Machine learning personalizes cannabis treatment for chronic pain

The results of the study indicate that certain clinical and genetic factors significantly influence therapy dropout. The Visual Analog Scale (VAS) score at the final follow-up emerged as the strongest predictor. Patients who reported persistently high pain levels were more likely to discontinue cannabis therapy, suggesting that inadequate pain relief remains a primary reason for dropout.

Chronic pain is a pervasive condition affecting nearly 30% of the global population, significantly impacting the quality of life and posing challenges for long-term treatment adherence. Traditional pain relief methods, including opioids and NSAIDs, often fail to provide sustained benefits and come with risks such as addiction and severe side effects. Given these limitations, medical cannabis has emerged as a promising alternative. However, treatment discontinuation remains a major issue, with a substantial number of patients opting out due to varying responses and side effects.

Addressing this concern, the research paper "Predicting Therapy Dropout in Chronic Pain Management: A Machine Learning Approach to Cannabis Treatment" by Anna Visibelli, Rebecca Finetti, Bianca Roncaglia, Paolo Poli, Ottavia Spiga, and Annalisa Santucci, published in Frontiers in Artificial Intelligence, explores the potential of machine learning (ML) to predict therapy dropout and improve treatment personalization.

Understanding the study: Data and methods

The study gathered a comprehensive dataset of 542 Caucasian patients undergoing cannabis-based therapy for chronic pain. These patients, recruited from the POLIPAIN CLINIC, were monitored for treatment adherence across multiple follow-ups. The dataset integrated genetic markers, clinical profiles, and pharmacological parameters, providing a robust foundation for analysis.

A machine learning model was developed to identify key predictors of therapy discontinuation. The authors used a random forest classifier, an ensemble learning method known for its high accuracy in handling complex datasets. The model underwent rigorous validation, achieving an impressive mean accuracy of 80%, with a peak of 86% and an Area Under the Curve (AUC) score of 0.86. To further refine the findings, the study employed SHapley Additive exPlanations (SHAP), a method that highlights the contribution of individual factors in the model’s decision-making process.

Key Findings: Why Do Patients Drop Out?

The results of the study indicate that certain clinical and genetic factors significantly influence therapy dropout. The Visual Analog Scale (VAS) score at the final follow-up emerged as the strongest predictor. Patients who reported persistently high pain levels were more likely to discontinue cannabis therapy, suggesting that inadequate pain relief remains a primary reason for dropout.

Another crucial factor was THC dosage. Higher THC intake correlated with increased dropout rates, likely due to adverse side effects such as dizziness, anxiety, or cognitive impairment. In contrast, CBD dosage appeared to have a stabilizing effect, reducing the likelihood of discontinuation, possibly by mitigating THC-induced discomfort.

Additionally, the study uncovered a genetic link to treatment adherence. The rs1049353 polymorphism in the CNR1 gene, which encodes the cannabinoid receptor CB1, was found to influence dropout rates. Patients with the CC genotype exhibited higher retention rates, whereas those with the CT genotype were more prone to discontinuation, possibly due to altered cannabinoid receptor sensitivity.

Implications for personalized pain management

The study underscores the potential of ML and pharmacogenetics in tailoring cannabis-based treatments. By incorporating genetic screening and predictive modeling, healthcare providers can design more effective treatment plans, optimizing dosage and minimizing side effects. For instance, patients identified as high-risk for dropout due to genetic markers or prior treatment responses could receive adjusted THC-CBD ratios or alternative therapeutic interventions.

Moreover, the findings highlight the need for continuous monitoring and adaptive treatment strategies. Patients who do not experience significant pain relief within the first few weeks of therapy could benefit from early intervention, such as dose modifications or complementary pain management approaches.

This research paves the way for AI-driven precision medicine in chronic pain management, offering a data-backed approach to enhancing patient adherence and therapeutic efficacy. Future studies may explore broader population groups and incorporate additional biomarkers to refine predictive accuracy further.