Developing Personalized Postoperative Follow‐Up Strategies for Patients With Locally Advanced Gastric Cancer Receiving Neoadjuvant Therapy Based on Dynamic Programming: A Multicenter Retrospective Cohort Study in China

Vollständiger Abstract

Worum geht es in dieser Arbeit?

ABSTRACT The optimal follow‐up strategy for patients with locally advanced gastric cancer (LAGC) receiving neoadjuvant therapy (NAT) remains unknown. Traditional follow‐up strategies based on relatively fixed intervals fail to fully account for dynamic changes in recurrence risk. This study aimed to develop a personalized postoperative follow‐up strategy using dynamic programming (DP) to optimize follow‐up arrangements based on individual patient characteristics and dynamic recurrence risks. This study included 3397 patients with LAGC who underwent surgery after NAT at 21 medical centers between 2018 and 2023. By integrating multiple prognostic indicators using a random survival forest model, we estimated individual time‐adjusted cumulative hazards. A conditional inference tree was then used to stratify patients into low‐, medium‐, and high‐risk groups. The DP algorithm was employed to determine the optimal follow‐up arrangements for recurrence detection. A Markov decision‐analytic model was used to identify the most cost‐effective follow‐up strategy. Compared with the guideline strategies, the DP‐based strategy significantly reduced the average delayed detection time, particularly in the high‐risk group. Furthermore, the cost‐effectiveness analysis showed that the DP‐based strategy achieved the best incremental cost‐effectiveness ratio. Finally, we determined that the optimal numbers of follow‐ups for the low‐, medium‐, and high‐risk groups were 9, 10, and 13, respectively. The study demonstrates that the DP‐based personalized follow‐up strategy significantly improved the efficiency of recurrence detection and resource utilization in LAGC. These findings highlight the potential of DP algorithms in clinical decision‐making and provide a foundation for future personalized follow‐up studies.