Machine Learning Model Identifies Early Adalimumab Discontinuation Risk in Rheumatoid Arthritis
Key Clinical Summary
- In a specialty pharmacy cohort, 37.7% of patients with rheumatoid arthritis (RA) were at high risk of discontinuing adalimumab within 6 months due to loss of efficacy.
- An Elastic Net machine learning (ML) model demonstrated strong predictive performance (AUC-ROC 0.886; F1 score 0.741) using routinely collected pharmacy and patient-reported data.
- Early identification of high-risk patients may enable targeted pharmacist interventions, reduce medication waste, and improve outcomes in specialty drug management.
Early discontinuation of biologic disease-modifying antirheumatic drugs (bDMARDs) remains a major challenge in RA, contributing to delayed disease control and increased health care costs. New research suggests that ML models using specialty pharmacy data may help identify patients at risk for early discontinuation of adalimumab, enabling more targeted intervention strategies.
Study Findings
In this retrospective study, investigators analyzed data from 300 patients with RA who initiated adalimumab through a specialty pharmacy between 2020 and 2023. All patients maintained adequate adherence prior to discontinuation, ensuring that early therapy cessation reflected lack or loss of efficacy rather than nonadherence.
Among the cohort, 37.7% were classified as high risk for discontinuing adalimumab within 6 months. Using 38 initial variables, researchers identified 19 predictive features with strong clinical relevance, including pain score, joint swelling, morning stiffness, RA duration, body mass index, infection history, comorbidities, and treatment initiation status.
Multiple ML models were evaluated, including Elastic Net, linear discriminant analysis, support vector machines, random forest, and extreme gradient boosting. While extreme gradient boosting performed best in training, Elastic Net demonstrated the strongest and most consistent performance on the test set, achieving an AUC-ROC of 0.886 and an F1 score of 0.741.
Clinical predictors of high risk were largely driven by symptom burden and comorbid conditions. Variables such as higher pain scores, joint swelling, infection history, depression, and fibromyalgia were positively associated with early discontinuation, while concomitant conventional synthetic DMARD use was associated with lower risk.
Clinical Implications
For managed care and specialty pharmacy stakeholders, these findings highlight the potential of ML-driven risk stratification to optimize biologic therapy management. Early discontinuation of high-cost biologics such as adalimumab can result in substantial financial waste and suboptimal patient outcomes.
Specialty pharmacies are uniquely positioned to operationalize these tools, as they routinely collect patient-reported outcomes and clinical data during onboarding and follow-up. Integrating predictive models into pharmacy workflows could enable proactive outreach, including enhanced counseling, symptom monitoring, and coordination with prescribers for timely therapy adjustments.
From a value-based care perspective, such models may support more efficient use of specialty medications by aligning interventions with patients most likely to benefit. Threshold customization also allows health systems to balance sensitivity and resource utilization, tailoring outreach efforts based on operational priorities.
Conclusion
ML models leveraging real-world specialty pharmacy data show promise in identifying patients with RA at risk of early biologic discontinuation. With further validation and integration into clinical workflows, these tools may enhance patient outcomes, reduce medication waste, and support value-based specialty pharmacy care.
Reference
Yoon AH, Gedeck P, Oelofsen M. Predicting early discontinuation of adalimumab in patients with rheumatoid arthritis using machine learning: A specialty pharmacy–based approach. JMCP. 2026;32(3). doi:10.18553/jmcp.2026.32.3.336
