Two-stage cluster sampling to assess SARS-CoV-2 seroprevalence without pre-enumeration: An example from Madagascar

Implementing population-based surveys in resource-constrained settings presents logistical challenges when detailed population enumeration is unavailable. We developed a field mapping system integrated into a cluster sampling approach to eliminate pre-enumeration requirements for a SARS-CoV-2 seroprevalence survey in Madagascar. We conducted a cross-sectional observational study in urban Fianarantsoa, Madagascar, between February and June 2021. Using probability proportional to size sampling, we selected clusters from administrative areas (fokontany) and randomly generated GPS coordinates within these clusters. Field teams navigated to coordinates using OpenStreetMap software on tablets, identified eligible households, and conducted health surveys with blood sampling. We employed a mobile-compatible system for real-time household mapping and data collection, functioning without continuous network connectivity. Sample size calculation targeted 650 households (SARS-CoV-2 seroprevalence 30%, precision ±5%, design effect 2.0). Our specific objectives were to develop and implement a geographic cluster sampling method that did not require pre-enumeration; to assess the feasibility of this method through participation rates; and to evaluate potential selection biases related to socioeconomic factors. We identified households at 95.3% (696/730) of randomly generated GPS coordinates. Of contacted households, 96.8% (674/696) participated, representing 1,121 individuals across 57 clusters. Participation rates varied geographically, with a modest inverse correlation with household wealth (participation decreased by 0.85% per wealth quintile increase, 95% CI: -3.54% to 1.84%). Demographic characteristics of our sample matched census data for urban Fianarantsoa, supporting the representativeness of our approach. This integrated field mapping system created a virtual household map simultaneously with survey implementation, enabling cost-effective two-stage cluster sampling without pre-enumeration. The approach enabled evaluation of selection bias, simplified logistics, and provided a permanent geo-referenced database of surveyed households. This methodology offers a practical solution for population-based surveys in resource-constrained settings with incomplete enumeration data and has applications beyond COVID-19 research for various public health surveillance activities.