RESUMEN
BACKGROUND: Environmental surveillance (ES) for poliovirus is increasingly important for polio eradication, often detecting circulating virus before paralytic cases are reported. The sensitivity of ES depends on appropriate selection of sampling sites, which is difficult in low-income countries with informal sewage networks. METHODS: We measured ES site and sample characteristics in Nigeria during June 2018-May 2019, including sewage physicochemical properties, using a water-quality probe, flow volume, catchment population, and local facilities such as hospitals, schools, and transit hubs. We used mixed-effects logistic regression and machine learning (random forests) to investigate their association with enterovirus isolation (poliovirus and nonpolio enteroviruses) as an indicator of surveillance sensitivity. RESULTS: Four quarterly visits were made to 78 ES sites in 21 states of Nigeria, and ES site characteristic data were matched to 1345 samples with an average enterovirus prevalence among sites of 68% (range, 9%-100%). A larger estimated catchment population, high total dissolved solids, and higher pH were associated with enterovirus detection. A random forests model predicted "good" sites (enterovirus prevalence >70%) from measured site characteristics with out-of-sample sensitivity and specificity of 75%. CONCLUSIONS: Simple measurement of sewage properties and catchment population estimation could improve ES site selection and increase surveillance sensitivity.