A novel cluster-based algorithm to estimate socio-economic status for health equity surveillance in Sub-Saharan Africa

Injury contributes an estimated 11% of disability-adjusted life years (DALYs) to the global burden of disease, which is more than human immunodeficiency virus (HIV), tuberculosis (TB), and malaria, combined1. Though this burden disproportionately affects sub-Saharan Africa (SSA)2, injury and other diseases addressed through surgical means are marked there by deep inequity in multiple sectors. There is a relative lack of funding, research, and support for public health and strengthening of clinical care compared to the estimated burden of disease potentially addressed through improvements in these areas 3–7. Within this inequity lies another pernicious disparity. In addition to being at higher risk for injuries, lower socioeconomic status (SES) populations in low- and middle-income countries (LMICs) are more likely to have worse outcomes after injury than those of higher SES backgrounds8. Acute, time-sensitive conditions such as injury and other surgical diseases are particularly stark revealers of inequity. In Cameroon, our analysis of data from a single institution pilot trauma registry suggested that individuals from lower SES strata were underrepresented in the hospital care-seeking population; among those that did present to the hospital, lower SES patients had higher injury severity scores (ISS) and were more likely to die than their wealthier counterparts9,10. Marked health disparities unmasked by injury research are likely to be part of a larger cross-cutting public health equity crisis. Current health disparities theory extends the understanding of health beyond biology to include social determinants of health11–13. Factors such as wealth, education, and environment contribute significantly to health outcomes14. As these are driven by upstream, human-made, social and political constructs, they are, by their nature, modifiable15,16. Health inequities are seen by the World Health Organization (WHO) and the United Nations (UN) as major challenges to improving the health of vulnerable communities and to populations as a whole17,18. The WHO has identified monitoring health inequities and intervention impact measurement as key strategies necessary for addressing disparities18. Health equity has also been identified as a cross-cutting metric of the Sustainable Development Goals that needs to be monitored in order to achieve success19. Currently, there is no widely-accepted, rapidly implementable SES metric to meet this urgent need. A critical barrier to addressing equity issues in resource- and time-constrained settings is the lack of population-appropriate, validated, easily implemented SES metrics applicable in diverse contexts. Measuring SES is a fundamental, but complex, aspect of monitoring equity that is especially challenging in LMIC settings. Braveman PA, Cubbin C, Egerter S, et al. consider SES as a “multidimensional construct comprising diverse socioeconomic factors (typically economic resources, power, and/or prestige)”20. Factors such as education, occupation, and wealth may seem related to, and are often used as surrogates of, economic status. However, education and occupation may affect health outcomes differently than wealth; though they may be correlated, these are not interchangeable metrics20. Common surrogates used to assess wealth include income, expenditures, and asset-based indices. Income can be difficult to assess in LMIC settings, where payment may not be monetary and inconsistent employment or self-employment are common21. Correspondingly, income may not be a reliable reflection of individual or household wealth. Data for methods such as consumption or expenditures are time-consuming and resource-intensive to collect 21. The inherent complexity of monitoring SES becomes even more challenging when attempted in the acute setting, where there is little time and tolerance for lengthy questionnaires. Of 47 trauma registries reviewed in LMIC settings, only 13 measured occupation and just three included education. Despite the suspected role of SES in injury in many contexts, none of the 47 trauma registries included an assessment of economic status22. Existing tools to measure SES are not feasible to regularly use in LMIC settings constrained by resources and time. In recognition of the increased need for health equity surveillance globally, the WHO has provided a Health Equity Monitoring Toolkit23. While a powerful resource to explore existing data sets (such as the DHS) and upload individual researchers’ data for analysis, this toolkit relies on DHS Wealth Index data to establish wealth quintiles; users need to have already established their own metric of SES or opted to utilize other preexisting methodology (such as the DHS’s). Similarly, while the commonly used Gini index is helpful in characterizing inequality using existing data, it does not address how to collect data to measure wealth at the individual level24. Although important adjuncts to SES research, these resources do not solve the critical issue of how researchers can accurately implement SES surveillance through primary data collection in the field. The Demographic and Health Survey (DHS), which has been conducted in 45 of the 54 countries in SSA25, uses the Wealth Index26 to measure SES. Of these 45 countries, 37 have data available in the last 10 years25. This nationally representative survey includes data on household characteristics, individual demographics, and diverse public health metrics. Developed by MEASURE DHS+ with support from the World Bank, the DHS methodology includes information on household assets and infrastructure, which are then applied to a principal components analysis (PCA) to create the Wealth Index26. We piloted an adapted DHS approach in a single- institution trauma registry in Cameroon27, in which participants answered 29 questions characterizing diverse household assets, such as cooking fuel, flooring material, and water source9,10. The DHS methodology allowed us to estimate wealth for our study population, explore comparisons with the DHS urban sample, and look for associations between clinical factors and SES. Our findings suggested a potential underlying disparity in access to trauma care9, so our collaboration wanted to continue to track SES in the trauma patient population. Collecting the 29 DHS Wealth Index variables required 24-hour/7-day a week research staff presence, which was not logistically or financially feasible for ongoing monitoring. Currently, researchers are forced to choose between two extremes: an inadequate, arbitrary surrogate of SES or a complex, lengthy methodology that is difficult and resource-intensive to implement. The lack of rapid, easily implemented tools to accurately measure SES in acute care settings is a significant barrier to progress in monitoring and ultimately mitigating health disparities in trauma. Important attempts to modify the DHS methodology to limit health equity surveillance resource utilization have been limited by challenges in accuracy, validation, and generalizability. Any measurement of wealth is imperfect. Wealth indices have been found to inconsistently correspond to consumption and display an “urban bias”, under-estimating wealth in rural settings28. Critics of the DHS Wealth Index also acknowledge that these inconsistencies could be reflective of consumption’s suboptimal metric of wealth in LMIC settings28. Despite these limitations, the DHS Wealth Index has been broadly used and accepted. Benefits of the DHS include its ubiquity, public availability, and inclusion of inherent potential comparison populations via its nationally representative samples. These advantages have prompted researchers to develop abbreviated versions consisting of fewer DHS variables for more feasible implementation. PCA methods consisting of fewer variables have low agreement with the original DHS Wealth Index and are subject to misidentification of non-linear relationships29,30. Regression with stepwise selection of individual asset variables was used to create a simplified linear model with only six variables31; however, when compared to the full Wealth Index, the misclassification error rate was 35%31. A common challenge is that index-based models of wealth force an assumption of linearity on the relationships between the wealth- associated variables, which likely interact in a more complex, multidimensional manner. Others have used subjective assessments (such as community focus groups) to develop subsets of DHS variables thought to be most contextually relevant32. While a participatory approach importantly incorporates a locally relevant lens on wealth, it is difficult to standardize and validate for use in diverse settings33,34. None of the abbreviated metrics derived from DHS has been widely adopted; however, these attempts at creating simplified asset-based wealth scales demonstrate both the common need for a simpler metric of economic status and the lack of a generally accepted, valid way of creating one. If health equity is to be addressed, equity-related metrics, such as wealth, need to be monitored in clinical and public health settings. Obtaining these data requires a statistically sound, logistically feasible, and broadly applicable metric of SES, which is currently lacking. We propose a five-year study to create an efficient method of tracking inequity in diverse clinical and public health settings in SSA and to make the resulting methodology freely available. While our group’s recognition of this need stemmed from the trauma setting, equity is a cross-cutting, pervasive, global health challenge. Between countries and regions, there exist marked disparities in health outcomes. In SSA, this disparity is further compounded by SES-related inequity within each country in disease risk, disease outcomes, and access to health care. A critical step towards improving health equity is to be able to identify it, measure it, and track it as interventions are implemented. Without a reliable, feasible, and efficient way to do this, efforts to reduce inequity are severely limited. To address this critical need, our team will draw on our deep expertise in data science, public health, and acute clinical care to apply and validate an innovative cluster-based algorithm for SES estimation, making the results freely available to other researchers. Potential benefits include low-cost, feasible, country-specific health equity surveillance methodology for baseline assessment to identify inequity, target vulnerable populations for intervention, and monitor the impact of interventions over time. This contribution is expected to be significant because it will create a free, publicly available method of SES estimation implementation in SSA to facilitate health equity surveillance across countries and diseases, unlocking the potential for health equity research across disciplines, conditions, and contexts in SSA.