Machine Learning Flags Who Will Still Hurt a Year After Rotator Cuff Surgery

Not everyone who has a torn rotator cuff repaired walks away pain-free. In a new study published in BMC Musculoskeletal Disorders, researchers report that machine learning models trained on preoperative psychological and psychosocial measurements can classify which patients are likely to report clinically significant shoulder pain a full year after surgery. The work, a retrospective single-center cohort study, is being presented by its authors as hypothesis-generating: a promising first step toward risk stratification that still requires multicenter external validation before any clinical use.

The study enrolled 491 patients with rotator cuff tears who had undergone surgical intervention. Persistent shoulder pain was defined as a Visual Analog Scale score of 4 or higher at the one-year follow-up. Patients meeting that cutoff were placed in a residual pain group, while the remainder were classified as a pain-free group. That binary outcome became the distinction the models were trained to make.

The investigators used a prespecified LASSO-based feature-selection strategy to identify candidate clinical variables, with additional stability analyses using forward selection and exhaustive search reported in the supplementary materials. Variables that were consistently selected across these procedures were considered candidate risk factors for persistent postoperative shoulder pain.

Nine commonly used machine learning models were trained to distinguish between patients with residual pain and those without. Performance was assessed with the area under the curve, specificity, sensitivity, and accuracy. According to the results, the models demonstrated effectiveness in classifying postoperative pain, and predictive performance was notably enhanced when the team turned to ensemble learning.

The ensemble approach combined the three models with the highest predictive performance into a single ensemble of the nine models. Among the tested configurations, the Support Vector Machine showed the highest accuracy when serving as the aggregator, the component that decides how the underlying models’ outputs are combined into a final classification. The authors report that the ensemble learning technique notably enhanced the classification of patients with residual pain.

Perhaps the most clinically interesting part of the study is which variables the ensemble models flagged as critical predictive factors. Rather than anatomical or surgical details, the consistently selected candidates were psychological and psychosocial measures: the Fear of Pain Questionnaire, the Pain Catastrophizing Scale, the Pittsburgh Sleep Quality Index subscale for perceived sleep quality, the Central Sensitization Inventory, and the Pain Surface Body Chart score. These instruments capture how patients perceive, fear, and catastrophize about pain, how they sleep, and how widespread their symptoms feel, all measured before the operation took place.

What it does offer is a concrete, data-driven shortlist of preoperative measurements that a future prospective study could test as part of a risk-stratification tool for patients contemplating rotator cuff repair.

The authors are explicit about the limits of their work. In the study’s conclusion, they describe the stacking ensemble of preoperative psychosocial variables as showing promise for preliminary risk stratification of clinically significant shoulder pain at one year after rotator cuff repair, and they state that these hypothesis-generating results require multicenter external validation before clinical use. The study is classified as Level 4 evidence in the Diagnosis category, and the authors note it was a retrospective observational study that was not registered as a clinical trial.

For now, the practical takeaway is cautious. Until that external validation arrives, the ensemble remains a research tool rather than a bedside decision aid, and the identified psychological risk factors remain candidates rather than confirmed predictors.

Subject of Research: Machine learning prediction of persistent shoulder pain one year after rotator cuff repair using preoperative psychological measurements

Article Title: Prediction of postoperative persistent pain after rotator cuff repair based on psychological measurements via machine learning

Article References: Hu, J., Guo, X., & Cheng, C. (2026). Prediction of postoperative persistent pain after rotator cuff repair based on psychological measurements via machine learning. BMC Musculoskeletal Disorders. https://doi.org/10.1186/s12891-026-10651-0

Image Credits: AI Generated

DOI: 10.1186/s12891-026-10651-0

Keywords: rotator cuff repair, persistent shoulder pain, machine learning, ensemble learning, pain catastrophizing, fear of pain, central sensitization, sleep quality, risk stratification, retrospective cohort, psychosocial factors, support vector machine