Real-world safety profile of mitotane in adrenocortical carcinoma: a retrospective cross-sectional study integrating pharmacovigilance and interpretable machine learning.
Mitotane remains the cornerstone of adrenocortical carcinoma. However, its real-world safety profile is poorly defined. This study systemically evaluated mitotane-associated adverse events (AEs) and characterized multivariable patterns of serious outcomes using a combined pharmacovigilance and machine learning approach.
Mitotane-related AE reports (2004-2025) were extracted from the FDA Adverse Event Reporting System (FAERS). Disproportionality analyses were performed via four algorithms. Detected signals were categorized using a clinical priority scoring system based on signal strength, mortality, seriousness, and existing evidence. Time-to-onset was evaluated utilizing Kaplan-Meier methods. Elastic net logistic regression and Extreme Gradient Boosting (XGBoost) models characterized features associated with serious outcomes, with performance assessed by the area under the receiver operating characteristic curve (AUROC).
Among 870 identified reports, 45.7% involved serious outcomes. We detected 75 safety signals, predominantly involving gastrointestinal, neurological, endocrine, and metabolic systems. Notably, 21 signals (28.0%) were assigned moderate clinical priority, with adrenal insufficiency as the highest-priority signal. Time-to-onset analysis showed that endocrine AEs had a significantly longer onset time than non-endocrine AEs. Both models demonstrated consistent characterization of multivariable patterns associated with serious outcomes (elastic net AUROC = 0.830, XGBoost AUROC = 0.869). Key contributors to serious outcomes included reporting country, medical event designation, organ system involvement, and medication complexity.
This large-scale study provides a real-world safety assessment of mitotane. Integrating signal detection, clinical prioritization, and interpretable machine learning reveals multivariable patterns underlying serious AEs. These findings support a risk-stratified monitoring approach to enhance the clinical safety of mitotane therapy.
Mitotane-related AE reports (2004-2025) were extracted from the FDA Adverse Event Reporting System (FAERS). Disproportionality analyses were performed via four algorithms. Detected signals were categorized using a clinical priority scoring system based on signal strength, mortality, seriousness, and existing evidence. Time-to-onset was evaluated utilizing Kaplan-Meier methods. Elastic net logistic regression and Extreme Gradient Boosting (XGBoost) models characterized features associated with serious outcomes, with performance assessed by the area under the receiver operating characteristic curve (AUROC).
Among 870 identified reports, 45.7% involved serious outcomes. We detected 75 safety signals, predominantly involving gastrointestinal, neurological, endocrine, and metabolic systems. Notably, 21 signals (28.0%) were assigned moderate clinical priority, with adrenal insufficiency as the highest-priority signal. Time-to-onset analysis showed that endocrine AEs had a significantly longer onset time than non-endocrine AEs. Both models demonstrated consistent characterization of multivariable patterns associated with serious outcomes (elastic net AUROC = 0.830, XGBoost AUROC = 0.869). Key contributors to serious outcomes included reporting country, medical event designation, organ system involvement, and medication complexity.
This large-scale study provides a real-world safety assessment of mitotane. Integrating signal detection, clinical prioritization, and interpretable machine learning reveals multivariable patterns underlying serious AEs. These findings support a risk-stratified monitoring approach to enhance the clinical safety of mitotane therapy.