Using measures of quality of care to assess equity in health care funding for primary care: analysis of Indonesian household data

Almost without exception, health systems worldwide provide health services that vary in terms of quality and access in ways that invariably favour higher income groups [1]. This occurs even in countries which have ostensibly achieved universal health coverage (UHC) and against a background of longstanding recognition of this type of disparity, countries are urged to ‘aim for affordable UHC and access for all citizens on the basis of equity and solidarity’ [2]. Nonetheless many low- and middle-income countries (LMICs) are implementing reforms to promote equity in health financing and delivery as a pathway towards UHC [3]. Measuring the distributional impact of these reforms is a high priority for these countries [4].

‘Benefit Incidence Analysis’, or BIA, is the traditional approach to estimating the distributional impact of government spending on health care [5]. It uses information on costs and the utilisation of health services to estimate the distribution of public spending across different socio-economic groups. BIA seeks to answer the question: who benefits from public expenditures on health care and by how much? Put differently, BIA measures by how much the income of a household would have to be raised if the household had to pay for the subsidized health services at full cost [6]. In practice, BIA studies estimate “benefits” or “public subsidies” to service users, who are typically ranked by socioeconomic status or some other variable of interest including geographical area, ethnic group or gender [7]. While BIA has traditionally focused on distribution of public sector subsidies, the analysis is increasing being extended to the private sector because of the growing and important role of the private health sector.

A key strength of a BIA is that it can provide a simple and transparent approach to assessing the extent to which public health spending benefits the poor. The approach, however, is not without its limitations. A key one that has been flagged by analysts is the failure to take account of variations in the quality of health services received by different individuals, leading to a potential under/over-estimation of the subsidy [8]. Increased evidence that the poorest segments of the population receive poorer quality of care [9], means that failing to take account of the quality of care in BIA could lead to a biased picture of who benefits most from government health spending.

Recently, Asante et al. attempted to address this critical methodological gap by introducing a quality score into the computation of BIA [8]. They developed a proxy measure for quality using area level deprivation indicators (availability of water, electricity, energy source for cooking, education, etc.). One limitation of their approach (acknowledged by the authors) is the use of deprivation indicators that are not directly related to health care quality. Second, they used area-level indicators by averaging the quality measures at the district level; this could not only mask variations in the quality of health services at the sub-district level, but most importantly across facilities used by individuals with varying socio-economic status (SES). In this paper, we address this important issue using data from linked household and primary health care facility surveys in Indonesia. We aim to provide a worked example of how to apply the quality-weighted BIA methodology using facility-level quality measures linked to individual utilization data.

Table 2 describes the basic characteristics of individuals in our sample. In total, we managed to link 784 individuals with 51 health facilities they visited, which represents about 19% of all the individuals in the sample who reported seeking primary health care in the previous month. Table 3 describes the health facilities surveyed: 84% were public health centres, and 16% private clinics. 37% offered inpatient services, and about half were open 24 hours a day. The average catchment of a health facility was 35,000 persons. All facilities were contracted with the BPJS-Health and therefore provided subsidised services to JKN patients. Quality scores were relatively high across all facilities, with an average of 82%. Scores for basic equipment were highest, with an average score of 99% (range 60 to 100%) compared to essential medicines with an average score of 60% (range 20 to 85%).

Table 4 presents the distribution of unadjusted and quality-adjusted subsidies. For the unadjusted subsidies, we found that the distribution of subsidies in the public sector was pro-poor (CI = − 0.04), while the distribution of subsidies in the private sector was significantly pro-rich (CI = 0.37). The overall distribution of subsidies was slightly pro-rich, but the CI was not statistically significant (0.032). When adjusting for quality, we found that the distribution of subsidies in the public sector became slightly less pro-poor (CI = − 0.03), while the distribution of subsidies in the private sector became more pro-rich (CI = 0.48). Overall the pro-rich distribution of total subsidies became statistically significant (CI = 0.037).

Figure 1 compares the mean level of need with the distribution of subsidies across wealth quintiles. The level of need was highest among the poorest quintiles, and the distribution of public subsidies was not proportional to the distribution of the need for health care. In other words, the poorest groups in our sample did not receive their fair share of public subsidies when considering their level of need. The level of inequality was slightly higher when using quality-adjusted subsidies, but the magnitude was small.

This study provides a worked example of how to apply a quality-weighted BIA methodology, and we summarise the main steps in Table 5 below. Our findings from Indonesia show that the distribution of subsidies for public primary health care facilities became less ‘pro-poor’ and subsidies for private primary health care facilities became more pro-rich after accounting for quality of care. The magnitude or the difference between the distributions of quality-adjusted and unadjusted subsidies was not large, and we believe that the gap between the two distributions is likely to be underestimated. As our measures of quality remain limited to structural indicators and do not include process or outcomes measures, we did not find large variations in quality across study sites though availability of essential medicines has the lowest score. More sensitive measures of quality and/or the inclusion of higher-level facilities such as tertiary care hospitals would have provided a more realistic picture of quality-adjusted subsidies in Indonesia, and potentially greater differences in quality-adjusted and unadjusted subsidies. Additionally, recent efforts to increase quality of care in Indonesia, such as accreditation of primary health care facilities, have certainly led to a standardisation of the basic level of infrastructure and equipment However, this study is aimed at illustrating in practice how to apply quality weights in BIA studies rather than producing precise quantitative estimates.

Indonesia has been the focus of few BIA studies [19]. The first study, published over 30 years ago, showed that primary health care was mildly progressive but hospital care was disproportionately used by the better-off [20]. Similar results were reported in 2001 [21]. A comparative analysis of Asian countries found that in Indonesia, the richest 20% of the population received more than 30% of the total subsidies, and that the distribution of health care utilisation was more pro-poor than the subsidy distribution [22]. The fourth study examined the marginal effects of decentralized public health spending on the benefit incidence, when the authority to manage public spending for health and other sectors was devolved to the district level [23]. This study found that increased public spending at the district level improved the targeting of public funds to the poor by increasing their utilisation of services and also their share of public expenditure. However, the authors concluded that effort to increase the use of health services by the poor was necessary, and that demand-side interventions, such as price subsidies or social health insurance, were needed.

To our knowledge, none of these studies took into account the quality of care that patients received. More recently, Sambodo et al. measured the benefit incidence of healthcare funding under JKN, taking into account regional variation in unit costs across districts [24]. As both primary and secondary care providers are paid prospectively and proportionally to the intensity of their activity under the JKN system, better-equipped service providers are more likely to receive larger provider payments. Sambodo et al. found that the distribution of benefits favoured the wealthier groups, but most importantly that standard BIA using national unit costs underestimates regional disparities in healthcare funding, and therefore underestimates the inequality in the benefit distribution. If one assumes that the variation in unit costs reflects the variations in quality of care (especially availability of basic amenities, basic medical equipment, essential medicines and diagnostic tools), then our findings are consistent with theirs in the sense that the level of inequality in the benefit distribution is underestimated if quality is not accounted for. However, this assumption is unlikely to hold if higher provider payments are not correlated with higher quality, but instead are reflective of higher level of inefficiency; hence the need to account for quality using robust measures.

A major strength of our analysis lies in the fact that we were able to link individuals with the facilities they visited. In most studies, data linkage is not possible at the individual level, since conducting a facility survey alongside a household survey can be resource-intensive. Some limitations should also be acknowledged. Due to time and budget constraints, only 51 primary healthcare facilities could be surveyed, and therefore data on quality was collected in only a fraction of health facilities that individuals in our survey visited in the previous month, making the picture incomplete. However, this study represents a methodological advancement by introducing quality weights into the BIA framework and we hope future studies will be able to validate these results with larger datasets. Another limitation is the use of supply-side readiness scores as a proxy for quality which do not take into account e.g. health systems responsiveness and people’s expectations [25]. Careful interpretation is needed since the concept of quality of care is considerably broader and more complex than the measure used here [26]. Inputs such as infrastructure, equipment, medicines, and diagnostic tests, are just one element or prerequisite to the provision of good quality care [27]. Finally, while concerns have been raised about the use of subjective measures of health care need, they tend to be more readily available to researchers than objective measures and have been validated in the analyses of inequalities in health [28].

From a methodological perspective, one of the challenges of accounting for quality of care in BIA is the lack of adequate data from LMIC settings or standardized measurement of quality, since incomplete and unreliable quality data are common, and they often poorly capture process and outcome measures of care [29]. Researchers often rely on secondary datasets made available through global agencies such as the WHO, World Bank or United States Agency for International Development. The Indonesian Family Life Survey, for example, conducts health facility surveys that incorporate various indicators, including structural and process indicators, which can be used to assess quality of care. However, quality data is collected in only a fraction of health facilities that individuals visited in the previous month, making the picture incomplete. Routine collection of quality indicators that can be linked to individuals are needed to enable a more comprehensive understanding of individuals’ pathways of care, including the quality of services they receive.

The implications of this study go beyond the methodological aspect. From a policy perspective, accounting for quality of care in health financing assessment is crucial in a context where quality of care is a nationwide issue. Recently, the World Bank conducted an assessment of a nationally representative sample of 686 Indonesian public and private primary health care facilities. Their report highlights significant gaps in the readiness of primary health care facilities to deliver a basic level of quality of care [30]. Additionally, large geographical inequalities in the quality of care were detected. In such a context, any health financing performance assessment is likely to be biased if quality is not accounted for.

Through this analysis, we have shown that accounting for quality in BIA studies may provide a more accurate picture of the level of inequality, since poor households may have no choice except to visit the lower quality health facilities in their communities. We recommend that future analysis looking at the level of inequality in the distribution of public health care subsidies should incorporate quality of care in order to get the most accurate picture of the health financing system. Table 5 provides ‘how to’ for future assessment of quality-adjusted BIA. Improvement of the method will lie in the scope of measurement (structural, process or outcomes) of quality of care using standardized indicators, as well as in the accuracy of linking individuals to the very facilities they reported visiting to avoid using area-level information. Policy should focus on strengthening and equalizing quality of care across all primary healthcare facilities, as recommended by the World Bank (World Bank Group, 2020).