A risk prediction model for severe, life-threatening, and fatal treatment-related toxicity.

Abstract

12143 Background: Adverse events are major contributors to morbidity and mortality in patients receiving systemic cancer therapy. Accurate prediction of high-grade toxicity could inform risk-adapted treatment strategies. Methods: This cohort study included data from 18 SWOG adult phase 2 and 3 treatment trials conducted from 1990 to 2022. Baseline patient-reported symptoms (fatigue, pain, and anxiety, classified as clinically meaningful yes vs. no), baseline sociodemographic (age, sex, race, ethnicity), and performance status (PS) variables were used. The aim was to develop a unified risk score applicable across multiple toxicity severity thresholds; adverse events were classified using the Common Terminology Criteria for Adverse Events. The primary outcomes were severe (Grade >3), life-threatening (Grade >4), and fatal toxic events. A training set was derived using a random 60% sample, balanced by study and arm; validation will occur in the remaining 40%. Best-subsets logistic regression was used for variable selection; final models were stratified by cancer type (hematologic, hormone-driven solid tumor, or other solid tumor) and treatment status (untreated vs refractory). Model discrimination was assessed using the concordance (c) statistic. Patients were divided into tertiles according to their total risk score. Toxicity rates and odds ratios were evaluated across tertiles (Ts). Results: Of 7,936 patients, 4,776 (60.2%; median age 64.9 (range, 18-95.3); female 28.0%; Black 10.8%) formed the training cohort. Increased toxicity was associated with age ≥60 years, fatigue, female sex, and PS ≥1. The multivariable risk models demonstrated good discrimination for grade ≥3 (c=0.76), grade ≥4 (c=0.77), and grade 5 toxicity (c=0.77). Severe toxicity increased from 15.8% in the lowest risk tertile (T1) to 64.6% in T3; life-threatening toxicity increased from 2.7% in T1 to 26.2% in T3 (Table). Compared with the lowest tertile, the highest risk tertile had markedly higher odds of grade ≥3 (OR 9.76), grade ≥4 (OR 12.96), and grade 5 toxicity (OR 7.15; all p<.001). Model validation results will be presented. Conclusions: This risk model accurately identified patients with substantially elevated risk for severe, life-threatening, and fatal toxicities. Risk stratification may support individualized treatment counseling, risk-guided monitoring, and proactive toxicity management. Observed toxicity rates and odds ratios by predicted-risk tertile. Observed Toxicity Odds Ratios by Model(95% CI, p-value) T1(n=1536) T2(n=1646) T3(n=1594) Increase per risk level T2 vs. T1 T3 vs. T1 Grade >3 15.8 37.7 64.6 3.11(2.86-3.39, p<.0001) 3.23(2.73-3.83, p<.0001) 9.76(8.23-11.59, p<.0001) Grade >4 2.7 9.1 26.2 3.58(3.12-4.11, p<.0001) 3.66(2.57-5.20, p<.0001) 12.96(9.32-18.02, p<.0001) Grade >5 0.2 0.6 1.4 2.54(1.54-4.18, p<.0001) 3.12(0.86-11.37, p=.08) 7.15(2.14-23.94, p=.001)