AI-driven chemotoxicity prediction in colorectal cancer: impact of race, SDOH, and biological aging.

Background: Patients with colorectal cancer (CRC) often experience chemotoxicity that impacts treatment adherence, survival, and quality of life. Early screening for chemotoxicity risk is vital, yet comprehensive predictive models are lacking. The objective of this study was to develop effective artificial intelligence (AI)/machine learning (ML) models, integrating racialized group, social determinants of health (SDOH) (Area Deprivation Index [ADI], employment status), and biological aging (Levine Phenotypic Age) to predict overall, gastrointestinal (GI), and hematological chemotoxicity.

Methods: We used electronic health records data from 1,735 adult patients with CRC. Sociodemographic/clinical variables, Levine Phenotypic Age (biological aging), and SDOH (including geospatial variations measured by ADI) were analyzed using descriptive statistics. Associations with chemotoxicity (overall, GI, hematological) were evaluated via univariate tests. Significant predictors from univariate tests were selected for AI/ML modeling. Six supervised ML models were trained on 80% of cases (n = 1,388), with 20% (n = 347) reserved for testing. Performance was assessed via accuracy, area under the curve (AUC), and F1-score. Permutation feature importance ranked predictors to define the most significant predictors of chemotoxicity.

Results: Support Vector Machine and XGBoost models demonstrated high accuracy in both the training and test datasets. Notably, the AUC (0.988) was highest for the Support Vector Machine model in predicting overall chemotoxicity within the training dataset. Key predictors of overall and GI toxicities included higher Levine Phenotypic Age, elevated inflammatory markers (e.g., C-reactive protein), and poor SDOH (e.g., higher ADI, unemployment). Hematological toxicity was linked to lower inflammatory markers, higher Levine Phenotypic Age, and younger chronological age. Race (non-Hispanic Black), body mass index, and lifestyle also influenced overall and GI toxicities.

Conclusions: ML-based chemotoxicity prediction models incorporating racialized group, SDOH, and biological aging had high accuracy. Greater biological aging, poor SDOH including ADI, and higher inflammation markers were common risk factors for overall and GI chemotoxicity. In contrast, chronological and biological ages and immune/inflammation markers were only linked to hematological chemotoxicity. Integrating these factors into predictive models can help clinicians identify at-risk patients and tailor interventions (e.g., anti-inflammatory, anti-aging strategies) to reduce chemotoxicity and improve survivorship outcomes.