16-59years in DZ i InPropsenior, The natural log ofthe proportionof visits by adults 60 and above in DZ i X" A set of variable that are expected to shift the productionfunction (rural, standalone, catchment area, distance to the nearest hospital, population density) includeddirectly inthe traditionalproductionfunction only independent, identicallydistributedrandom error for DZ i = non-negativeunobservable random variable associatedwith technical inefficiency for DZi that is assumedto follow a half-normaldistribution(in the SF models only) 94. Ifthe second-order input parameters are not found to be significantly different from zero, the functional form reduces to a Cobb-Douglas, which is a special case of the translog: 27A Cobb-Douglas specificationofthe productionfunction will be used, so the empiricalestimatewill be based on the natural logarithm of the output (visits) and inputs (full-time equivalents for physicians, nurses, and paramedical staff). 44 95. The starting point for the cost function i s also the flexible translog specification: lnc,=a;lnYCi+a,lnY, +a,lnY, + where, natural log of the total annual expenditures of DZ i natural log of the annual total number of consultations in DZ i(output 1) natural log of the annual total number of laboratory tests in DZ i(output 2) natural log of the annual total number of diagnostic services in DZ I (output 3) InPropchild, = The natural log of the proportion of visits by children 0-15 years in DZ i InPropfem, = The natural log ofthe proportion of visits by females 16-59years in DZ i InPropsenior, = The natural log of the proportion of visits by adults 60 and above in DZ i X" - - A set of variable that are expectedto shift the productionfunction (rural, standalone, catchment area, distance to the nearest hospital, population density) included directly in the traditional productionfunction only independent, identically distributed random error for provider i - - non-negative unobservable random variable associatedwith cost inefficiency for provider ithat is assumed to follow a half-normal distribution (in the SF models only) 96. Again, if the second-order input parameters are not found to be significantly different from zero, the functional form reduces to a Cobb-Douglas cost function, which i s a special case o f the translog: 97. A third, even less restrictive functional form for the cost function i s estimated that is consistent with economic theory and accounts for the multi-product nature o f DZ output. The functional form, which is based on the specification used by Granneman, Brown, and Pauly (1986) to estimate a hospital cost function, i s as follows: 45 lnc,=qlnY, +qlnY, +a,lnY, + TraditionalProduction Function Specifications 98. The results of the estimation of alternative traditional production functions are presented below in output tables (1) - (4). The estimation of the traditional production function with a translog functional form (1) has a high R2 (0.698), but none of the variables in the model were found to be statistically significant, including the variables that are hypothesized to affect efficiency. The coefficients on the staff variables (marginal product of labor) have particularly wide confidence intervals. Becauseof the small sample size, the additional higher order variables of the translog functional form significantly reduce the degrees of freedom. To gain efficiency, the model was estimated using an aggregate variable for the total number of medical staff (physicians, nurses, and parmedical staff). The estimation results (2) still show that no variables inthe model are statistically significant. 99. The traditional production function estimated with a Cobb-Douglas functional form (3) has a slightly lower R2 (0.643) due to fewer variables, but it yields a significant coefficient for the physician variable of the appropriate sign. None of the other variables in the estimate of the traditional Cobb-Douglas production function are found to be statistically significant, other than the proportion of pediatric consultations. Very similar results were obtained for the estimate of the traditional Cobb-Douglas production function with the aggregate staff variable (4). 100. A likelihood ratio test of the significance of the higher order input terms in the translog model (Table A.6) failed to reject the null hypothesis that those terms are jointly equal to zero. Therefore, within the comparison of specifications for the traditional production function, the Cobb-Douglas functional form with the disaggregatedstaff variable is preferred. Comparison Group LR test statistic Prob > chi2 Conclusion Translog vs. Cobb- LR chi2( 15) = 22.37 0.0985 Fail to reject the nullhypothesis Douglas that the higher order input terms (disaggregatedstaff arejointly equal to zero; Cobb- variable) Douglas functional form is preferred. Translog vs. Cobb- LRchi2(6) = 5.17 0.5219 Fail to reject the null hypothesis Douglas that the higher order input terms (aggregate staff arejointly equal to zero; Cobb- variable) Douglas functional form is preferred. 46 (1) Traditional Production Function-Translog Functional Form Source I ss df MS Number o f obs = 134 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 28, 1 0 5 ) = 8 . 6 5 Model I 92.1794265 28 3.29212237 Prob > F = 0 . 0 0 0 0 Residual I 39.9601573 105 .380572927 R- squared = 0.6976 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0.6169 Total 132.139584 133 .993530705 Root MSE = . 6 1 6 9 1 lnconsult lnphys 6.151925 5.437026 1 . 1 3 0.260 -4.628694 16.93254 lnnurse -7.539458 6.783765 -1.11 0.269 - 2 0 . 9 9 0 4 1 5.911495 lnpararned ,1797572 ,6913796 0 . 2 6 0.795 - 1 . 1 9 1 1 2 1 1.550635 lnspace 1 , 9 7 1 4 3 1 2.150546 0.92 0 . 3 6 1 -2.292705 6.235567 Inequip2 -1,09357 1.092616 - 1 . 0 0 0.319 -3.260025 1.072884 input11 - ,440692 2,264761 - 0 . 1 9 0.846 -4.931295 4.049911 input22 1.820739 2.926112 0 . 6 2 0 . 5 3 5 - 3 . 9 8 1 1 9 9 7,622678 input33 .0663837 .0426845 1 . 5 6 0.123 - .0182518 .1510193 input44 ,0001724 .2759469 0 . 0 0 1 , 0 0 0 5469794 ,5473242 input55 . l o 5 7 0 1 3 .2020655 0.52 0 . 6 0 2 -- .,2949573 .5063599 input12 .1269393 2,335172 0.05 0 . 9 5 7 -4.503276 4.757154 input13 - .0905494 ,168241 -0.54 0 . 5 9 2 -.4241401 ,2430414 input14 - ,512825 ,6692206 - 0 . 7 7 0.445 -1.839766 ,8141158 input15 ,3156803 1 . 2 1 0.229 - .2435608 1 . 0 0 8 3 1 input23 1688801 .2070125 - 0 . 8 2 0.416 - .5793477 .2415874 input24 - -.,3823748 . 095411 .974195 - 0 . 1 0 0.922 - 2 , 0 2 7 0 6 1,836238 input25 - .5509507 .3802463 - 1 . 4 5 0 . 1 5 0 -1.304909 .2030075 input34 .0860194 1 . 4 3 0 . 1 5 5 - ,0472622 .2938587 input35 - ..1232982 0191572 .0358795 - 0 53 * 0 . 5 9 5 - . 0902996 .0519852 input45 ,2399645 ,1484086 1 . 6 2 0 . 1 0 9 - .0543024 .5342314 d ist-hosp - .000226 ,0022212 - 0 . 1 0 0 . 9 1 9 -.0046303 .0041782 r u r a l .1125404 .1379725 0 . 8 2 0 . 4 1 7 - . 1610336 .3861144 standalone .0705887 ,1500785 0 . 4 7 0 . 6 3 9 - . 2269892 ,3681667 catchsqkrn 6 . 5 3 e - 0 6 ,0002162 0.03 0 . 9 7 6 - .0004222 .0004352 popdensity .000037 0 . 3 9 0 . 6 9 7 ,0000878 lnconsped - ..0000144 0296301 .0164589 - 1 . 8 0 0 . 0 7 5 - -. 0000589 .062265 .0030049 lnconsf em .0154749 .0154919 1 . 0 0 0 . 3 2 0 - .0152426 .0461925 lnconssen .0025959 .0185615 0 . 1 4 0 . 8 8 9 - ,0342081 ,0394 -cons 8.703384 1 2 . 3 5 5 0 1 0 . 7 0 0 . 4 8 3 - 1 5 . 7 9 4 3 2 33.20109 (2) Traditional Production Function-Translog Functional Form with Aggregate Staff Variable Source 1 ss df MS Number o f obs = 134 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 17, 1 1 6 ) = 1 2 . 5 6 Model 85.6280637 1 7 5.03694493 Prob > F = 0 . 0 0 0 0 Residual II 46.5115201 116 .40096138 R-squared = 0.6480 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0.5964 Total I 132.139584 133 .993530705 Root MSE = .63322 47 1nequip2 - .6674284 .E455246 - 0 . 7 9 0 . 4 3 2 -2.342096 1,00724 inputaa .7367295 .4779858 1 . 5 4 0 . 1 2 6 -.2099815 1 . 6 8 3 4 4 1 inputab - .4293479 .3172511 - 1 . 3 5 0 . 1 7 9 ,199008 inputac -- ..0147692 2195491 .1345653 - 1 . 6 3 0 . 1 0 5 --1.057704 ,0469744 inputbb .2507734 - 0 . 0 6 0.953 - ..4860725 5114576 .4819192 inputcc .1594461 .2037047 0 . 7 8 0.435 - .2440167 .562909 inputbc ,2000753 .1402518 1 . 4 3 0.156 - ,0777111 .4778617 d i st-hosp - ,0010174 .0022221 - 0 . 4 6 0.648 .0054186 .0033837 r u r a l - ,0192557 .1291629 -0.15 0.882 -- .2750791 .2365678 standalone .1906912 .1445651 1 . 3 2 0 . 1 9 0 - .0956381 .4770205 catchsqkm ,0000213 ,0002173 0 . 1 0 0.922 .0004092 .0004517 popdensity 2 . 4 8 e - 0 6 .0000354 0 . 0 7 0.944 -- .0000727 lnconsped - ,0345661 .0166271 - 2 . 0 8 0.040 - .0016341 lnconsfem .0279629 .0154238 1 . 8 1 0.072 -- ..0000677 . 0674981 0025859 ,0585117 lnconssen -.009268 .0183211 - 0 . 5 1 0 . 6 1 4 - 0455553 * .0270192 -cons - .4272314 7.333201 - 0 . 0 6-- ------ 0 . 9 5 4 -14.95156 1 4 . 0 9 7 1 (3) Traditional Production Function-Cobb-Douglas Functional Form Source I ss df MS Number o f obs = 134 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 13, 1 2 0 ) = 1 6 . 6 0 Model 84.9204857 1 3 6.53234505 Prob > F = 0 . 0 0 0 0 Residual II 47.2190981 120 .393492484 R- squared = 0.6427 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0.6039 Total I 132.139584 133 .993530705 Root MSE = ,62729 lnphys .E783758 .2944835 2 . 9 8 0.003 .2953189 1.461433 lnnurse -2203022 .3256819 0 . 6 8 0 . 5 0 0 - .E651298 lnparamed .0185331 .0320075 0 . 5 8 0.564 - ..4245253 0448394 .0819057 lnspace - .1980679 .1360107 - 1 . 4 6 0 . 1 4 8 - ,4673596 .0712238 1nequip2 ,0145793 .054487 0.27 0 . 7 8 9 -.0933011 .1224598 d ist-hosp - .0016499 .002092 - 0 . 7 9 0 . 4 3 2 -.0057919 ,0024922 r u r a l .0488684 .1322805 0 . 3 7 0.712 - .2130377 .3107745 standalone .0803838 ,1456392 0 . 5 5 0.582 - ,2079716 ,3687392 catchsqkm ,0002024 - 0 . 5 1 0 . 6 0 8 -.000505 .0002965 popdensity -.-0001042 . 000014 .0000335 - 0 . 4 2 0 . 6 7 7 - .0000804 .0000524 lnconsped - .0350518 .0161221 - 2 . 1 7 0 . 0 3 2 - - .0031311 lnconsfem .0191465 ,015147 1 . 2 6 0 . 2 0 9 -..0669724 0108435 .0491365 lnconssen .0015316 .0181234 0 . 0 8 0.933 -.0343515 .0374148 -cons 9.770757 .9751572 1 0 . 0 2 0 . 0 0 0 7.840013 11.7015 (4) Traditional Production Function-Cobb-Douglas Functional Form with Aggregate Staff Variable Source I ss df MS Number o f obs = 134 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 11, 1 2 2 ) = 1 9 . 2 3 Model 83.7977461 11 7.61797692 Prob z F = 0 . 0 0 0 0 Residual 1I 48.3418376 122 ,396244571 R- squared = 0.6342 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0.6012 Total 1 132.139584 133 .993530705 Root MSE = ,62948 I n e q u i p 2 I .0542497 0 . 4 3 0.668 - .0840553 .1307301 d i s t - h o s p - ..0233374 0017181 .0020989 - 0 . 8 2 0.415 ----.0058731 .0024368 rural - ,0190353 .1264772 - 0 . 1 5 0 . 8 8 1 .2694096 .231339 s t a n d a l o n e .1381556 .1394716 0 . 9 9 0.324 .1379423 .4142535 c a t c h s q k m - .0000757 .0002019 - 0 . 3 7 0 . 7 0 9 .0003241 p o p d e n s i t y IIIII - .0000124 .0000334 - 0 . 3 7 0 . 7 1 1 - ..0004754 0000786 l n c o n s p e d - .0375841 .0160947 - 2 . 3 4 0 , 0 2 1 - ,0694451 - ..0000538 0057231 l n c o n s f e m II .0235478 .0149731 1 . 5 7 0 , 1 1 8 - . 0060931 .0531886 l n c o n s s e n II - .0051797 .0177206 - 0 . 2 9 0 . 7 7 1 - ,0402595 .0299001 1 0 . 8 4 0 . 0 0 0 6.715601 . - - - - ---c-o--n- -- - -8.216432- - - - -.7581485- --- - ---- s , -- ---- - - - - - - - . .-- - - - - - . --- - - -- - - - - - _ -9.717262_ - - - - - _ _ StochasticProduction Frontier Specifications 101. The results o f estimations o f alternative stochastic production frontier models are presented below inoutput tables (5) - (8). Inthe estimation o f the SF model with a translog functional form (5), as inthe traditional production function with a translog functional form, none o f the variables inthe model were found to be statistically significant, and the staff variables have particularly wide confidence intervals. Again, the translog model was estimated usingan aggregate variable for the total number o fmedical staff. The estimation results (6) still show that no input variables inthe model are statistically significant, but the proportion o f pediatric consultations and the proportion o f adult female consultations become statistically significant (p<0.05). The proportion o f pediatric consultations has a negative effect on DZ output, and the proportion o f adult female consultations has a positive effect on output. 102. The SF production function estimated with a Cobb-Douglas functional form (7) yields a significant coefficient for the physician variable (pchi2 Conclusion Translog vs. Cobb- 1ILRchi2(15)= 24.81 II 0.0525 Narrowly fail to reject the null Douglas hypothesisthat the higher order (disaggregatedstaff input terms arejointly equal to variable) zero; Cobb-Douglasfunctional form is preferred. Translog vs. Cobb- LRchi2(6) = 6.12 0.4102 Fail to reject the null hypothesis Douglas that the higher order input terms (aggregate staff arejointly equal to zero; Cobb- variable) Douglas functional form i s Dreferred. 49 103, A likelihoodratio test of the significance of the higher order input terms inthe translog SF productionfunction model (Table A.7) failedto reject the null hypothesis that the higher order input terms arejointly equal to zero, although only narrowly (p=0.0525). Therefore, the Cobb-Douglas functional form with the disaggregatedstaff variable is marginally preferred. (5) Stochastic Production Frontier Model-Translog Functional Form S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number of obs = 1 3 4 Wald c h i 2 ( 2 3 ) = 3 0 9 . 0 8 Log likelihood = -106.50093 P r o b > c h i 2 - 0 0000 I _ _ _ _ _ _ _ _ - _ _ _ _ lnconsult lnphys 5 . 2 7 1 7 7 8 4 . 3 7 9 2 1 9 1 . 2 0 0 . 2 2 9 - 3 . 3 1 1 3 3 2 1 3 . 8 5 4 8 9 lnnurse - 6 . 5 0 7 1 5 8 5 . 5 0 5 3 3 9 - 1 . 1 8 0 . 2 3 7 - 1 7 . 2 9 7 4 2 4 . 2 8 3 1 0 8 1nparamed . 1 7 5 7 7 5 6 . 5 6 3 3 4 8 9 0 . 3 1 0 . 7 5 5 - -. 9880576 , 928368 1 . 2 7 9 9 1 9 lnspace 2 . 6 6 8 1 8 6 1 . 8 6 5 4 6 5 1 . 4 3 0 . 1 5 3 6 . 3 2 4 4 2 9 I n e q u i p 2 - 1 . 0 4 6 6 9 9 .E941723 - 1 . 1 7 0 . 2 4 2 - 2 . 7 9 9 2 4 4 .7058467 input11 - .E48616 1 . 8 5 7 4 9 1 - 0 . 4 6 0 . 6 4 8 - 4 . 4 8 9 2 3 1 2.791999 i n p u t 2 2 1 . 7 3 3 4 3 2 2 . 5 9 3 2 4 2 0 . 6 7 0 . 5 0 4 3 . 3 4 9 2 2 9 6.816093 i n p u t 3 3 0 6 1 3 9 5 1 . 0 3 5 5 4 3 9 1 . 7 3 0 . 0 8 4 -- 0 0 8 2 6 9 7 ,13106 input44 -.. 0 7 9 1 1 0 1 . 2 3 7 1 3 6 1 - 0 . 3 3 0 . 7 3 9 5438884 ,3856682 input55 . 0 9 0 3 2 4 6 , 1 6 7 0 0 7 6 0 . 5 4 0 . 5 8 9 -- ..- 2 3 7 0 0 4 3 .4176536 input12 . 3 1 2 5 7 2 7 1 . 9 7 8 1 5 2 0 . 1 6 0 . 8 7 4 - 3 . 5 6 4 5 3 4 4.18968 input13 . 0 6 8 1 5 6 . 1 4 8 2 5 1 5 - 0 . 4 6 0 . 6 4 6 - .3587236 .2224115 i n p u t 1 4 -.-3 3 4 0 5 9 1 .5456563 - 0 . 6 1 0 . 5 4 0 -- . 1620523 1 . 4 0 3 5 2 6 .7354076 input15 3 5 3 4 4 8 1 .2630152 1 . 3 4 0 . 1 7 9 .E689485 input23 - .. 1 6 0 9 9 6 9 , 1 8 1 1 9 1 5 - 0 . 8 9 0 . 3 7 4 - .5161257 ,1941318 input24 - ,2455022 , 8 0 2 2 6 8 1 - 0 . 3 1 0 . 7 6 0 - 1 . 8 1 7 9 1 9 1 , 3 2 6 9 1 4 input25 - ,5217213 , 3 1 5 4 7 3 8 - 1 . 6 5 0 . 0 9 8 - 1 . 1 4 0 0 3 9 .096596 input34 , 1 0 6 5 2 8 4 . 0 6 8 8 4 1 1 . 5 5 0 . 1 2 2 0283975 .2414543 input35 - ,0111132 .029623 - 0 . 3 8 0 . 7 0 8 0 6 9 1 7 3 1 . 0 4 6 9 4 6 7 i n p u t 4 5 . 1 1 9 2 0 0 7 1 . 9 6 0 . 0 5 1 0 0 0 5 5 9 6 .4666987 lnconsped -..2330696 0257469 .0140708 - 1 . 8 3 0 . 0 6 7 0 5 3 3 2 5 2 .0018314 lnconsfem . 0 1 2 4 8 8 3 1 . 5 6 0 . 1 1 9 ----- ...., 0 5 0 0 1 2 .043952 lnconssen - ..0194754 0100124 . 0 1 5 1 9 6 - 0 . 6 6 0 . 5 1 0 -.0039796 .0197713 -cons 5 . 6 2 2 7 6 9 1 0 , 3 9 2 7 3 0 . 5 4 0 . 5 8 8 - 1 4 . 7 4 6 6 1 2 5 . 9 9 2 1 5 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - / l n s i g 2 v I - 2 . 0 5 4 5 7 4 .3588256 - 5 . 7 3 0 . 0 0 0 - 2 . 7 5 7 8 5 9 - 1 . 3 5 1 2 8 9 / l n s i g 2 u I - . 7 5 9 2 7 3 3 . 3 4 0 8 0 3 9 - 2 . 2 3 0 . 0 2 6 - 1 . 4 2 7 2 3 7 -.0 9 1 3 1 sigma-v . 3 5 7 9 7 6 8 .0642256 .251848 .5088284 sigma-u 1I .6841099 . 1 1 6 5 7 3 7 .4898685 .9553715 s i g m a 2 . . . . . . .lambda. . . . . .1...9.1.1.0.4.5. . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2...2.4.4.2 1I , 5 9 6 1 5 3 8 . 1 2 8 8 4 7 3 , 3 4 3 6 1 7 7 .E486899 . . . . . 1 6 9 9 8 5 7 1 . 5 7 7 8 7 9 1 1 Likelihood-ratio t e s t of sigma-u=O: c h i b a r 2 ( 0 1 ) = 6 . 8 4 P r o b > = c h i b a r 2 = 0 . 0 0 4 (6) Stochastic Production Frontier Model-Translog Functional Form with Aggregate Staff Variable S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 3 4 Wald c h i 2 ( 1 2 ) = 2 3 2 . 9 2 Log l i k e l i h o o d = -117.55236 P r o b > c h i 2 - 0 . 0 0 0 0 50 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - _ - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - l n a l l s t a f f 1 . 0 2 2 1 9 8 1 . 1 1 7 0 6 5 0 . 9 2 0 . 3 6 0 - 1 , 1 6 7 2 0 9 3 . 2 1 1 6 0 5 lnspace 2 . 3 4 9 6 2 3 1 . 5 8 5 1 9 1 1 . 4 8 0 . 1 3 8 - .757295 5 , 4 5 6 5 4 l n e q u i p 2 - . 7 3 0 6 1 7 5 .7327387 - 1 . 0 0 0 . 3 1 9 - 2 . 1 6 6 7 5 9 .7055239 inputaa , 6 6 2 9 3 3 1 ,4154815 1 . 6 0 0 . 1 1 1 - . 1 5 1 3 9 5 8 1 . 4 7 7 2 6 2 inputab - . 3 6 6 8 7 1 3 . 2 8 1 0 5 2 9 - 1 . 3 1 0 . 1 9 2 - . 9 1 7 7 2 4 9 .1839822 inputac - . 1 9 7 3 4 4 . 1 1 7 1 9 3 1 - 1 . 6 8 0 . 0 9 2 - . 4 2 7 0 3 8 3 .0323503 inputbb - . l o 4 8 2 1 3 .2315565 - 0 . 4 5 0 . 6 5 1 - . 5 5 8 6 6 3 8 .3490212 inputcc , 1 6 3 1 5 6 7 .1781605 0 . 9 2 0 . 3 6 0 - .1860313 .5123448 inputbc .1935872 .1205254 1 . 6 1 0 . 1 0 8 - .0426382 .4298127 lnconsped 11IIIIIIII - . 0 3 0 5 8 6 1 ,0149466 - 2 . 0 5 0 . 0 4 1 - . 0 5 9 8 8 1 -.0012912 l n c o n s f e m .0315592 . 0 1 3 2 1 4 9 2 . 3 9 0 . 0 1 7 .0056584 .0574599 lnconssen-cons 1 II - . 0 1 7 8 9 0 7 , 0 1 6 1 6 8 7 -1.11 0.269 - . 0 4 9 5 8 0 7 ,0137994 - 1 . 0 4 5 2 6 3 6 . 5 0 9 2 1 4 - 0 . 1 6 0 . 8 7 2 - 1 3 . 8 0 3 0 9 1 1 . 7 1 2 5 6 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - / l n s i g 2 v I - 1 . 7 3 9 9 9 9 .3037199 - 5 . 7 3 0 . 0 0 0 - 2 . 3 3 5 2 7 9 - 1 . 1 4 4 7 1 9 / l n s i g 2 u I - . 7 4 7 6 0 7 1 .3654972 - 2 . 0 5 0 . 0 4 1 - 1 . 4 6 3 9 6 8 - . 0 3 1 2 4 5 7 sigma-v , 4 1 8 9 5 1 7 .063622 , 3 1 1 1 0 0 4 .5641926 sigma-u . 6 8 8 1 1 2 1 .1257515 . 4 8 0 9 5 3 7 .9844985 s i g m a 2 . . . . . . .lambda. . . . . .1...6.4.2.4.6.2. . . . ..1771864. . . . . . . . . . . . . . . . . . . . . . . . .1.. . . . . . . . . . . .1...9.8.9.7 I111 . 6 4 9 0 1 8 7 .1396148 ,3753788 .9226586 . . . . . . . . . . . 2 9 5 1 8 3 4 1 Likelihood-ratio t e s t o f sigma-u=O: c h i b a r 2 ( 0 1 ) = 5 . 6 0 P r o b > = c h i b a r 2 = 0 . 0 0 9 (7) Stochastic Production Frontier Model-Cobb-Douglas Functional Form S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number of obs = 134 Wald c h i 2 ( 8 ) = 2 3 1 . 7 9 Log likelihood = -118,90632 P r o b > c h i 2 -- 0 . 0 0 0 0 - _ _ _ _ - _ _ _ _+_--- _- - - - - - - - - - - - - - - - - - - - - . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . - - - - - _ - - - - lnphys .E19724 .2557332 3 . 2 1 0 . 0 0 1 . 3 1 8 4 9 6 1 1 . 3 2 0 9 5 2 lnnurse . 2 6 4 4 9 8 6 . 2 8 9 6 0 4 5 0 . 9 1 0 , 3 6 1 . 3 0 3 1 1 5 8 .E32113 1nparamed I1I . 0 0 0 5 6 6 2 .0283709 0 . 0 2 0 . 9 8 4 -- . 0 5 5 0 3 9 8 ,0561722 lnspace . 1 8 2 3 0 2 9 ,1235168 - 1 . 4 8 0 . 1 4 0 - ,4243913 .0597855 1nequip2 11 I .0161645 ,0490804 0 . 3 3 0 . 7 4 2 - .0800314 .1123604 lnconsped .0309514 .0149152 - 2 . 0 8 0 . 0 3 8 - ,0601846 - . 0 0 1 7 1 8 1 lnconsfem --. 0 2 1 6 1 5 9 .0130698 1 . 6 5 0 . 0 9 8 - ,0040004 .0472323 lnconssen - . 0 0 7 6 2 0 3 .0162604 - 0 . 4 7 0 . 6 3 9 - . 03949 .0242494 -cons I 1 0 . 1 2 7 9 7 .9023769 1 1 . 2 2 0 . 0 0 0 8.359348 1 1 . 8 9 6 6 / l n s i g 2 v - 1 . 6 1 7 7 , 2 7 6 5 1 8 7 - 5 . 8 5 0 . 0 0 0 - 2 . 1 5 9 6 6 7 - 1 . 0 7 5 7 3 4 / l n s i g 2 u I1 - . E 5 9 3 2 7 5 .3990806 - 2 . 1 5 0 . 0 3 1 - 1 . 6 4 1 5 1 1 - . 0 7 7 1 4 4 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - sigma-v . 4 4 5 3 6 9 8 .Of515765 . 3 3 9 6 5 2 1 .5839926 sigma-u 11 II , 6 5 0 7 2 7 9 .1298464 . 4 4 0 0 9 9 .9621624 s i g m a 2 . 6 2 1 8 0 1 , 1 3 5 4 7 6 6 . 3 5 6 2 7 1 7 .E873304 lambda 1 . 4 6 1 0 9 5 ,1790212 1 . 1 1 0 2 2 1 . 8 1 1 9 7 1 Likelihood-ratio t e s t o f sigma_u=O: c h i b a r 2 ( 0 1 ) = 4 . 3 0 P r o b > = c h i b a r 2 = 0 . 0 1 9 (8) Stochastic Production Frontier Model-Cobb-Douglas Functional Form with Aggregate Staff Variable S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number of obs = 1 3 5 Wald c h i 2 ( 5 ) = 1 9 9 . 3 1 Log likelihood = -126.44537 P r o b > c h i 2 -- 0 . 0 0 0 0 51 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . lnconsult I C o e f . S t d . E r r . 2 P > l Z I [ 95% C o n f , I n t e r v a l ] - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - l n a l l s t a f f .956683 . 0 7 4 0 8 8 7 1 2 . 9 1 0 . 0 0 0 .E114717 1 . 1 0 1 8 9 4 I n e q u i p 2 . 0 0 5 8 1 9 7 .0514313 0 . 1 1 0 . 9 1 0 - .0949838 . l o 6 6 2 3 2 lnconsped - . 0 4 3 5 2 9 4 .0152646 - 2 . 8 5 0 . 0 0 4 - . 0 7 3 4 4 7 4 - . 0 1 3 6 1 1 4 l n c o n s f e m II IIII . 0 1 9 1 3 1 2 . 0 1 3 3 5 3 4 1 . 4 3 0 . 1 5 2 . 0 0 7 0 4 1 1 .0453034 lnconssen .000313 .0157098 0 . 0 2 0 . 9 8 4 -- , 0 3 0 4 7 7 7 .0311037 -cons 8 . 1 5 5 6 9 2 .3966978 2 0 . 5 6 0 , 0 0 0 7 . 3 7 8 1 7 8 8 . 9 3 3 2 0 5 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - / l n s i g 2 v I - 1 . 4 1 0 4 9 2 . 2 6 2 3 7 9 1 - 5 . 3 8 0 . 0 0 0 - 1 . 9 2 4 7 4 6 -.E962384 / l n s i g 2 u I - . 9 4 1 5 8 8 .4738148 - 1 . 9 9 0 . 0 4 7 - 1 . 8 7 0 2 4 8 - . 0 1 2 9 2 7 9 sigma-v I . 4 9 3 9 8 7 . 0 6 4 8 0 5 9 ,3819854 .6388285 sigma-u .6245062 . 1 4 7 9 5 0 2 ,3925372 .9935569 s i g m a 2 .6340312 , 1 4 4 6 3 5 1 .3505517 .9175108 lambda 1II 1 . 2 6 4 2 1 6 , 2 0 0 3 2 8 .E715801 1 . 6 5 6 8 5 2 Likelihood-ratio t e s t of sigma-u=O: c h i b a r 2 ( 0 1 ) = 2 . 8 6 P r o b > = c h i b a r 2 = 0 . 0 4 5 104. A comparison o f the results of the estimations o f the traditional production functions and the SF productionmodels indicates that the SF model i s preferred. First, both approaches yield similar estimates ofthe marginal productivity ofthe inputs,with only physicians showing a marginal productivity statistically significantly different from zero inboth approaches. Inaddition, the magnitude o f the coefficient is similar inbothestimation approaches. Second, likelihood ratio tests rejected the null hypothesis o f zero share o f the variance attributable to inefficiency (sigma-u=O) ineach specification o f the SF model. This indicates that the error term should include the stochastic inefficiency component. 105. The SF production model also appears to be robust to alternative specifications with regard to the inefficiency estimates, although the translog specification gives slightly higher estimates o fthe share o f inefficiency intotal variance than the Cobb- Douglas specification. The mean share o f the total variance attributable to the inefficiency term (sigma-u2/ sigma-u2+sigma-u2) ranges from 0.615 inthe Cobb- Douglas specification with the aggregate staff variable to 0.776 inthe translog specification with the disaggregated staff variable (Table A.8). Table A.8. Estimates of Inefficiency as a Share of Total Variance Across 106. Inaddition, a plot of the ranking o f DZs by the estimated production efficiency score shows little variation across alternative specifications (Figure A.3). The Translog specification does yield marginally more variation inefficiency levels, and slightly higher inefficiency levels among poor performers, but these differences do not have any practical significance. 52 107 Based on the overall comparison o f alternative production model specifications, it is concluded that the SF production model with a Cobb-Douglas functional form i s preferred, because (i)likelihood ratio tests rejected the nullhypothesis o f zero share o f the variance attributable to inefficiency; and (ii) likelihood ratio tests failed to reject the nullhypothesis that the input interactionterms inthe translog specification arejointly equal to zero. Furthermore, due to the small sample size and large number o f parameters, the translog specification was not able to detect a positive marginal product for physician labor, which was estimated by the Cobb- Douglas specification. Figure A.3: Ranking of DZs by Production Efficiency Score across Different Model Specifications - _ _ _ - - ~ ~ _ - - P) B 0.9 gv) 0.8 CL E 0.7 .- .-3EX 0.6 2 0.5 u- $ 0.4 0.3 5.- 0 20 0.2 0.1 0 20 40 60 80 100 120 140 160 DZ 108. The SF production frontier model with the Cobb-Douglas functional form and disaggregated staff variable was also estimated usingthe service point, rather than the DZ, as the unit o f analysis. All variables inthe model were deflated by the number of service points inthe DZ. The estimation results (9) are very similar to the model usingDZ as the unit of analysis, with a slight reduction inthe estimated marginal product o f labor (0.820 vs. 0.758). The ranking o f DZs by the estimated production efficiency score shows little variation using service point as the unit o f analysis (Figure A.4). (9) Stochastic ProductionFrontier Model-Cobb-Douglas Functional Form with Service Point as the Unit o f Analysis S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l N u m b e r o f obs = 1 3 4 Wald c h i 2 ( 8 ) = 2 9 4 . 1 2 Log likelihood = -119.24803 P r o b > c h i 2 - 0 . 0 0 0 0 53 _ - _ _ _ _ _ _ _ -+_- ----- - _ - - - - - - - -- - - - --- - - . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . _ _ _ _ _ _ _ - _ c lnphysprsvpt .7581315 ,2440227 3 . 1 1 0 . 0 0 2 1.236407 l n n u r s s v p t .3680271 .2634607 1 . 4 0 0 . 1 6 2 -..2798559 1483464 .8844005 l n p a r a s v p t I1I - . 0 1 2 6 4 8 2 ,0236073 -0.54 0 . 5 9 2 - . 0589177 .0336213 l n s p a c e s v p t .1509045 ,117238 - 1 . 2 9 0 . 1 9 8 - .3806868 ,0788778 l n e q u i p s v p t II - .027248 .0479218 0 . 5 7 0 . 5 7 0 - .0666771 .121173 l n c o n s p e d -.0315769 .0149816 - 2 . 1 1 0 . 0 3 5 - ,0609404 - .0022135 l n c o n s fem .0222799 .0130342 1 . 7 1 0 . 0 8 7 - .0032667 ,0478265 l n c o n s s e n -.0079308 .0162642 - 0 . 4 9 0 , 6 2 6 - . 039808 ,0239464 -c o n s III 9 . 5 9 6 5 9 1 .632841 1 5 . 1 6 0 . 0 0 0 8.356245 10.83694 / l n s i g 2 v - 1 . 6 0 8 2 3 7 .275203 - 5 . 8 4 0 . 0 0 0 -2.147625 -1.068849 / 1 n s i g 2 u I1 -.8606408 .4004277 - 2 . 1 5 0 . 0 3 2 -1.645465 -.075817 sigma-v .4474822 .0615742 ,3417033 .5860064 sigma-u .6503007 .1301992 .4392299 .962801 sigma2 .6231313 '1357444 ,3570771 .8891856 lambda III1 1.453244 .1793231 1.101777 1 . 8 0 4 7 1 1 Likelihood-ratio t e s t of sigma-u=O: c h i b a r 2 ( 0 1 ) = 4 . 2 5 P r o b > = c h i b a r 2 = 0 . 0 2 0 Figure A.4: Ranking of DZs by Production Efficiency Score across Different Units of Analysis (Service Point) 1 Cobb-DougiasServicePolniI I 0 ~ICobb-DouglasSDZ 0 20 40 60 80 100 120 140 160 DZ Traditional Cost Function Specifications 109. The results o f estimations o f alternative traditional cost functions are presented below in output tables (10) -(15). Inthe estimation o f the traditional cost function with a translog functional form (10), none o fthe variables were found to be statistically significant, with the exception o f population density (pchi2 I Conclusion Translogvs. Cobb- II LRchi2(6) = 7.65 I 0.2647 I Failto reject the null hypothesis Douglas that the higher order output terms arejointly equalto zero; Cobb- Douglasfunctional form is preferred. LRchi2(9) = 13.84 0.1281 Failto reject the null hypothesis Douglas that the higher order input interactionterms arejointly equal to zero; Cobb-Douglasfunctional form is preferred. (10) Traditional Cost Function-Translog Functional Form Source 1 ss df MS Number o f obs = 129 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 17, 111) = 3 . 1 0 Model 82.8137863 1 7 4.87139919 Prob > F = 0.0002 Residual II 174.605358 111 1.57302124 R- squared = 0.3217 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0.2178 Total I 257.419144 128 2.01108706 Root MSE = 1 . 2 5 4 2 h e x p t o t I Coef. S t d . E r r . [ 95% Conf , I n t e r v a l ] lnconsult - 2 . 9 9 9 4 0 1 1.940762 - 1 . 5 5 0 . 1 2 5 - 6 . 8 4 5 1 5 1 .E46349 Inlab3 - .0094253 .9689288 - 0 . 0 1 0.992 -1.929422 1 . 9 1 0 5 7 2 lndiag2 .0190828 .453271 0 . 0 4 0.966 - .E79104 .9172696 output11 .2749549 .1675201 1 . 6 4 0 . 1 0 4 - .0569975 .6069072 output22 .0194462 .0161413 1 . 2 0 0 , 2 3 1 - .0514312 output33 .0013201 .0157775 0 . 0 8 0 . 9 3 3 -.0125388 . 029944 .0325843 output12 - .0026792 .1414627 - 0 . 0 2 0.985 - .282997 ,2776385 output13 . 0 0 6 6 1 1 ,0705707 0 . 0 9 0 . 9 2 6 - .1332296 .1464516 output23 .0105421 - 0 . 8 7 0 . 3 8 6 - .0300658 ,0117141 lnconsped -- .. 0091759 0368855 .0353808 - 1 . 0 4 0 . 2 9 9 - ,1069949 ,0332239 55 lnconsfem .0171054 .0321345 0.53 0.596 - ,0465711 ,080782 lnconssen .0375989 0.76 0 . 4 5 1 - ,0460763 . l o 2 9 3 3 1 r u r a l -..0284284 .265754 -1.22 0.227 ----.0006064 ,8496552 .2035634 standalone -.3230459 0568452 ,3294928 - 0 . 1 7 0.863 .7097573 .5960668 catchsqkm .000209 .0004115 0 . 5 1 0.613 .0010244 dist-hosp - .0030714 ,0043312 - 0 . 7 1 0.480 .011654 .0055112 popdensity -.0002596 ,0000672 - 3 . 8 7 0 . 0 0 0 -.0003927 -.0001265 -cons 33.98678 1 3 . 9 9 9 7 7 2 . 4 3 0 . 0 1 7 6.245293 61.72826 (1 1) Traditional Cost Function-Granneman Functional Form Source I ss df MS Number o f obs = 129 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 20, 1 0 8 ) = 2 . 9 5 Model 90.9930843 20 4.54965421 Prob > F = 0 . 0 0 0 2 Residual II 166.42606 1 0 8 1.54098204 R- squared = 0.3535 1nexptot lnconsult - 1 5 . 7 3 6 6 1 1 8 . 1 1 3 4 7 - 0 . 8 7 0.387 -51.64065 20.16743 lnlab3 ,6447621 1.067923 0 . 6 0 0 , 5 4 7 -1.472047 2 . 7 6 1 5 7 1 lndiag2 - .1532164 ,5349269 - 0 . 2 9 0.775 -1.213534 .go71016 outputaa 1.264505 1.482565 0 . 8 5 0.396 - 1 . 6 7 4 1 9 7 4.203207 outputbb - ,074541 .0407871 - 1 . 8 3 0.070 - .1553881 .006306 outputcc - ,0114247 .0341531 - 0 . 3 3 0 , 7 3 9 -- .0791221 .0562728 outputaaa -.0314452 ,0407979 - 0 . 7 7 0,443 .1123136 .0494233 outputbbb .0053522 ,0025441 2 . 1 0 0.038 .0003093 .010395 outputccc .0035413 0 . 1 6 0.874 - .0064583 ,0075805 outputab -..0005611 0629922 .080498 -0.78 0.436 - .2225532 ,0965686 outputac ,035643 0 . 4 1 0 . 6 8 5 - ,0561501 .0851509 outputbc - ..0145004 0055937 .0054709 - 1 . 0 2 0.309 - . 016438 .0052506 lnconsped - ,03267 .0355014 - 0 . 9 2 0.359 . l o 3 0 3 9 9 ,0377 lnconsfem .0243423 ,0321363 0 . 7 6 0 . 4 5 0 -- .0393614 .0880459 lnconssen .0209361 .0373821 0.56 0 . 5 7 7 0531617 .0950339 r u r a l - .3250216 ,263146 - 1 . 2 4 0 . 2 1 9 .1965795 standalone - . 0686798 .3265123 - 0 . 2 1 0.834 --- ...E466227 7158838 ,5785243 catchsqkm ,0004131 0 . 1 8 0 . 8 5 7 - .0007444 .0008933 d ist-hosp 0039054 .0043075 - 0 . 9 1 0 . 3 6 7 - .0124437 .0046328 popdensity -- ..0000744 .0002792 .0000673 - 4 . 1 5 0 . 0 0 0 - .0004125 - .0001458 -cons 83.82586 75.4039 1.11 0.269 - 6 5 . 6 3 7 7 6 233.2895 (12) Traditional Cost Function-Cobb-Douglas Functional Form 1nconsult .49147 .1309315 3.75 0 . 0 0 0 .232167 .7507729 lnlab3 .0173462 .0331667 0 . 5 2 0.602 .0483388 .0830312 India92 .026185 0 . 9 9 0 . 3 2 5 -- .0259516 ,0777645 lnconsped - ..0259064 0368532 .034787 -1 . 0 6 0 . 2 9 2 - .lo5747 ,0320406 lnconsfem .0174279 .0307008 0 . 5 7 0 . 5 7 1 - .0433735 .0782293 lnconssen .016613 .0353676 0 . 4 7 0 . 6 3 9 - .0534307 ,0666567 56 r u r a l I -.3780416 .2568745 - 1 . 4 7 0.144 - . 886768 .1306848 standalone -.1584582 .2986551 -0.53 0.597 -.7499289 .4330125 c a t c h s q k m -.0005238 II II .0002635 .0003975 0.66 0 . 5 0 9 ,0010508 d i s t - h o s p -.0034592 .0042453 - 0 . 8 1 0 . 4 1 7 -.0118667 p o p d e n s i t y -.0002466 .0000654 - 3 . 7 7 0 . 0 0 0 -.0003762 - ..0049484 0001171 . . . . . . . -.cons. . . . . .12.76626. . . .1.714988. . . . . . . . . . . . . .0.000. . . . . . .9,369819 . _-1--- - - - - - I . . . . . . . . . . . . . . . . . . 7.44 . . . . . . . . . . 6 , 1 6 2 7 1 Stochastic Cost Frontier Specifications 112. The results o f estimations o f alternative stochastic cost frontier models are presented below inoutput tables (13) - (15). Inthe estimation o f the SF model with a translog functional form (13) and the Grannemanfunctional form (14), as inthe traditional cost function with a translog functional form, none o f the output variables inthe model were found to be statistically significant, and they have particularly wide confidence intervals. 113. The SF cost function estimated with a Cobb-Douglas functional form (5) yields a significant coefficient on the output variable for consultations (p c h i 2 -- 0.0017 57 1nconsult - 3 , 0 9 9 3 4 9 1 . 9 1 3 7 1 6 - 1 . 6 2 0 . 1 0 5 - 6 , 8 5 0 1 6 3 . 6 5 1 4 6 5 1 l n l a b 3 , 3 8 9 1 1 9 ,9414174 0 . 4 1 0 . 6 7 9 - 1 , 4 5 6 0 2 5 2 , 2 3 4 2 6 3 l n d i a g 2 .2179666 .4509746 0 . 4 8 0 . 6 2 9 - ,6659274 1 . 1 0 1 8 6 1 output11 .3247155 .1673142 1 . 9 4 0 . 0 5 2 - .0032143 .ti526453 output22 .0146896 . 0 1 5 8 4 0 1 0 . 9 3 0 . 3 5 4 - .0163565 . 0 4 5 7 3 5 7 output33 .0145506 0 . 0 7 0 . 9 4 2 - ,0274548 ,0295824 output1 2 --- ..0010638 0587569 .1370685 - 0 . 4 3 0 . 6 6 8 - . 3 2 7 4 0 6 3 ,2098925 o u t p u t 1 3 .0216184 .0701447 - 0 . 3 1 0 . 7 5 8 - , 1 5 9 0 9 9 6 .1158628 o u t p u t 2 3 .0163442 . 0 0 9 9 7 5 9 - 1 . 6 4 0 , 1 0 1 - .0358965 . 0 0 3 2 0 8 1 lnconsped - .0054311 ,0343125 - 0 . 1 6 0 . 8 7 4 - . 0 7 2 6 8 2 5 .0618202 lnconsfem - .0072884 .0312546 - 0 . 2 3 0 . 8 1 6 - . 0 6 8 5 4 6 2 ,0539694 lnconssen .0473487 ,037265 1 . 2 7 0 . 2 0 4 - ,0256893 ,1203868 - - _ - - - ---cons_ -+_- -3- 1-- - - - - - - - - _ _ - - _ - _. .-. .2...2.9. . . .0...0.2.2. . . . . . .4...5.2.9.1.1.1. . . . . .5.7... . . - - . 0 8 2 4 2 1 3 . 5 4 7 8 5 6 3 5 7 2 . . _ / l n s i g 2 v .4728047 ,1246959 3 . 7 9 0 . 0 0 0 - 2 2 8 4 0 5 3 , 7 1 7 2 0 4 1 / l n s i g 2 u - 1 0 , 2 5 0 4 3 8 4 6 , 8 3 1 5 - 0 . 0 1 0 . 9 9 0 - 1 6 7 0 . 0 1 1 6 4 9 . 5 0 9 sigma-v 1 . 2 6 6 6 8 4 , 0 7 8 9 7 5 1 1 . 1 2 0 9 7 9 1 . 4 3 1 3 2 7 sigma-u .005945 2.517187 0 s i g m a 2 . . . . . . .lambda. . . . . .,0046933. . . .2...5.2.2.6.6.6. . . . . . . . . . . . . . . . . . . . . . . -.4...9.3.9.6.4. . . . . . . . . . . . . . . . . 1III 1 . 6 0 4 5 2 3 .2007016 1 . 2 1 1 1 5 5 1 . 9 9 7 8 9 1 . . . . . . . 4 . 9 4 9 0 2 7 Likelihood-ratio t e s t of sigma-u=O: c h i b a r 2 ( 0 1 ) = 0 . 0 0 P r o b > = c h i b a r 2 = 1 . 0 0 0 (14) Stochastic Cost Frontier Model-Granneman Functional Form S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 2 9 Wald c h i 2 ( 1 5 ) = 3 6 . 6 4 Log likelihood = -211.48204 P r o b > c h i 2 - 0 . 0 0 1 4 lnexptot I C o e f . S t d . E r r . [95% C o n f . I n t e r v a l ] lnconsult - 3 . 4 7 9 6 9 7 1 7 . 9 4 2 1 4 - 0 . 1 9 0 . 8 4 6 - 3 8 . 6 4 5 6 4 3 1 , 6 8 6 2 5 l n l a b 3 1 . 0 4 7 2 5 5 1 . 0 2 7 4 9 2 1 . 0 2 0 . 3 0 8 - . 966593 3 . 0 6 1 1 0 2 l n d i a g 2 .2445287 ,5264096 0 . 4 6 0 , 6 4 2 - .7872151 1 , 2 7 6 2 7 3 outputaa . 2 4 7 7 6 0 9 1 . 4 6 9 3 2 8 0 . 1 7 0 . 8 6 6 --2 . 6 3 2 0 6 8 3 . 1 2 7 5 9 outputbb - .0738552 .0405432 - 1 . 8 2 0 . 0 6 9 1 5 3 3 1 8 4 . 0 0 5 6 0 8 outputcc 0 0 4 6 2 2 9 .0339408 0 . 1 4 0 . 8 9 2 -..0618998 . 0 7 1 1 4 5 7 outputaaa - .. 0023807 ,0403964 - 0 . 0 6 0 . 9 5 3 -.0 8 1 5 5 6 1 . 0 7 6 7 9 4 7 outputbbb ,0051309 ,0025148 2 . 0 4 0 . 0 4 1 , 0 0 0 2 0 1 9 ,0100598 outputccc - .0010945 ,0035284 - 0 . 3 1 0 . 7 5 6 - ,00801 , 0 0 5 8 2 1 1 outputab - . 0906107 ,0771892 - 1 . 1 7 0 . 2 4 0 - ,2418987 , 0 6 0 6 7 7 3 outputac - ,0036975 ,035122 - 0 . 1 1 0 . 9 1 6 - ,0725354 .0651403 outputbc - . 0098778 ,0050948 - 1 . 9 4 0 . 0 5 3 - .0198634 , 0 0 0 1 0 7 7 lnconsped . 0 0 2 6 4 6 ,0342695 0 . 0 8 0 . 9 3 8 -.-0616575 . 0 6 4 5 2 1 . 0 6 9 8 1 3 lnconsfem - . 0 0 0 6 5 2 1 ,0311258 - 0 . 0 2 0 . 9 8 3 . 0 6 0 3 5 3 2 lnconssen . 1 1 2 4 6 8 2 _ - _ _ _-cons- - - - - - - 1I .0401937 ,0368755 1 . 0 9 0 . 2 7 6 - .0320809 + - - ----- --- - - - _7-4_. - _ _ - ----------- . .0...6.7.9. . . . . .-.1.1.5...0.3.6. .1. 3 0 . 8 3 4 3 6 4 2 5 0 5 0 . 4 1 7 6 , 7 0 4 8 - -1- - ------ / I n s i g 2 v . 4 4 0 9 1 6 4 .1247033 3 . 5 4 0 . 0 0 0 . 1 9 6 5 0 2 4 , 6 8 5 3 3 0 3 / I n s i g 2 u II - 1 0 . 2 7 6 0 3 8 5 7 . 1 0 9 9 - 0 . 0 1 0 . 9 9 0 - 1 6 9 0 . 1 8 1 6 6 9 . 6 2 8 sigma-v 1 . 2 4 6 6 4 8 .0777305 1 . 1 0 3 2 4 1 . 4 0 8 6 9 7 sigma-u 1I 1I .0058693 2 . 5 1 5 3 3 4 0 s i g m a 2 1 . 5 5 4 1 6 5 .1944366 1 . 1 7 3 0 7 6 1 . 9 3 5 2 5 4 ____----______------____________________-------------------------------------- lambda . 0 0 4 7 0 8 1 2 . 5 2 0 7 8 5 - 4 . 9 3 5 9 4 1 4 . 9 4 5 3 5 7 L i k e l i h o o d - r a t i o t e s t o f sigma-u=O: chibar2 (01) = 0.00 Prob>=chibar2 = 1 . 0 0 0 58 (15) Stochastic Cost Frontier Model-Cobb-Douglas Functional Form S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 29 Wald c h i 2 ( 6 ) = 2 2 . 0 2 Log l i k e l i h o o d = -217.44106 P r o b > c h i 2 -- 0.0012 l n e x p t o t I C o e f . S t d . E r r . z P > l Z I [95% C o n f . I n t e r v a l ] - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - l n c o n s u l t ,5438858 .1249132 4 . 3 5 0 . 0 0 0 .2990605 . 7 8 8 7 1 1 1 l n l a b 3 .0041014 .0334023 0 . 1 2 0 . 9 0 2 - .0613659 .0695688 l n d i a g 2 III III . 0 1 0 1 5 0 1 .0262563 0 . 3 9 0.699 - ,0413113 .0616116 l n c o n s p e d -.0022149 .0347415 - 0 . 0 6 0.949 -.070307 .0658772 l n c o n s f e m - . 0 0 8 9 5 0 1 .0303311 - 0 . 3 0 0.768 .068398 ,0504978 l n c o n s s e n .0309559 ,0358212 0 . 8 6 0.387 -.-0392524 . l o l l 6 4 2 -c o n s I 12.06162 5.483936 2.20 0.028 1 . 3 1 3 3 22.80993 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - / l n s i g 2 v .533302 .1245433 4 . 2 8 0.000 .2892017 .7774023 / l n s i g 2 u 1I -13.9128 1 3 8 3 9 . 6 1 - 0 . 0 0 0 . 9 9 9 -27139.06 27111.23 sigma-v 1,305585 .0813009 1.155578 1,475064 sigma-u II .0009525 6.591236 0 sigma2 I I 1.704552 .2123923 1 . 2 8 8 2 7 1 2.120834 lambda .0007296 6 . 5 9 3 4 8 1 -12.92226 12.92372 L i k e l i h o o d - r a t i o t e s t of sigma-u=O: chibar2(01) = 0.00 Prob>=chibar2 = 1 . 0 0 0 115. A comparison o f the results o fthe estimations o f the traditional cost functions and the SF cost models indicates that the SF model is preferred. Bothapproaches yield similar estimates of the marginal cost o f output, with only consultations showinga marginal cost that i s statistically significantly different from zero inboth approaches. Inaddition, the magnitude o f the coefficient i s similar inboth estimation approaches. Although the likelihood ratio tests failed to reject the null hypothesis o f zero share o f the variance attributable to inefficiency (sigma-u=O) in each specification o f the SF model, which indicates that the error term does not include a stochastic inefficiency component, the SF production'functiondoes statistically confirm an inefficiency component, and the contrast between the production and cost function estimations i s a key result o f the study. 116. As inthe SF production model comparison, the choice o f functional form does not appear to affect the inefficiency estimates or the ranking o f DZs by efficiency score (FigureA.5). Although the Cobb-Douglas functional form yields the lowest efficiency scores, the difference i s less than 0.01 and not practically significant. The Cobb-Douglas specification is therefore preferred, becausethere is sufficient power to detect a positive marginal cost o f consultations. 59 Figure AS: Ranking of DZs by Cost Efficiency Score across Different Model Specifications 1.005 1.0045 I---- 1.004 I 1.0035 1.003 1.0025 - - ~ _ _ _ _ _ Translog ~ 1.002 Granneman 1.0015 -~ - _ _- - ___ - __- ~ , 1.001 - 1.0005 , I , I 1 I 60 ANNEXTABLE LISTOFPHCCENTERSSURVEYED,JANUARYDECEMBER 1: - 2007 # Name Municipality District 1 DZ Dr Milorad Vlajkovic Barajevo Beogradski 2 DZ Dr Sima Milosevic Cukarica Beogradski 3 DZ Milivoje Stojkovic Grocka Beogradski 4 DZ Mladenovac Mladenovac Beogradski 5 DZ Novi Beograd Novi Beograd Beogradski 6 DZ Milutin Ivkovic Palilula Beogradski 7 DZ Rakovica Rakovica Beogradski 8 DZ Savski Venac SavskiVenac Beogradski 9 DZ Sopot Sopot Beogradski I O DZ Stari Grad Stari Grad Beogradski 11 DZ Vozdovac Vozdovac Beogradski 12 DZ Zemun Zemun Beogradski 13 DZ Zvezdara Zvezdara Beogradski 14 DZ Djordje Kovacevic Lazarevac Beogradski 15 DZ Vracar Vracar Beogradski 16 DZ Bor Bor Borski 17 DZ Kladovo Kladovo Borski 18 DZ Negotin Negotin Borski 19 DZ Kucevo Kucevo Branicevski 20 DZ Petrovac Petrovacna Mlavi Branicevski 21 DZ Jovan Serbanovic Pozarevac Branicevski 22 DZ Veliko Gradiste Veliko Gradiste Branicevski 23 DZ Zabari Zabari Branicevski 24 DZ Zagubica Zagubica Branicevski 25 DZ Bojnik Bojnik Jablanicki 26 DZ Lebane Lebane Jablanicki 27 DZ Leskovac Leskovac Jablanicki 28 DZ Medvedja Medvedja Jablanicki 29 DZ Vlasotince Vlasotince Jablanicki 30 DZ Bac Bac Juznobacki 31 DZ Mladen StojanoviC Backa Palanka Juznobacki 32 DZ Backi Petrovac Backi Petrovac Juznobacki 33 DZ Becej Becej Juznobacki 34 DZ Dr Djordje Bastic Srbobran Juznobacki 35 DZ Temerin Temerin Juznobacki 36 DZ Titel Titel Juznobacki 37 DZ Veljko Vlahovic Vrbas Juznobacki 38 DZ Zabalj Zabalj Juznobacki 39 DZ Alibunar Alibunar Juznobanatski 40 DZ Bela Crkva Bela Crkva Juznobanatski 41 DZ Kovacica Kovacica Juznobanatski 42 DZ Kovin Kovin Juznobanatski 43 DZ Opovo opovo Juznobanatski 44 DZ Pancevo Pancevo Juznobanatski 45 DZ 1.oktobar Plandiste Juznobanatski 46 DZ Vrsac Vrsac Juznobanatski 47 DZ Lajkovac Lajkovac Kolubarski 61 48 DZ Ljig Ljig Kolubarski 49 DZ Mionica Mionica Kolubarski 50 DZ Obrenovac Obrenovac Kolubarski 5 1 DZ Osecina Osecina Kolubarski 52 DZ Ub Ub Kolubarski 53 DZ B B Kolubarski 54 DZ Bogatic Bogatic Macvanski 55 DZ Dr DarinkaLukic Koceljeva Macvanski 56 DZ Krupanj Krupanj Macvanski 5 1 DZ Loznica Loznica Macvanski 58 DZ Mali Zvornik Mali Zvornik Macvanski 59 DZ Sabac Sabac Macvanski 60 DZ Cacak Cacak Moravicki 61 DZ Lucani Guca Moravicki 62 DZ Ivanjica Ivanjica Moravicki 63 DZ Aleksinac Aleksinac Nisavski 64 DZ Doljevac Doljevac Nisavski 65 DZ Gadzin Han Gadzin Han Nisavski 66 DZ Merosina Merosina Nisavski 67 DZ Nis Nis Nisavski 68 DZ Razanj Razanj Nisavski 69 DZ Dr Ljubinko Djordjevic Svrljig Nisavski I O DZ Bosilegrad Bosilegrad Pcinjski 71 DZ Bujanovac Bujanovac Pcinjski 12 DZ Presevo Presevo Pcinjski 13 DZ Surdulica Surdulica Pcinjski 14 DZ Vladicin Han Vladicin Han Pcinjski 75 DZ Vranje Vranje Pcinjski 16 DZ Dr Obren Pejic Babusnica Pirotski 17 DZ Bela Palanka Bela Palanka Pirotski 18 DZ Dimitrovgrad Dimitrovgrad Pirotski 19 DZ Pirot Pirot Pirotski 80 DZ Vladimirci Vladimirci Podrinjsko-Kolubarski 81 DZ Smederevo Smederevo Podunavski 82 DZ Smederevska Palanka Smederevska Palanka Podunavski 83 DZ Dr MilanBane Djordjevic Velika Plana Podunavski 84 DZ Cuprija Cuprija Pomoravski 85 DZ Despotovac Despotovac Pomoravski 86 DZ Jagodina Jagodina Pomoravski 87 DZ Paracin Paracin Pomoravski 88 DZ Rekovac Rekovac Pomoravski 89 DZ Svilajnac Svilajnac Pomoravski 90 DZ Brus Brus Rasinski 91 DZ Cicevac Cicevac Rasinski 92 DZ Krusevac Krusevac Rasinski 93 DZ Sava Stanojevic Trstenik Rasinski 94 DZ Vlastimir Godic Varvarin Rasinski 95 DZ A A Raski 96 DZ Novi Pazar Novi Pazar Raski 91 DZ Raska Raska Raski 62 98 DZ Tutin Tutin Raski 99 DZ Nikola Dzamic Vrnjacka Banja Raski 100 DZ Dr Janos Hadzi Backa Topola Severnobacki 101 DZ Kula Kula Severnobacki 102 DZ Dr Marton Sandor Mali Idjos Severnobacki 103 DZ Subotica Subotica Severnobacki 104 DZ Ada Ada Severnobanatski 105 DZ Coka Coka Severnobanatski 106 DZ Kanjiza Kanjiza Severnobanatski 107 DZ Kikinda Kikinda Severnobanatski 108 DZ Novi Knezevac Novi Knezevac Severnobanatski 109 DZ Senta Senta Severnobanatski 1I O DZ Novi Becej Novi Becej Srednjebanatski 111 DZ Secanj Secanj Srednjebanatski 112 DZ Srpska Crnja Srpska Crnja Srednjebanatski 1I 3 DZ Zitiste Zitiste Srednjebanatski 1I 4 DZ Dr Bosko Vrebalov Zrenjanin Srednjebanatski 115 DZ Milorad Mika Pavlovic Indjija Indjija Sremski 1 I 6 DZ Pecinci Pecinci Sremski 1I 7 DZ Rurna Ruma Sremski 1I 8 DZ Sid Sid Sremski 1 I 9 DZ Sremska Mitrovica Sremska Mitrovica Sremski 120 DZ StaraPazova Stara Pazova Sremski 121 DZ Arandjelovac Arandjelovac Sumadijski 122 DZ Batocina Batocina Sumadijski 123 DZ Dr Vojislav Dulic Knic Sumadijski 124 DZ Kragujevac Kragujevac Sumadijski 125 DZ Lapovo Lapovo Sumadijski 126 DZ Raca Raca Sumadijski 127 DZ Sveti Djordje Topola Sumadijski 128 DZ Blace Blace Toplicki 129 DZ Kursumlija Kursumlija Toplicki 130 DZ Prokuplje Prokuplje Toplicki 131 DZ Zitoradja Zitoradja Toplicki 132 DZ Sjenica Sjenica Uzicki 133 DZ Boljevac Boljevac Zajecarski 134 DZ Knjazevac Knjazevac Zajecarski 135 DZ Sokobanja Sokobanja Zajecarski I36 DZ Zajecar Zajecar Zajecarski 137 DZ Apatin Apatin Zapadnobacki 138 DZ Odzaci Odzaci Zapadnobacki 139 DZ Sornbor Sombor Zapadnobacki 140 DZ Arilje Arilje Zlatiborski 141 DZ Evelina Haverfild Bajina Basta Zlatiborski 142 DZ Cajetina Cajetina Zlatiborski 143 DZ Nova Varos Nova Varos Zlatiborski 144 DZ Pozega Pozega Zlatiborski 145 DZ Priboj Priboj Zlatiborski 146 DZ Prijepolje Prijepolje Zlatiborski 147 DZ Uzice Uzice Zlatiborski 63 ANNEXTABLE VARIABLESINTHE PHCFACILITYSURVEY,JANUARY 2: - DECEMBER 2007 ~ Category Variable Output variables # of visits in differentservice categories: consultation preventive immunization dental visit home visit injection overnightstay physiotherapy diagnostic service laboratorytest # of visits by differentage groups: adults, age 60 and older women, age 16-59 men, age 16-59 children, age 0-15 # visits by diabetics # of individuals enrolledin the DZ Input variables total expenditure of the DZ total expenditureon specific inputs: personnel, other supplies, utilities, maintenance # of staff in each category (physicians, nurses, paramedical, administrative, technical) # of machines(x-ray, ultrasound, CT, other) # of beds # of rooms (consultatiodprocedure,laboratory, pharmacy, public space) # of vehicles Policy variables % of revenue from capitation externally-drivenstaffcuts (available from human resourcesstrategy) Other control whether the DZ is public or private variables whether DZ is part of a hospital radius ofthe catchmentarea distanceto the nearest hospital population density ruraliurban % ofrevenue from the health insurance fund % o f revenue from patients % of revenue from other sources iemographicstructure of DZ enrolledpopulation 64 ANNEXTABLE STAFFINGINPRIMARYHEALTH CENTERS,JANUARY 3: CARE - DECEMBER 2007 Staff Mean (range) Category All DZs Rural Urban Stand-alone I n Health Center Total number of staff Per DZ I 239.4 I 167.1 I 312.7 I 224.9I 276.9 (48- 1,276) (50 - 692) (48 - 1,276) (50 - 1,276) (48 - 810) Per servicepoint 35.8 30.5 41.2 24.5 65.0 (3.5 - 729) (3.5 - 584) (5.9 - 729) (3.5 - 195) (6.6 - 729) - Per 1,000 population 5.4 5.7 5.2 5.5 5.2 (2.8-9.8) (3.5-8.1) (2.8-9.8 (2.8-9.8) (3.1 -8.1) Total # of physicians PerDZ I 61.8 41.3 I 81.5 I 60.0 I 66(1 (0-348) (11- 198) (0-348) (0-348) (0-218) Per service point II 8.5 III 7.8 II 9.1 II 6.3II 13.8 (0 - 169) (0.58 - 169) (0 - 128) (0 -48) (0 - 169) Per 1,000 population 1.3 1.4 1.3 1.4 1.2 (0 -3.0) (0.73 - 2.1) (0 -3.0) (0 -2.6) (0 - 3.0) Total # of nurses Per DZ 85.8 112.1 (0 682) -118.0 (27 - 374) (0 -148.9 682) (0 -682) (0 -132.2 440) Per service point 16.6 15.3 17.9 12.1 27.7 (0 -263) (1.7 - 263) (0 -254) (0 -95) (0 -263) Per 1,000 population 2.7 2.9 2.5 2.7 2.5 (0 - 5.6) (1.5 -4.8) (0 - 5.6) (0 -5.6) (0 -4.8) Total # of paramedical staff Per DZ 3.5 1.9 5.0 3.1 4.2 (0.01 -44) (0.01 - 19) (0.01 -44) (0.01- 44) (0.01 - 19) Per service point 0.43 0.29 0.57 0.27 0.82 (0.0005 - (0,001 - 6) (0.0005 - (0.0006 - 5) (0.0005 - 10) 10) 10) Per 1,000 population 0.06 0.05 0.07 0.05 0.07 (0 - 0.34) (0 -0.30) (0 -0.34) (0 - 0.30) (0 - 0.25) Total # of administrative staff Per DZ 13.7 10.9 16.4 14.0 13.2 (0 - 83) (0 -66) (0 -83) (0 75) - (0 - 83) Per service point 2.7 2.4 3.1 1.7 5.2 (0 - 83) (0 - 66) (0 - 83) (0 - 13) (0 - 83) Per population 0.35 0.40 0.30 0.39 0.25 (0 - 1.2) (0 - 1.2) (0 -0.77) (0 - 1.2) (0 - 0.76) Total # of technical staff I I (0-256) (0-98) 1 I (0-256) (0- 175) (0 -256) Per service point 6.6 II 4.7 8.3II 3.9II 13.0 (0 256) - (0 - 80) (0 256) - (0 - 37) (0 256) - Per 1,000 population 0.88 0.98 0.78 0.92 0.78 (0 - 2.3) (0 - 1.7) (0 -2.3) (0- 1.8) (0 -2.3) 65 ANNEXTABLE QUESTIONNAIRE FORPRIMARY HEALTH 4: CARECENTERS Question Answer Definitions and Remarks 101 DZNumber Pre-assignednumber filled in by CESID) - 102 DZName Nameof the DZ Nameof the Municipality where DZ is located (e& 103 .Municipality Savski Venac, Jagodina.,,) Code Municipality 104 Town Nameof the town where DZ is located Nameof District where DZ is located(e.g. I05 District Sevemobacki, Toplicki, etc.,,) Code District "Rural" - oooulationliving . . .,incatchment area of the DZ Location (rural or urban) as classified by 1) rural is mainly rural; "Urban" -population living in 106 DZ 2) urban catchmentarea of the DZ is mainly urban 1 I O 111 112 113 114 115 122 square meters I 1Ithe patients All rooms in DZ whose mainouruose IS different than . . mentionedabove e.g.public space, waiting rooms, Public space, waiting area, and other administrative offices, change rooms, utility rooms, 123 room, sq. meters heating, maintenance,garage, kitchen if any, etc... 124 Visit date filled by CeSID 125 Name of interviewer filled by CeSID Code Interviewer 126 Person interviewed Name and position of the interviewedperson Code position of interviewed person 127 Name of personwho entered data I filled by CeSID Code of data entry person 66 Revenue from different sources of payers, in Serbian Dinars, from January 1,2007 to December31,2007 \Me: I f tou reccitctl in-kind donations, pleaw cstiniatc and inwrt cash-,altic of in-hind donation\ such a 4di'iigs Quest1 Answer Definitions and Remarks Do the numbers on revenues given below refer to the period from January 1,2007 to December 31,2007. Ifyes, please proceedto 201 question 205. 1) yes 2) no 202 Start date, DD.MM.YYYY Put in the date in the form dd.mm.yyyy. 203 End date, DD.MM.YYYY Put in the date in the form dd.mm.yyyy. 204 205 206 207 designedperiod. Only funds paidon behalfof MoF. Since MoF executes all paymentsfor Government of Serbia, paymentson behalfof other Ministeriesand institutionsshouldbe in "MoH' part for MoH, "Other government" for 208 Ministry of Finance others, etc... Total ammount of funds paidby MoHin designedperiod. Including infrastructure investments, vertical programs,prevention, vaccination... Excluding donations, projects and serviceson market (e.g. check-upsof MoH 209 Ministry of Health employees) Total ammount of funds paidby other Ministries indesignedperiod. Excluding 210 Other Ministries Idonations, projectsand services on market I Total ammount of funds paidby MunicipaUCity Government in-designed period Includinginfrastructureinvestments, vertical programs, prevention Excluding 211 Municipality/City donations, projects and services on market Total ammount of fimds paidby other Governmentinstitutions indesignedperiod (Distnct, Province) Including infrastructure investments Excluding donations and services 212 Other Government on market Total ammount of funds paid directly by individual patients indesignedperiod, including copaymentsandmarketedandon- demand services Including institutionsand companiespaying for marketedand on-demand etc....).Revenuesnot relatedto provision of Other (lump sum) healthcareservices. 220 rotal, Serbian Dinars should add up to sum over answers 206 to 219 ~~ ~~~ ~~~ ~~ ~ ~~ Control Total, sum of 206-219 0 Error for Total 0 Control HIF Revenues, sum of 206 and 207 0 67 Error for HIF Revenues 0 221 Money owed to DZ on January 1,2007 Total amrnount of funds owedto DZ at the beginningand at the end of the observed 222 Money owed to DZ on December 31,2007 period. Expenditures, in Serbian Dinars, from January 1,2007 to December 31,2007 Question Answer Description and Remarks Do the numbers on expenditures given below refer to the oeriod from Januarv 1.2007 to December 31,2007. 1) yes 301 302 303 304 305 306 307 308 309 Total amount of funds spent on goods and services in designedperiod.Should addup 310 Goods and services 311 to 318 Total amount of funds spent for drugs/pharmaceuticalsin designedperiod, 31I -Pharmaceuticals /Drugs including donations and projects Total amount of funds spent for medical materialin designedperiod, including 312 -Medical material Idonations and projects 313 314 (cleaningmaterial, medical and other equipmentmaintenance,phone, internet, 315 medical and nonmedical equipment(e.g. computers). Including donations and 316 donations and projectsand including 317 318 319 320 321 68 322 Total obligations January 1,2007 323 Total obligations December 31,2007 outstanding loans by DZ Staff, beds, equipments \ate: pleaw report rvcragc anniirl number for the period from Jaiiiirry 1,2007 to December31, 2007. Lg. ifjoii started the )ear aith 10 doctors but at the ciid of the year you had 8, the ruerage i59 doctors Question 401 402 403 404 405 406 407 408 409 Laboratoryroom all rooms in DZ whose - mainpurposeis linked to the laboratory -number of Laboratory room activities Pharmacyroom - allrooms in DZ whose main purpose is linked to the pharmacy -Pharmacy room activities Consultationand proceduresroom - all rooms in DZ whose main purpose is providing healthcareservicesto the 3 -Consultation and procedures rooms patients All rooms in DZ whose main purpose is different than mentionedabove e g public space, waiting rooms, administrative offices, change rooms, utility rooms, heating, maintenance,garage, kitchenif 4 -Public space and otber rooms any, etc 69 .i:I 9 9 l l I E /l I l s;z s; - N m b 0 v1 m 0v1 Ten most frequent reasons for visits in DZ, total number per year I'letise enter for the total of 12 months the most frequent reasons for visit (diagnoses) Rank Iis the most frequriitly swn diagnosis(c.g. respiratory inliwion) Rank 10 is the least freqiierit ainong the ten diagnoses Question 601 602 603 604 605 606 607 608 609 Irin ." " I " , I 1 I I 611 Total number of top 10diagnosis Year of Year of -X-ray machine Number Production Purchase Price 410 1 Generate list of machineswith following 2 information for every machine:(1) year of 3 production;(2) year purchase;(3) price 4 r; 9 10 11 12 13 73 Generate list of machineswith following information for every machine:(1) year of -Ultrasound machines 411 - IProduction IPurchase IPrice production; (2) year purchase; (3) price 1 I I I 3 4 5 6 8 9 I I I 10 11 12 13 18 ...20 19 25 26 *" I I I 74 Year of Year of -CT Production Purchase Price 412 1 Generate list of machineswith following information for every 2 machine:(1) year of production; (2) year purchase; (3) price 3 4 5 6 1 1 12 13 14 15 16 1I 1 3 I 1 26 21 28 -Vehicles Production Purchase Price 413 1 2 Generate list of machines with following information for every machine:(I)year of production; (2) year purchase; (3) price 19 20 28 29 30 75 Year of Year of -Other Production Purchase Price 414 1 Generate list of machineswith following information for every 3 machine:(1) year of production; (2) year purchase; (3) price 4 " I I I 9 76