Two-part Statistical Model for Identifying Baseline Predictors of Chronic Postsurgical Pain.

Area of Science:

• Biostatistics
• Pain Medicine
• Health Services Research

Background:

• Many patients report no postsurgical pain, creating zero-inflated data problematic for standard statistical models.
• Traditional models struggle with predicting postsurgical pain and identifying its risk factors.
• A two-part model, common in healthcare expenditure analysis, was hypothesized to outperform traditional methods for postsurgical pain prediction.

Purpose of the Study:

• To compare the predictive performance of a two-part model against traditional logistic and linear regression models for postsurgical pain.
• To identify patient and clinical risk factors associated with postsurgical pain using different modeling approaches.
• To determine if the two-part model offers superior identification of chronic postsurgical pain predictors.

Main Methods:

• A prospectively collected dataset of 3925 patients with chronic postsurgical pain was analyzed.
• A two-part model was compared with logistic and linear regression using a training/testing split and 400 repetitions.
• The two-part model estimated pain probability (3 months) and then severity (numeric rating scale) among affected patients.

Main Results:

• The two-part model showed superior predictive performance (higher R², lower RMSE and MAE) compared to linear regression.
• The two-part model identified 7 unique preoperative risk factors for postsurgical pain, including race, education, and anxiety.
• Patient sex, surgical type, and surgical site pain were significant factors across all models.

Conclusions:

• A two-part model provides a statistically superior approach for analyzing postsurgical pain data.
• This model enhances the identification of patient and clinical risk factors for chronic postsurgical pain.
• It effectively differentiates factors influencing pain onset from those affecting pain severity.