The efficiency of routine infant immunization services in six countries: a comparison of methods.
Abstract Background Few studies have systematically examined the efficiency of routine infant immunization services. Using a representative sample of infant immunization sites in Benin, Ghana, Honduras, Moldova, Uganda and Zambia (316 total), we estimated average efficiency levels and variation in efficiency within each country, and investigated the properties of published efficiency estimation techniques. Methods Using a dataset describing 316 immunization sites we estimated site-level efficiency using Data Envelopment Analysis (DEA), Stochastic Frontier Analysis (SFA), and a published ensemble method combining these two approaches. For these three methods we operationalized efficiency using the Sheppard input efficiency measure, which is bounded in (0, 1), with higher values indicating greater efficiency. We also compared these methods to a simple regression approach, which used residuals from a conventional production function as a simplified efficiency index. Inputs were site-level service delivery costs (excluding vaccines) and outputs were total clients receiving DTP3. We analyzed each country separately, and conducted sensitivity analysis for different input/output combinations. Results Using DEA, average input efficiency ranged from 0.40 in Ghana and Moldova to 0.58 in Benin. Using SFA, average input efficiency ranged from 0.43 in Ghana to 0.69 in Moldova. Within each country scores varied widely, with standard deviation of 0.18–0.23 for DEA and 0.10–0.20 for SFA. Input efficiency estimates generated using SFA were systematically higher than for DEA, and the rank correlation between scores ranged between 0.56–0.79. Average input efficiency from the ensemble estimator ranged between 0.41–0.61 across countries, and was highly correlated with the simplified efficiency index (rank correlation 0.81–0.92) as well as the DEA and SFA estimates. Conclusions Results imply costs could be 30–60% lower for fully efficient sites. Such efficiency gains are unlikely to be achievable in practice – some of the apparent inefficiency may reflect measurement errors, or unmodifiable differences in the operating environment. However, adapted to work with routine reporting data and simplified methods, efficiency analysis could triage low performing sites for greater management attention, or identify more efficient sites as models for other facilities.