Assessing trends in non-coverage bias in mobile phone surveys for estimating insecticide-treated net coverage: a cross-sectional analysis in Tanzania, 2007-2017.

Abstract

Introduction: Monitoring insecticide-treated net (ITN) coverage and use generally relies on household surveys which occur on a relatively infrequent basis. Because indicators of coverage are used to forecast the need for ITNs and aid in planning ITN distribution campaigns, higher frequency monitoring could be helpful to guide programme strategies. The use of mobile phone-based survey (MPS) strategies in low-income and middle-income countries has emerged as a rapid and comparatively inexpensive complement to large-scale population-based household surveys, considering the dramatic growth trend of mobile phone ownership.

Methods: The potential for non-coverage bias in the calculation of ITN coverage estimates from MPSs was assessed through the use of five consecutive Tanzania-specific Demographic and Health Surveys (DHS). Primary comparisons were made between all households included in the data sets (the reference standard) and mobile phone-owning households (the comparator). Deviations in ITN coverage estimates between the reference standard and mobile phone-owning households were used as a proxy for assessing potential non-coverage bias, with estimates calculated using a bootstrap method.

Results: By the 2017 DHS, regional measures of non-coverage bias for ITN coverage indicators rarely exceeded a ±3 percentage point difference when comparing mobile phone-owning households to the overall sample. However, larger differences were observed when comparing mobile phone-owning households to non-mobile phone-owning households, particularly in periods without recent mass ITN distributions.

Conclusion: Results suggest that MPSs can reliably estimate ITN coverage at the population level when both ITN coverage and mobile phone ownership are high. However, as ITN coverage declines, the gap between phone-owning and non-phone-owning households widens, indicating potential non-coverage bias and underscoring the need for caution in interpreting MPS data under such conditions.