Assessing frontline HIV service provider efficiency using data envelopment analysis: a case study of Philippine social hygiene clinics (SHCs)

HIV/AIDS is one of the gravest public health issues in the world. According to the 2014 Global Statistics [1], 1.2 million people died of AIDS-related illnesses, with about 36.9 million people infected with HIV, of which 2 million accounts for the annual newly-infected cases. The global response to date, through the prevention and treatment services for sufferers and population at risk, has succeeded in decreasing the number of People Living with HIV (PLHIV) steadily [2, 3].

In recent years, country-led frontline service delivery programs/interventions have been carried out to halt the epidemic across the globe [4‐6]. These programs/interventions, focusing on better access to basic prevention, care and treatment (PCT) services, have catered to both the PLHIV population and key populations such as, male having sex with male (MSM), female sex workers (FSW) and people who inject drugs (PWID). Each country devised their own intervention as a response to the epidemic. In Ghana, TB integration into the HIV services through “one-stop shops” were found to have improved HIV screening [7]. While in Asia, Hong Kong and the Philippines have social hygiene clinics (SHC), which provide basic PCT services to key populations, particularly for FSW [8].

With the increasing number of HIV/AIDS-related frontline health service providers, there is a growing need to assess whether these providers have been efficiently delivering the health services subject to the available resources. In literature, effectiveness/outcomes and costing studies of the frontline HIV/AIDS services have been extensively studied independently [9‐27]. However, efficiency studies, which account for both effectiveness/outcomes and costs at the same time, were relatively limited [10, 28‐34]. Efficiency studies provide information about the efficient/inefficient use of finite resources, which could be subsequently utilised as potential suggestions in the improvement of both management and policy. As a tool to analyse efficiency, data envelopment analysis (DEA) has been used most often, since it is non-parametric, which imposes no restrictive hypothesis on the data generating process and requires minimal assumptions about the technology [35, 36]. Although DEA has been widely used in different disciplines [37‐42], its utilisation in HIV/AIDS research is limited [28‐30]. HIV/AIDS research using DEA has been instrumental to various areas across the globe as a means for resource allocation and service delivery optimisation efforts. Health centers in Rwanda were performing at an average efficiency at 78%, which suggests that there are still room for improvement in service delivery [28]. Inclusion of efficiency determination in national HIV/AIDS programmes in resources-needs estimation, using macro-level data, can narrow the spending gap in the HIV/AIDS response [29]. Lepine, et al. [35] has observed that Avahan NGOs in India could have reduced their inputs by 43% given the outputs they reached. In a more comprehensive analysis by Zanakis, et al. [43], countries with lower population density that manage to provide better health system performance and per capita support with lower GNP and better media information, tend to have lower HIV/AIDS indicators. In Kenya, Omondi Aduda, et al. [44] has observed that significant improvement in the technical efficiency among the outreach facilities were attributable to the voluntary medical male circumcision using DEA.

Most of the HIV-related DEA studies have focused on the efficiency of the delivery of health services to the general population, with little or no focus on the key populations. Key populations such as the MSM, FSW and PWID also access the same basic health services but are at higher risk of acquiring sexually transmitted diseases. Their access to these health services may vary from the general population, hence, determining the efficiency of service delivery for these populations would provide insightful suggestions to current service provider.

To the best of our knowledge, this is the first Philippine case study which aims to determine the less efficient frontline service providers, SHCs, among the aggregate and key populations, as well as to identify its possible determinants of efficiency, as a mean to improve the SHC’s operational practice compared to the efficient ones in search for better management.

Table 1 shows the summary of the inputs and outputs as well as the area-specific meta-predictors. In the aggregate population, SHCs incur the greatest cost with the HIV test 380 (± 173), which constitutes the largest percentage (84%) of input costs. While for the output side, condom access and access to gram staining services were of the same percentage (at 50%). The wide standard deviation as observed in the input side with nearly 50% of the mean, is indicative of the variations among the cost of goods/services.

Table 2 shows both raw and bias-corrected technical efficiencies derived from DEA and bootstrapping, respectively. At the aggregate level, 3 SHCs were found to be inefficient, with Pasay as the lowest with a technical efficiency score of 0.54. When disaggregated into the key populations, MSM was observed to have the least average technical efficiency score at 0.81, which seconded FFSW (at 0.84) and has still more means to improve compared to the perfect average efficiency score of RFSW (at 1.00). Though there is an overestimation of the efficiencies, with values of the bias-corrected DEA efficiencies lower than the raw technical efficiencies, the ranking of the SHCs with respect to the efficiency scores are still the same, in both the aggregate and key populations. Consistent and evident low efficiency can be observed in Pasay SHC.

Table 3 shows the input and output adjustments needed for the SHC to operate on the efficient frontier line. At the aggregate level, Baguio has much to improve in relation to the reduction of its input costs for gram stain (− 0.50 Php), condom (− 4.90 Php) and particularly HIV testing (− 465 Php), and likewise the increment the number of people accessing its condom services (+ 139 pax). Across the key populations, there is no clear pattern of reductions or increments among the SHCs, however, Baguio and Pasay SHCs have been observed to need more adjustments compared to the remaining SHCs. On the other hand, from among the 9 SHCs, Davao and Quezon City SHCs don’t need any change in their input or output components, which may be partially indicative, yet inconclusive signs of good management performance.

Income and HIV prevalence, through AMTP 5 indicator, were used as the area-specific meta-predictors, instead of the output dataset’s candidate meta-predictors. HIV prevalence was observed to be negatively associated with the efficiency scores, with notable significant inverse relationship observed at the aggregate level (at − 0.60). On the other hand, income is positively associated to the efficiency scores, with a significant association observed in the FFSW (at 2.42e^− 4). We have observed a positive association with income and negative association with HIV prevalence, as shown in Table 4. RFSW results were not shown due to the homogeneity of the efficiency scores, which can’t be analysed using Tobit regression.

The frontline service delivery is an effective response to the HIV/AIDS epidemics [4, 5]. In the Philippines, SHCs operate as one of the major frontline health service providers. In this case study, we have successfully identified the potentially inefficient use of finite resources among sampled SHCs. As exemplified by Pasay SHC, DEA was able to objectively identify its inefficiency among the SHCs, and even across the different key populations. Likewise, further analyses showed that both income and HIV prevalence were statistically related to efficiency.

Efficiency is an important index for benchmarking a DMU’s performance. Performance, according to ranking, based on either input or output variables (Table 1), would only provide limited information of the DMUs performance. Through DEA, inefficiency can be detected and be resolved by adjusting either the input or output side. At the aggregate population, in Table 2, an example of Pasay SHC, with the lowest efficiency score even after bootstrapping, can overcome this inefficiency by following the adjustments/suggestions of DEA, such as finding cheaper unit cost of inputs (e.g. 12.60 Php cheaper gram stain service and 58.30 Php cheaper HIV testing kit as shown in Table 3). Pasay SHC’s inefficiency can be attributed to the higher priced inputs in relation to the outputs (services provided) (shown in Table 3), which may be reflective of the bargaining transactions of the local government unit (LGU). The devolved functions of the government system in the Philippines is also apparent in health, particularly in the procurement of resources. LGUs have jurisdiction over the procurement of resources in their respective areas. While the procurement process is a standard protocol across LGUs, the bargaining transactions may vary, thus resulting to varied procurement prices. In achieving optimum efficiency, aside from Pasay, other inefficient SHCs in both the aggregate and key populations, such as Baguio, Cagayan de Oro, Cebu and others, can also benefit from DEA’s input reduction and output increment features (as shown in Table 3). The potential adjustments in the inputs/outputs provide an indication of future directions regarding SHC management. In Table 3, inputs can be reduced, or outputs can be increased per specific units to achieve optimum efficiency. While these DMUs are currently operating away from the efficient frontier, there are opportunities where resource use (input) and service delivery (output) can be optimised to approach the frontier, and prospectively become efficient. Specifically, the temporal changes in technology, in terms of the operational management of the respective inputs and outputs, provide an opportunity for DMUs to optimise their efficiency. In Nepal, Silwal and Ashton [76] notes the potential for hospitals to increase their outputs given the additional resources through the years. While in Greece, Polyzos [77] observed that middle-sized hospitals were able to achieve 100% technical efficiency in the span of 2009–2011 as a result of technological changes (i.e. spending cuts and constant reforms).

It should be noted that although we assume SHCs as DMUs, Philippine SHCs don’t decide independently, instead, they are under the mandate of the LGUs. In most cases, bargaining of goods/services are dependent on the LGUs, rather than the SHCs. The complicated functionality of the SHC-LGU nexus in relation to decision-making and bargaining of goods, is further affected by the uncertainty introduced by the local market. Market price dictates unit cost, which then affects the DMUs’ decision. These interrelated complexities in the functional and structural relation of SHCs, LGUs and local markets were not accounted, and brings us to one of the limitations of the study, which assumes that SHCs are DMUs, and that unit costs of goods/services from local markets have low variance in a short time horizon, thus such complexities were assumed to be null in this study.

Studies conducted in the field of HIV service facility efficiency assessment focused more on the aggregate population [28‐30], which assumes a homogeneous population rather than the disaggregated, heterogeneous populations. Santos, et al. [48] notes that aggregation of data might undermine the performance of other DMUs providing the same services. In the same note, Gori, et al. [78] observed that disaggregated data provides better estimation of the effects than aggregated models. Disaggregation of service provision can provide better insights regarding the utilisation of resources per service, compared to aggregate ones [79]. This scenario was also observed in this case study, wherein among the key populations, efficiency estimates were varied, with some estimates consistently lower than the aggregate population; a case exemplified by the Pasay SHC. This suggests that Pasay SHC, should not only focus at the aggregate level, but most importantly at the disaggregated key populations in addressing the issue of inefficiency. We have also observed that SHCs performed better in providing services to the RFSW compared to the other two key populations. The difference in the performance (in terms of the efficiencies) may be related to the nature of the access by key populations. RFSW as the term implies are registered under the SHCs, whereby they are required to access the services in a timely manner. MSMs and FFSWs, on the other hand, are populations which access the services in a discretionary manner.

After bootstrapping the DEA efficiency scores (as shown in Table 2), we have determined that though there is consistent overestimation of results in the raw efficiencies, the ranking of the efficiencies even after being bias-corrected are relatively the same, which only shows that the results of this study are robust. We further examined the variations among the efficiencies by determining which exogenous area-specific, meta-predictors may be able to explain the difference among the efficiency scores. Both income and HIV prevalence (through the AMTP classification), were observed to be related with the efficiency scores. As shown in Table 4, income is positively associated with efficiency, an increase in the income of the local area may contribute to the increase in the efficiency of the SHCs. Similarly, if there is a larger proportion of income, this may translate to higher allocation for the specific programs, such that of HIV PCT [80, 81]. The devolved system of government of the country gives LGU exclusive allocative powers with respect to the management of its territorial resources as well as its own transactions. The crude assumption of greater income equates greater allocation is anchored on the premise of prioritisation. We can assume that everything has a proportional increase or decrease based on income and the level of prioritisation.

On the other hand, high prevalence areas have been significantly associated to less efficiency, in the aggregate population, as shown in Table 4. The same trend was observed in the statistically non-significant coefficients of MSM and FFSW. This inverse association is an eventual paradox to the usual understanding of the economies of scale. We believe that this paradox might have been due to two following limitations of the data sources: 1) the nature of the data, leaning towards a capital-intensive perspective, and 2) non-inclusion of treatment unit costs due to lack of complete data across SHCs. The capital-intensive approach did not include the number of health personnel in the SHCs. Highly prevalent areas may have greater number of service personnel, but since the available data included only the service delivery time and the salary wage per health personnel, labour, such as actual number of health service providers, were not accounted. Likewise, after thorough data management, most treatment variables 1in the input data source were not included in the study, as only few SHCs have sufficient unit cost data for treatment services. This may be attributable to the paradox, whereby highly prevalent areas may provide more treatment services, but the model was only able to capture the prevention and control intervention components.

With the various data management techniques used in this study, caution should be regarded with the limitations in the 1) specifications of the protocols, 2) generalisability of the results, and 3) data availability. The protocols in data generation, data matching, data imputation, and variable selection are based on the published literature and the observed plausibility for the analysis to be carried out. Since this study is a case study of the Philippine SHCs, the number of SHCs used in this study compared to the 70 SHCs all over the country may prove that the results acquired can’t be fully generalised with other units. We also acknowledge the possibility that the locations whereby the data are available may already be performing better than the rest of the SHCs with no available data, which, in turn may bias the results towards efficient performers. Though this may be the case with the current data source, there is also another possibility whereby the SHCs with no data may perform better than the study SHCs. While this possibility is plausible, we can’t ascertain the validity of this assumption especially with the limited data. As mentioned earlier, some results of this study are constrained by the data availability of actual number of health service personnel and the complete unit cost data of the SHCs. The inclusion of these variables into DEA warrants further investigation.

We were able to determine the inefficiently performing SHCs in the Philippines. Though currently inefficient, these SHCs may adjust their inputs and outputs to become efficient in the future. While there were indications of income and HIV prevalence to be associated with the efficiency variations, the results of this case study may only be limited in generalisability, thus further studies are warranted.