Estimating trial-record geography across ascertainment and counting specifications: an empirical methods comparison in metastatic colorectal cancer

Background Geographic analyses of clinical trials often rely on one data source and one rule for allocating multinational records across countries. The extent to which these methodological choices alter geographic estimates is uncertain. Methods We conducted a cross-source methodological study by linking metastatic colorectal cancer trial records from ClinicalTrials.gov and a July 2026 TrialTrove/Citeline snapshot to construct their deduplicated union. Geographic estimates used 186 IARC entities; the primary period was 2020–2025. A 3-by-2 design crossed two source-specific ascertainment pipelines and a deduplicated-union specification with fractional counting, which conserved one unit per record, and full counting, which assigned one unit to every documented country. The primary measure was the high-income-country observed-to-expected (O/E) ratio using all-stage colorectal cancer incidence as a burden proxy. We assessed cross-source country-field agreement and used total variation distance (TVD) to compare field, cohort-composition, complete-pipeline, and counting-estimand changes. We calculated percentile intervals from 5,000 paired component bootstrap resamples. Results The two source cohorts and the union contained 2,394, 864, and 2,841 eligible records or components across all years; 781, 259, and 891 began in 2020–2025. High-income-country O/E ranged from 1.03 to 1.82 across the six primary specifications. Among 140 matched components with country fields in both sources, 134 (95.7%) had identical normalised country sets and mean Jaccard similarity was 0.984. Changing only the country field produced a fractional TVD of 0.012, compared with composition TVDs of 0.187 and 0.131 and a complete-pipeline TVD of 0.240. Fractional-to-full TVDs ranged from 0.191 to 0.296. Conclusions Within the analysed specifications, TVD point estimates for cohort-composition and counting contrasts exceeded those for country-field substitution among matched records. Trial-geography studies should report the complete measurement chain and should not interpret documented countries as activated sites, enrolment, access, or database accuracy.