Predictive Modeling and Comparative Analysis of Palbociclib- and Abemaciclib-Associated Nephrotoxicity in Breast...

IntroductionCDK4/6 inhibitors (palbociclib and abemaciclib) are pivotal in hormone receptor-positive breast cancer, yet nephrotoxicity remains insufficiently studied. This study aimed to establish predictive models for nephrotoxicity and compare the renal safety profiles of the two agents.MethodsA retrospective cohort of 205 patients (145 abemaciclib, 60 palbociclib) was analyzed using six renal biomarkers, demographics, and treatment parameters. Machine learning models, including logistic regression and random forest, were trained to predict nephrotoxicity, defined as post-treatment acute kidney disease (AKD). Interaction effect analysis of age and single dosage on the predicted probability of nephrotoxicity was conducted.ResultsUsing the KDIGO AKD functional criteria supplemented by cystatin C, nephrotoxicity occurred in 56.6% (116/205) of the cohort. Using strictly pre-treatment variables, multivariable analysis identified a higher single drug dosage (150 mg vs. 100 mg; OR=3.99, p=0.004) and abnormal baseline serum creatinine (OR=0.04, p<0.001) as key independent predictors. Among six machine learning algorithms, the parametric logistic regression achieved the best discrimination (test area under the receiver operating characteristic curve [AUC]=0.704), outperforming complex models like random forest (AUC=0.641). Crucially, interaction analysis revealed a profound age-dependent dosage effect: younger patients exhibited a steep dose-dependent surge in nephrotoxicity risk, whereas older cohorts displayed a flattened, stable risk trajectory.ConclusionThis study demonstrates that a parsimonious machine learning approach using routine pre-treatment variables can effectively predict CDK4/6 inhibitor-associated nephrotoxicity, uncovering a critical age-dependent dosage effect. These findings underscore the necessity of individualized dose optimization and serve as an exploratory proof-of-concept that requires prospective, multi-center external validation before clinical application.