Validation of the CARG tool in prediction of chemotherapy completion and side effects in older adults with cancer: A retrospective cohort study

INTRODUCTION: Predicting treatment-related toxicity in older adults remains challenging. The Cancer and Aging Research Group (CARG) toxicity tool is one of the most established toxicity prediction tools; however, validation studies have shown variable predictive ability. Our primary objective was to validate the CARG toxicity tool in a cohort of Canadian older adults on systemic therapy. Secondary objectives were to compare the tool's performance with Eastern Cooperative Oncology Group Performance Status (ECOG-PS), Geriatric-8 (G8), and Vulnerable Elders Survey (VES-13) scales and examine whether these tools could predict treatment discontinuation, emergency room (ER) visits, and hospitalizations. MATERIALS AND METHODS: This retrospective analysis included patients aged ≥65 who completed a Comprehensive Geriatric Assessment (CGA) and received systemic therapy at an academic Canadian cancer centre. The area under the receiver operating characteristic curve (AUC) was calculated for the CARG and compared to ECOG-PS, G8, and VES-13. Logistic regression was performed to examine the association of the CARG toxicity tool with Grade ≥ 3 toxicity, ER visits, and hospitalizations. RESULTS: One hundred sixty-eight patients were included, 87 (51.8%) female, mean age 79.1 (SD 5.9). At least one Grade ≥ 3 toxicity was observed in 118 patients (70.2%). In toxicity prediction, the CARG toxicity tool had modest performance (AUC 0.64, 95% CI 0.55-0.73) but outperformed the ECOG-PS (AUC 0.57), VES-13 (AUC 0.55), and G8 (AUC 0.54), though this was not statistically significant (p = 0.17). On multivariable logistic regression, the high-risk category on the CARG toxicity tool was significantly associated with Grade ≥ 3 toxicity (OR 2.93 (95% CI 1.10-7.79, p = 0.03). High-risk CARG toxicity scores were correlated with treatment incompletion (OR 2.12, CI 1.04-4.30, p = 0.04) but not with hospitalization or ER visits. ECOG-PS ≥ 2 (OR 1.99, 95% CI 1.04-3.80, p = 0.04) and abnormal G8 scores (OR 2.71, 95% CI 1.13-6.51, p = 0.03) showed stronger association with these outcomes, respectively. The CARG tool performed similarly when restricted to chemotherapy regimens alone and for solid vs hematological malignancies. DISCUSSION: The CARG toxicity tool demonstrated modest performance in toxicity prediction in a Canadian older adult cohort but appears superior to other tools. It was also associated with treatment incompletion although the ECOG-PS and G8 showed stronger association with hospitalization and ER visits.