Concordance of coverage estimates from Routine and Survey data of Measles Second Dose Vaccine in Western Kenya

Angela Moturi^1,2Moses Musau Mutinda^2*Samuel K. Muchiri²Peter M Macharia^2,3Bob Snow^2,4Emelda A Okiro^2,4

• ¹Faculty of Public Health and Policy, London School of Hygiene & Tropical Medicine, London, United Kingdom
• ²Population and Health Impact Surveillance Group, KEMRI-Wellcome Trust Research Programme Nairobi, Nairobi, Kenya
• ³Department of Public Health, Instituut voor Tropische Geneeskunde, Antwerp, Belgium
• ⁴Centre for Tropical Medicine and Global Health, Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom

Background: Missed opportunities for key vaccinations continue to exacerbate disease outbreaks. Accurately monitoring immunisation coverage is fundamental to identifying gaps in vaccine delivery and informing timely action. This study assesses the agreement between routine and survey-based coverage estimates for the second dose of the measles vaccine (MCV2) in Western Kenya. Methods: This study utilised model-based geostatistics estimates MCV2 coverage from the 2022 Kenya Demographic and Health Survey (DHS), monthly immunisation data from routine health information systems (2019–2022) imputed for missingness and population data from WorldPop for 2019 across 62 Western Kenyan subnational areas (sub-counties). Routine MCV2 coverage was computed using MCV2 doses as a numerator and two separate denominators: (i) Pentavalent 1 doses to account for children already receiving prior vaccines at health facilities (service-based coverage) and (ii) surviving infants to account for all eligible children (population-based coverage). Concordance was assessed using the 95% confidence intervals (CIs) of survey-modelled estimates, intra-class correlation coefficient (ICC), and Bland-Altman (BA) plots. Results: Survey-modelled estimates differed substantially in 55 (89%) and 39 (63%) sub-counties compared to population and service-based coverage estimates respectively. The different approaches showed poor congruence in survey-modelled versus population-based coverage estimates (ICC: 0.10, p = 0.229) and survey-modelled versus service-based coverage estimates (ICC: 0.42, p = <0.001); there was moderate congruence of population versus service-based coverage estimates (ICC: 0.65, p = <0.001). Survey-modelled vs. population-based coverage estimates showed the highest bias in BA plots of 18.80 percent points (p.p) compared to 11.02 p.p. and 7.79 p.p. between survey-modelled vs. service-based coverage and population vs. service-based coverage estimates, respectively. Conclusions: Substantial discrepancies among survey‐modelled, routine population, and service-based coverage estimates expose important variations in each approaches' results. While all approaches offer distinct insights, improving survey models, routine data quality and refining estimates of population catchment is imperative for reliable fine‐scale vaccine delivery monitoring.