Sponsor exposure to Russia and Ukraine and the regional footprint of industry trials after 2022: an exposure difference-in-differences on ClinicalTrials.gov

Before February 2022, Russia and Ukraine hosted sites for a sizeable share of multinational industry trials. This paper asks if sponsors more reliant on the two countries changed their regional footprint differently after the invasion. Using all interventional studies starting in 2015-2025 on ClinicalTrials.gov (278,548 records; 53,977 industry trials after the sample rules), it compares, within sponsor, trials started in 2023-2024 with those started in 2017-2021, using each sponsor's pre-war share of trials with a Russian or Ukrainian site as a continuous exposure. The design and primary test follow an analysis plan written on 26 October 2026, not lodged with a registry; the order of that plan and access to the outcome data cannot be verified. A sponsor with full exposure lost 56 share points of trials with a Russian or Ukrainian site relative to a sponsor with none (95% cluster-bootstrap interval -82 to -31; 304 sponsors). The planned primary hypothesis, that exposed sponsors increased their share of trials with a Central and Eastern European site, is not supported: the coefficient is -1.1 share points (95% interval -18.9 to +24.4). Of six secondary regions, North America shows the only nominally significant shift (+13.2 points, interval +1.8 to +26.0), not surviving Holm correction (adjusted p = 0.15). The countries-per-trial estimate is imprecise (+0.77, interval -1.51 to +4.35) and compatible with both a fall and a rise. The overall share of industry trials with a Russian or Ukrainian site fell from 5.3% (2021 starts) to 1.3% (2024 starts). A post-hoc joint bootstrap puts the Central and Eastern European change at -2% of the Russian and Ukrainian loss (95% interval -30% to +47%). The estimates are exposure-associated differences; the outcomes record only regional presence, so the data give no precise evidence about whether, or where, lost sites were replaced.