Trends in HIV, hepatitis B, and hepatitis C eligibility language in immune checkpoint inhibitor oncology trials: A ClinicalTrials.gov rule-based NLP analysis (2010–2025).

Abstract

e23011 Background: Exclusion of people with HIV or viral hepatitis can limit generalizability of immune checkpoint inhibitor (ICI) trial evidence. Despite growing emphasis on risk-stratified inclusion, prevalence and temporal trends of HIV/HBV/HCV eligibility language remain incompletely characterized. We quantified these patterns in ICI oncology trials on ClinicalTrials.gov. Methods: Interventional ICI oncology trials first posted 2010–2025 on ClinicalTrials.gov were included (N = 8,924). Eligibility criteria were parsed with transparent, section-aware rule-based NLP at the criterion-line level and aggregated to trial-level classifications via prespecified precedence. For HIV/HBV/HCV, language was categorized as no mention, absolute exclusion, conditional eligibility, or mention–unclear. Primary HIV absolute exclusion was blanket exclusion for explicit HIV-positive status/known HIV infection/AIDS (strict); sensitivity included HIV-test–positive exclusions. HBV/HCV absolute exclusion was explicit chronic/known infection or serologic/virologic positivity without conditional qualifiers; conditional eligibility allowed controlled/suppressed/treated infection or limited exclusions to active/uncontrolled/detectable infection. Prevalence was summarized overall and by era (2010–2014, 2015–2019, 2020–2025). Temporal trends used adjusted logistic regression (binomial GLM, logit) estimating OR/year, adjusted for phase and industry sponsorship. Results: HIV/HBV/HCV language: no mention 2,345 (26.28%); conditional eligibility for ≥1 infection 5,352 (59.97%); absolute exclusion for ≥1 infection 990 (11.09%; sensitivity 12.19%, similar temporal pattern); mention–unclear 237 (2.66%). Infection-specific absolute exclusion was 9.72% (HIV), 1.48% (HBV), 0.30% (HCV). By era (2010–2014/2015–2019/2020–2025; n = 336/3,429/5,159), absolute exclusion for ≥1 infection was 10.12%/13.09%/9.83% and conditional eligibility 43.45%/64.30%/58.17%. Adjusted models showed decreasing HIV absolute exclusion (OR/year 0.956, 95% CI 0.935–0.977; p = 5.97×10⁻⁵) and absolute exclusion for ≥1 infection (OR/year 0.959, 95% CI 0.940–0.980; p = 9.25×10⁻⁵). HBV/HCV absolute exclusion was uncommon (132/27 events) with no significant temporal change. Stratified manual validation evaluated classification performance (n = 1,659). Conclusions: In ICI oncology trials posted 2010–2025, conditional eligibility for HIV/HBV/HCV was common, yet absolute exclusion persisted in ~1 in 9 trials (strict), peaking in 2015–2019 and declining in 2020–2025 largely due to decreasing HIV blanket exclusion. These findings support standardizing risk-stratified, evidence-based eligibility language to improve trial representativeness while maintaining safety.