Answers are grounded in this paper and cite their sources.
Characterizing study sample representativeness is critical for the validity of biobank-derived findings, yet selection biases are rarely quantified across the full clinical spectrum. Here we systematically evaluate the Estonian Biobank (EstBB) - comprising ∼20% of the adult population - by comparing two recruitment waves against a 30% national reference dataset (Est-Health-30). Analyzing prevalence ratios (PR) across 1,028 ICD-10 categories, we reveal a bifurcated landscape of representativeness. While EstBB achieves population parity (PR ∼ 1.0) for widespread chronic conditions like Type 2 diabetes, 47% of diagnostic categories exhibit significant deviations (>1.3-fold). We identify a distinct “managed symptomatic” phenotype: a systematic enrichment of outpatient diagnoses - such as melanocytic nevi (PR=2.07, 95% CI 1.93…2.21) and major depression (PR=1.53, 1.4…1.66) - coupled with a depletion of high-mortality conditions like lung cancer (PR=0.69, 0.64…0.75) and vascular dementia (PR=0.45, 0.38…0.54). These biases evolved across recruitment phases, with the later EstBB2 cohort exhibiting a healthier, prevention-oriented profile. To support research integrity, we provide an interactive open-access dashboard for phenotype refinement. Accounting for such selection-driven “clinical visibility” is essential to avoid collider bias in risk prediction and causal inference models.
No takes yet. Share an insight, caveat, or question.
Pajusalu et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: