Photo Credit: PeopleImages

The following is a summary of “Predicting Treatment Outcomes in Patients with Low Back Pain Using Gene Signature-Based Machine Learning Models,” published in the December 2024 issue of Pain by Lian et al. 

Low back pain (LBP) is a global health challenge with uncertain molecular mechanisms and variable treatment outcomes, complicating response prediction.  

Researchers conducted a retrospective study to develop gene signature-based machine learning models using transcriptomic data from peripheral immune cells to predict treatment outcomes in individuals with LBP.  

They retrieved transcriptomic data of individuals with LBP from peripheral immune cells using the GEO database. Participants were recruited, and treatment outcomes were evaluated after 3 months, grouping them into resolved pain and persistent pain categories. Bioinformatic analysis identified differentially expressed genes (DEGs) between the groups, 5 machine learning models—Lasso, Elastic Net, Random Forest, Support Vector Machine (SVM), and Gradient Boosting Machine (GBM)—selected key genes. These genes trained 45 models combining 9 algorithms, including Logistic Regression, K-Nearest Neighbors, Decision Tree, Random Forest, SVM, GBM, Multilayer Perceptron, Naive Bayes, and Linear Discriminant Analysis, 5-fold cross-validation ensured robust evaluation, splitting data into training and validation sets. Model performance was measured by accuracy, precision, recall, and F1 score, with final performance reported as mean and standard deviation across all folds.  

The results showed 61 DEGs between individuals with resolved and persistent pain. From these, 45 machine learning models were developed using various feature selection methods and classification algorithms. Elastic Net with Logistic Regression achieved the highest accuracy at 88.7% ± 8.0% (mean ± standard deviation), followed by Elastic Net with Linear Discriminant Analysis at 88.7% ± 7.5%, and Lasso with Multilayer Perceptron at 87.7% ± 6.7%. A total of 15 models surpassed 80% accuracy, showing the robustness of the machine learning approach in predicting LBP treatment outcomes. The SHapley Additive exPlanations (SHAP) method highlighted the contributions of core genes to model performance, emphasizing the predictive roles.  

Investigators concluded the potential of transcriptomic data from peripheral immune cells and machine learning models in predicting treatment outcomes for LBP, offering a foundation for personalized treatment strategies. 

Source: link.springer.com/article/10.1007/s40122-024-00700-8