Utilizing machine learning to identify multimodal signatures for patients who would benefit from the addition of tremelimumab to durvalumab and chemotherapy (TRIDENT)

Abstract Purpose: POSEIDON (NCT03164616) was a randomized, open-label, multicenter phase 3 trial comparing first-line durvalumab with or without tremelimumab in combination with chemotherapy versus chemotherapy alone in patients with metastatic non-small-cell lung cancer (NSCLC). Overall survival (OS) and progression-free survival were significantly increased in the tremelimumab plus durvalumab and chemotherapy arm. We conducted a post hoc analysis (TRIDENT) to identify patients who may receive greater OS benefit from the addition of tremelimumab to durvalumab and chemotherapy. Experimental design: This analysis included clinical, genomic, and radiomic data from the POSEIDON trial (data cut-off March 12, 2021). Machine learning models leveraging multimodal data were trained to identify subpopulations of patients that benefit from the addition of tremelimumab to first-line durvalumab and chemotherapy. Results: Using clinical and genomic data, the model was able to predict treatment benefit from adding tremelimumab to first-line durvalumab and chemotherapy, with the top ranked 50% of patients with non-squamous tumors achieving a HR of 0.56 (95% CI: 0.33–0.97). EGFR wild-type, FGFR3 wild-type, CDKN2A wild-type, KRAS mutations, and STK11 mutations were the factors most associated with higher OS benefit. Conclusions: By utilizing machine learning models to analyze POSEIDON data, we yielded genetic signatures identifying patients with non-squamous metastatic NSCLC who may derive greater OS benefit from the addition of tremelimumab to first-line durvalumab and chemotherapy. Such approaches could be used in future to enhance precision in tailoring therapies for individual patients.