Negative prognostic factors and clinical improvement prediction modeling for extracorporeal shockwave therapy in...

The efficacy of extracorporeal shockwave therapy (ESWT) for treating shoulder calcific tendinitis can be influenced by various prognostic factors. This study aimed to identify prognostic factors associated with the failure of ESWT for symptom relief and to evaluate the predictive capability of the eXtreme Gradient Boosting (XGBoost) algorithm of artificial intelligence techniques in this context.

This retrospective study enrolled patients with persistent shoulder pain attributed to calcific tendinitis who underwent ESWT after failed conservative treatment between January 1998 and December 2022. Age, sex, duration of symptoms, calcification classification and size, pre-ESWT visual analog scale (VAS), and pre-ESWT Constant-Murley score (CMS) served as potential input attributes. The difference in VAS and CMS were defined as the output attributes. The XGBoost model was used to predict treatment outcomes based on these factors. The dataset was balanced using the synthetic minority oversampling technique, and the model's performance was assessed using 10-fold cross-validation. Spearman's rank correlation coefficient analysis was adopted to explore the relationships between significant continuous input attributes and post-ESWT VAS and CMS scores.

A total of 296 patients with calcific tendinitis were enrolled and completed the 1-year follow-up. The findings revealed that a prolonged symptom duration (>10 months), severe pain (pre-ESWT VAS >5), and higher pre-ESWT CMS (>55) were significant prognostic factors for the failure of ESWT for symptom relief. Using these factors as inputs, the XGBoost model demonstrated high accuracy, precision, recall, and F1 score. By reducing the input attributes to age, calcification size, pre-ESWT CMS, and symptom duration, the model maintained a high prediction rate, suggesting that these factors are sufficient for effective prediction.

The present study identified significant prognostic factors associated with the failure of ESWT in the treatment of shoulder calcific tendinitis. By using artificial intelligence techniques, particularly the XGBoost algorithm, we demonstrated an effective ability to predict the VAS and the CMS outcomes following ESWT. By employing the trained XGBoost model, clinicians can offer accurate predictions regarding the outcome of ESWT in this clinical scenario, aiding in treatment decision-making and optimizing patient care.