AI’s impact on medical research: A bibliometric study of citation impact and disruption
Huan Wang, Liting He, Huanhuan Huang, Yaoyue Hu, Shuangjiang Zheng, Jingdong Ma, Hao WuBackground
Artificial intelligence (AI) is increasingly used in medical research, but its scientific role remains unclear. Although AI may improve research visibility and efficiency, it is uncertain whether it primarily supports disruptive breakthroughs or high-impact incremental advances.
Objective
This study aimed to characterize trends in AI adoption in medical research and to examine whether AI-related outputs differ from non-AI outputs in citation impact and scientific disruptiveness.
Methods
We conducted a large-scale bibliometric analysis of medical research outputs indexed in OpenAlex and the PubMed Knowledge Graph from 1980 to 2022. Outputs were classified as AI-related using predefined criteria based on AI subfield classification, AI-related terminology, or AI-related references. Citation impact was assessed using citation counts, and scientific disruptiveness was assessed using disruption-based metrics. Multivariable regression models were used to examine the associations of AI adoption with citation impact and scientific disruptiveness.
Results
Among 17,800,073 medical research outputs, 530,907 were classified as AI-related using at least one of three criteria: the OpenAlex subfield taxonomy, AI-related terms, or references to AI-related research. The proportion of AI-related outputs increased markedly after 2015 but remained relatively low, reaching 5.77% in 2022. AI-related papers received more 5-year citations than non-AI-related papers during both the machine-learning era (mean, 28.90 vs 22.97; P < .001) and the deep-learning era (mean, 37.80 vs 30.28; P < .001). In adjusted analyses, AI-related status was positively associated with log-transformed citation counts (β = 0.115; 95% CI, 0.111–0.118) but was associated with lower odds of ranking among the top 5% of papers according to the disruption index (OR = 0.728; 95% CI, 0.712–0.745). AI adoption was particularly prevalent in health informatics and other data-intensive medical subfields.
Conclusions
The use of AI in medical research was associated with greater citation impact but a lower probability of ranking among the most disruptive papers. These contrasting associations suggest that AI-related research has, to date, been more closely associated with influential incremental advances than with changes that disrupt established citation patterns. Because the disruption index reflects subsequent citation relationships rather than scientific creativity itself, these findings should not be interpreted as evidence that AI reduces originality or innovation. Instead, they highlight the importance of combining AI-enabled analytical capabilities with human conceptual reasoning and scientific judgment.