Analyzing collaborations in clinical trials in Korea using association rule mining.

Abstract

Identifying how trial sites collaborate is essential for multicenter trials. The ways in which collaboration among trial sites is established can vary according to study phase and clinical trial domains. In this study, we employed association rule mining to reveal trial collaboration. We used trial approval data provided by the Ministry of Food and Drug Safety in Korea and organized the trial sites. We collected trial information from 2012 to 2023 and categorized the trials according to study phase and clinical trial domain. We performed association rule mining based on study phase and clinical trial domain. We identified 209 valid trial sites and analyzed 11,107 clinical trials conducted during this period. By study phase, phase 1 trials accounted for the largest number (5,451), followed by phase 3 (2,492), others (1,826), and phase 2 (1,338). We found that phase 1 clinical trials had the highest lift metrics. The mean lift for phase 1 trials was 5.40, which was significantly greater than that of phase 2 (1.68) and phase 3 trials (1.72). Additionally, the network structure for trial collaboration in phase 1 trials was highly condensed, with several trial sites located in Seoul and Gyeonggi-do. Different trial collaboration characteristics were noted among clinical trial domains, with mean and variability of the lift metrics for pediatrics being the highest. In conclusion, association rule mining can identify collaborations among trial sites. Collaboration in phase 1 trials is relatively more exclusive than in other phases, and aspects of collaboration differ among clinical trial domains.