Abstract
INTRODUCTION: Low household socioeconomic status is associated with unhealthy behaviours including poor diet and adverse health outcomes. Different methods leading to variations in SES classification has the potential to generate spurious research findings or misinform policy. In low and middle-income countries, there are additional complexities in defining household SES, a need for fieldwork to be conducted efficiently, and a dearth of information on how classification could impact estimation of disease risk. METHODS: Using cross-sectional data from 200 households in Kisumu County, Western Kenya, we compared three approaches of classifying households into low, middle, or high SES: fieldworkers (FWs), Community Health Volunteers (CHVs), and a Multiple Correspondence Analysis econometric model (MCA). We estimated the sensitivity, specificity, inter-rater reliability and misclassification of the three methods using MCA as a comparator. We applied an unadjusted generalized linear model to determine prevalence ratios to assess the association of household SES status with a self-reported diagnosis of diabetes or hypertension for one household member. RESULTS: Compared with MCA, FWs successfully classified 21.7% (95%CI = 14.4%-31.4%) of low SES households, 32.8% (95%CI = 23.2-44.3) of middle SES households, and no high SES households. CHVs successfully classified 22.5% (95%CI = 14.5%-33.1%) of low SES households, 32.8% (95%CI = 23.2%-44.3%) of middle SES households, and no high SES households. The level of agreement in SES classification was similar between FWs and CHVs but poor compared to MCA, particularly for high SES. None of the three methods differed in estimating the risk of hypertension or diabetes. CONCLUSIONS: FW and CHV assessments are community-driven methods for SES classification. Compared to MCA, these approaches appeared biased towards low or middle SES households and not sensitive to high household SES. The three methods did not differ in risk estimation for diabetes and hypertension. A mix of approaches and further evaluation to refine SES classification methodology is recommended.