2010-2026年ClinicalTrials.gov注册的AI临床试验持续与全球疾病负担错配
该横断面元研究将2010-2026年ClinicalTrials.gov注册的2771项AI临床试验映射至全球疾病负担三级病因,发现试验数量与DALYs的log-log回归斜率为0.31,远低于比例分配的预期斜率1.0;160种病因中56种(35.0%)无AI试验,占全球DALYs的14.4%;2023年7月全球AI健康公平倡议启动前后错配模式无实质性重新定向。
要点
- AI临床试验数量随疾病负担增加呈亚比例增长,DALYs斜率仅0.31。
- 56种病因无AI试验,占全球DALYs的14.4%,腹泻病单独56.5百万DALYs零试验。
- 2023年全球AI健康公平倡议启动后未观察到研究优先级向高负担被忽视疾病重新定向。
核心结论
- AI临床试验活动与全球疾病负担仍显著错配,未观察到向高负担被忽视疾病的重新定向。
研究设计与样本
- 横断面元研究,将2010-2026年ClinicalTrials.gov注册的2771项AI临床试验映射至全球疾病负担三级病因。
主要发现
- 试验数量对全球DALYs的log-log回归斜率为0.31,10倍疾病负担增加仅对应2.0倍AI试验活动增加,比例分配下预期斜率为1.0。
- 死亡(斜率0.56)和患病率(斜率0.14)也呈亚比例模式。
- 160种全球疾病负担三级病因中56种(35.0%)在整个研究期间无AI试验,合计占全球DALYs的14.4%;腹泻病单独承载56.5百万DALYs且零试验。
- 至少有一项试验的病因中,错配比最高包括斑秃(19.06,n=2项试验)、炎症性肠病(12.94)和非黑色素瘤皮肤癌(12.41);最 underrepresented 为新生儿疾病(0.05)、疟疾(0.02)和结核病(0.10)。
- 错配模式在不同负担指标间广泛一致(DALY vs 死亡 Spearman r=0.77;DALY vs 患病率 r=0.60)。
关键数值与推荐
- 2023年7月5日前后比较错配比:75种合格病因中32种(42.7%)错配恶化,43种(57.3%)改善;最大变化反映已重度研究疾病的极端过度代表性降低,而非高负担被忽视疾病的获益。
临床怎么用
- 资助者和研究者可能需要采取 deliberate 措施,将AI研究导向高负担、研究不足的疾病。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVE: To determine whether artificial intelligence (AI) clinical trial activity aligns with global disease burden, and whether the 2023 launch of a global AI health equity initiative was associated with a reorientation of research priorities. STUDY DESIGN AND SETTING: We conducted a cross-sectional metaresearch study mapping 2771 AI clinical trials registered on ClinicalTrials.gov between 2010 and 2026 to Global Burden of Disease level 3 disease causes, quantifying mismatch between trial allocation and burden across three metrics: disability-adjusted life years (DALYs), deaths, and prevalence. RESULTS: Log-log regression of trial count on global DALYs returned a slope of 0.31, indicating that a 10-fold increase in disease burden was associated with only a 2.0-fold increase in AI trial activity against an expected slope of 1.0 under proportional allocation. Slopes for deaths (0.56) and prevalence (0.14) showed similar subproportional patterns. Fifty-six of 160 Global Burden of Disease level 3 causes (35.0%) had no AI trials identified by our title-based search criteria across the entire study period, collectively representing 14.4% of global DALYs; diarrheal diseases alone carried 56.5 million DALYs with zero trials. Among causes with at least one trial, inflammatory bowel disease (mismatch ratio 12.94), nonmelanoma skin cancer (12.41), and alopecia areata (19.06, n = 2 trials) were among the most overrepresented relative to their burden. Neonatal disorders (0.05), malaria (0.02), and tuberculosis (0.10) were the most underrepresented. Mismatch patterns were broadly consistent across burden metrics (DALY vs deaths Spearman r = 0.77; DALY vs prevalence r = 0.60). To assess whether the July 2023 launch of the Global Initiative on AI for Health (GI-AI4H) by WHO, ITU, and WIPO was associated with a reorientation of research priorities, we compared mismatch ratios before and after July 5, 2023. Among 75 eligible causes, 32 (42.7%) experienced worsened mismatch and 43 (57.3%) improved; however, the largest shifts reflected reductions in extreme overrepresentation among already heavily researched conditions rather than gains for high-burden neglected ones. CONCLUSION: AI clinical trial activity remains substantially misaligned with global disease burden. No observable reorientation toward high-burden neglected conditions was detected in registration data following the establishment of a formal international equity mandate. PLAIN LANGUAGE SUMMARY: Artificial intelligence is increasingly used in medical research, but it is unclear whether AI clinical trials focus on the diseases that cause the most harm worldwide. We looked at every AI-related clinical trial registered on the website ClinicalTrials.gov between 2010 and 2026 and compared how many trials existed for each disease against how much illness, death, and disability that disease causes globally. We found that AI trials often focus on diseases that are already well-studied, such as certain skin conditions and some cancers, whereas diseases that cause enormous suffering worldwide, such as diarrheal diseases, malaria, and newborn health problems, had few or no identified AI trials at all. In fact, 56 diseases that together account for more than one in every 7 years of healthy life lost worldwide had no identified AI trials studying them. Diseases that cause more global harm did tend to attract more AI trials overall, but not nearly in proportion: a tenfold increase in a disease's burden was linked to only about a twofold increase in AI trial activity. In 2023, the World Health Organization and other international bodies launched an initiative meant to direct AI health research toward the diseases most in need. We compared AI trial activity before and after this initiative launched and found no clear shift toward the neglected diseases it aimed to help. These findings suggest that decisions about which diseases to study with AI are being shaped more by where research is already easy or already funded than by where AI could help the most people. Funders and researchers may need to take deliberate steps to direct AI research toward high-burden, understudied diseases if this pattern is to change.
引用
- Kirby K, Forbes C, Livsey T, et al. AI clinical trials registered on ClinicalTrials.gov showed persistent misalignment with global disease burden, 2010-2026. J Clin Epidemiol. 2026. PMID: 42805511. doi:10.1016/j.jclinepi.2026.112543
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。