Production and cost of health services in developing countries

Missing data represents the highest barrier to estimate the efficiency and costs of providing health care services in low- and middle income countries. For this reason, previous studies on hospitals inefficiency suffer from many limitations: inputs are often expressed in terms of recurrent expenditures, which may inherently incorporate inefficiency. Hospitals output is mainly defined as outpatient visits and inpatient. This naïve categorization is likely to not adequately represent the complex production process of hospitals. No reliable case-mix information is available, so that national hospitals classification systems are often used with the aim of comparing homogenous facilities.

We use detailed facility-level data collected as part of the Access, Bottlenecks, Costs, and Equity project to improve previous analyses of hospitals efficiency in developing countries. The dataset contains information across a range of service delivery platforms from different low- and middle income countries such as Kenya, Zambia, Uganda and India. In each country we collected data in about 200 facilities over a 5-years period, resulting in what we believe to be the largest ever health facility costing exercise. In addition, about 12’000 exit interviews were conducted with the aim of gather information of consumer perception of health facility quality.

We specify a production model with 5 inputs and 7 outputs. Inputs include the number of beds as proxy for capital, and 4 categories for labor (doctors, nurses, other medical staff and administrative staff). As with respect to output, outpatient visits include: basic outpatient services, ART (Antiretroviral treatment), malaria, antenatal care and emergency. For inpatient services we use inpatient days, births and surgery. Non-discretionary variables such as demand-constraints, quality and electricity availability are also considered.

We exploit super-efficiency models and information about the frequency with which a facility is used as a peer to detect and remove outliers from the sample. To avoid biased efficiency estimates due to heterogeneous technology, we propose an innovative approach that adjust outputs across facilities. We first identify all pharmaceuticals and equipment related to the production of each output and build a score which reflects the extent to which technology is available in the facility.

We then use bootstrap DEA models using the adjusted-outputs to compute technical efficiency scores by controlling for measurement error and noisy data. We include minimal weight restrictions to reflect the relative importance of inputs and outputs in the production process of health facilities. Weight restrictions are chosen to maintain the radial nature of efficiency valid.

We finally use output weights provided by DEA to calculate the marginal rate of transformation between outputs. This information is critical to the estimation of average costs for each output.

Our analysis shows reasonable average costs for each output. We find evidence of important inefficiency (40% on average) with massive variation across facilities. Inefficiency substantially increases average costs to produce health services (35% on average). Also, we find evidence of efficiency increases over time of about 10%, likely due to the scale-up of ART treatment and related services. Additional evidence is necessary to assess the causal relationship.