Comparing sampling approaches for assessing sub-national vaccination coverage: a pilot study in Central African Republic and the Democratic Republic of Congo

Vaccination coverage surveys (VCS) are essential for monitoring program performance, especially where administrative data are unreliable. While national and sub-national estimates are common, district-level data remain limited despite their value for local planning and equity monitoring. This study compared three alternative sampling approaches (ASAs)-Kinshasa School of Public Health VCS (KSPH-VCS), Geographic Information System (GIS)-based sampling, and Lot Quality Assurance Sampling (LQAS)-to the WHO Vaccination Coverage Cluster Survey (WHO-VCCS) in terms of validity and precision among children aged 12-23 months at the health district (HD) level. A cross-sectional study was conducted in five HDs (urban and rural) in the Central African Republic (CAR) and the Democratic Republic of Congo (DRC). All four approaches were applied simultaneously. For 17 vaccination indicators per HD, comparability to WHO-VCCS was assessed using four metrics: (i) absolute difference in percentage points (pp) between each ASA and the WHO-VCCS estimates, (ii) inclusion within WHO's 95% confidence intervals (CIs), (iii) CI width ratio, and (iv) proportion of estimates outside WHO CIs. Survey-adjusted logistic regressions, design effects (DEFF), and intra-cluster correlation coefficients (ICC) were analyzed. Aggregated results at regional/national levels were also simulated. GIS showed the highest concordance (49.3% of estimates within 5 pp.; 41.2% within WHO CIs) but had the widest CIs (median width ratio relative to WHO-VCCS: 2.2). KSPH-VCS had narrower CIs (1.7) and moderate validity (22.4% within 5 pp), while LQAS had the highest proportion of estimates outside WHO CIs (median: 18.5 pp). KSPH-VCS and GIS were statistically comparable to WHO-VCCS in 3 of 5 HDs; LQAS in 4. At aggregated levels, all ASAs performed similarly. KSPH-VCS and GIS offer viable options to WHO-VCCS, especially for producing aggregated estimates with acceptable precision. Further studies should assess cost, feasibility, and adaptability across settings.