Background: Patients with K-RAS wild-type (WT) metastatic colorectal cancer (mCRC) are currently considered the optimal candidates for upfront treatment with combinations of chemotherapy and EGFR inhibitors. These combinations significantly extend overall survival (OS) compared to chemotherapy alone. However, a proportion of patients would not achieve this goal. This study has investigated the ability of an artificial intelligence-based analytical tool that, based on patients clinical parameters, can identify those on which the addition of EGFR inhibitors to initial chemotherapy is
likely to be effective.
Methods: Through www.projectdatasphere.org we accessed the dataset of the PRIME (NCT00364013) study. This phase III clinical trial compared panitumumab-FOLFOX4 (P/FOL) vs. FOLFOX4 (FOL) as first-line therapy for mCRC patients. This dataset was used to generate synthetic state representations (SSRs) for each patient through the integration of 36 clinical features (demographics, anthropometrics, medical history, blood tests and adverse events) collected, respectively, during the screening phase and the first month of inclusion in the trial. These SSRs were then inputted into a deep learning framework (DLF) to identify subgroup of patients based on their similarities. The resultant subpopulations were correlated with overall survival (OS).
Results: A total of 460 K-RAS WT patients were eligible for this exploratory analysis. DLF identified two different subpopulations of patients: SPA (n 162) and SPB (n 298). Feature contribution analysis showed major differences between both subpopulations. Indeed, SPA included significantly less patients with ECOG0 (51.5% vs. 61.7%; p 0.03), white race (89.5% vs. 94.6%; p 0.04) and that suffered previous surgery (84.5% vs. 93.9%, 2) (1.80 vs. 1.85; p 0.01) and BMI (24.9 vs. 26.4; < 0.001). In addition, they had lower levels of haemoglobin (g/dl) (12.0 vs. 12.4; < 0.001),
albumin (g/dl) (3.78 vs. 3.92; p < 0.001) and creatinine (mg/dl) (0.84 vs. 0.87; p 0.006) and higher levels of platelets (G/L) (342.2 vs. 275.9; p < 0.001), white blood cells (G/L) (8.38 vs. 7.16; p < 0.001), ALP (UI)(291.1 vs. 152.8; p < 0.001), LDH (UI) (883.7 vs. 390.5; < 0.001) and CEA (ng/dl) (1223.2 vs. 110.1; p < 0.001). Patients in SPA had a lower risk of death when treated with P/FOL (n 79) compared to FOL (n 83) (median OS: 23.6 months vs. 17.1 months; hazard ratio 0.68, 95% CI 0.48-0.99; p 0.04). Patients in SPB showed no significant differences between P/FOL (n 153) and FOL (n 145) (median OS: 27.1 vs. 23.6, respectively; hazard ratio 1.21, 95% CI 0.85-1.71; p 0.27). Of note, SPA patients treated with P/FOL achieved a similar OS than SPB patients treated with FOL.
Conclusions: Based on significantly singular clinical and laboratory features, our DLF system identifies two mCRC subpopulations with different survival outcome depending on whether they are treated with P/FOL or FOL. Specifically, SPA patients achieve significantly longer survival if P is added to FOL. In contrast, SPB patients would not receive any benefit from a P/FOL combination compared to FOL. Further work is required to validate this approach as a novel predictive biomarker tool for treatment decision making on K-Ras WT mCRC patients.
Citation: Ann Oncol 2021 32(Issue S3):S145
Please, request poster pdf at email@example.com