THE BANGLADESH EPI

The EPI in Bangladesh was established in 1979 and became fully operational in 1985. It aimed to reduce morbidity and mortality from six vaccine-preventable diseases. Therefore, a fully vaccinated child received six standard EPI antigens against diphtheria, pertussis, and tetanus (DPT), tuberculosis (TB), polio, and measles through eight vaccinations. Pregnant women were also given vaccinations to prevent maternal and neonatal tetanus. Since 1985, vaccination coverage has increased from 2 percent for all antigens to a reported 92 percent for BCG and 62 percent for measles (16). However, immunization coverage rates were much lower in urban compared with rural areas. Therefore, in 1988, the United States Agency for International Development implemented a program to strengthen vaccination services in urban areas in conjunction with the array of government and nongovernment funders and providers of service.

DATA AND RESULTS

Our sample was obtained by means of a 1999 cost analysis of EPI services undertaken in a random sample of 25 percent of the facilities (132 of 511) providing EPI services in Dhaka City Corporation. To be parsimonious, five outputs (the amount of doses given for DPT, TB, polio, measles, and TT in 1999) and one input (total program costs of the EPI by site) were specified. Only program sites with full information were included. The final data set consisted of 117 of a possible 132 total clinics. Hence, 89.3 percent of all clinics sampled were included. The type of missing data that resulted in sites being excluded from the sample included ownership form, type of vaccination site, duration of operation, as well as some of the outputs. The descriptive statistics are given in Table 1.

Turning next to our efficiency results given in Table 2, we found that overall efficiency (TE CRS) was only 0.33. In other words, if program sites were technically efficient and operated at the correct scale, costs on average could have been reduced by 67 percent without sacrificing the current level of outputs produced. By decomposing this overall measure into pure technical efficiency (TE VRS) and scale efficiency, we found that more of the overall inefficiency was due to sites incurring too much cost in producing the array of vaccinations rather than operating at the wrong size. However, both sources of this overall inefficiency must be addressed for these sites to become less wasteful of scarce resources.

Given the findings that the sites in this sample exhibited variable returns to scale, the types of diseconomies of scale were examined next. Table 3 shows that the majority of the program sites exhibited increasing returns to scale (suggesting that they are too small), 17 program sites exhibit decreasing returns to scale (suggesting that they are too large), and only six program sites were the “right” size.

In Tables 4 and 5, we assessed whether differences in efficiency followed systematic patterns due to factors beyond managerial control. Table 4 displays statistically significant differences between the efficiency of two ownership forms, and shows that scale efficiency is relatively greater in government-owned program sites. As outreach sites were statistically significantly less scale efficient than fixed sites, we infer that satellite sites are too small given the best practice frontier.

Although the EPI program has been in existence in Dhaka City Corporation since 1988, not all sites began providing EPI services at the same time. Table 5 shows that the length of time a program site has been in operation is positively correlated with scale efficiency.

DISCUSSION/POLICY IMPLICATIONS

The sites in our sample were, on average, relatively inefficient both in terms of technical inefficiency as well as scale inefficiency. To become technically efficient, program sites would have had to decrease their costs by an average of 50 percent, and if they had been operating at the right size, costs could have been reduced by a further 36 percent. Sites that were relatively more inefficient, on average, were not government-owned satellites. Therefore, the governmentally owned sites, perhaps due to more centralized control, appeared to be better at long-term planning. We also found that sites that had been practicing longer were relatively more scale efficient, which is perhaps attributable to a learning curve effect.

The presence of pure technical inefficiencies suggests that, if such cost data were used as the numerator of a cost-effectiveness ratio, a cost-effectiveness analysis would not reflect the minimum efficient point of production at a given level. However, to ascertain whether this outcome is likely to be the case, researchers need to begin using a larger sample size of provider units for costing, especially if results are intended for use beyond the geographical focus of an evaluation.

Our evidence provides empirical support to Jacobs and Baladi's (9) contention that assuming constant returns to scale might not be realistic. The presence of increasing returns has two particular implications. First, that this intervention cannot be treated as perfectly divisible within a population and retain the same level of incremental cost-effectiveness. Second, it suggests that, if constant returns to scale are assumed when increasing returns to scale exist, an intervention is likely to be overprovided in that form. Finally, the potential learning effect raises questions about how relevant it is to transfer cost-effectiveness ratios over time or across countries as levels of technology differ (12). Therefore, we conclude that ignoring the possible existence of technical inefficiencies and variable returns to scale would make the generalizability of cost-effectiveness ratios suspect and could worsen rather than improve the allocation of resources.