Inferring Population Health from Selective Laboratory Testing Data

Laboratory data provide rich information about population health, yet suffer from a systematic bias: biomarkers measurements are observed only for individuals accessing diagnostic services, with testing shaped by repeated measurements, supply and demand side factors. Ignoring these mechanisms may lead to biased population-level inference. This study develops a statistical framework to address nonignorable missingness and selection bias in population-based laboratory data, linking clinical measurements about blood test with administrative records for all seven local health districts of the Modena province. We first model the propensity to undergo laboratory analysis as functions of observed individual socio-demographic characteristics and General Practitioner-linked attributes. We then analyse biomarker distributions accounting for the selection process. This working paper studies and compares several more traditional selection models along with newly developed modelling approaches. Expected results include quantification of coverage gaps across population subgroups, bias-corrected prevalence estimates, and identification of underserved populations with unmet diagnostic needs. This work provides a methodological template for developing evidence-based practices in Laboratory Information System design and for bias-aware analysis of Real World Data in integrated health data warehouse.