OBJECTIVE

The objective is to construct a random forest model for predicting the occurrence of Myofascial pelvic pain syndrome (MPPS) and compare its performance with a logistic regression model to demonstrate the superiority of the random forest model.

METHODS

We retrospectively analyze the clinical data of female patients who underwent pelvic floor screening due to chronic pelvic pain at the Pelvic Floor Rehabilitation Center of the Third Affiliated Hospital of Zhengzhou University from January 2021 to December 2023. A total of 543 female patients meeting the study's inclusion and exclusion criteria are randomly selected from this dataset and allocated to the MPPS group. Furthermore, 702 healthy female patients who underwent pelvic floor screening during routine physical examinations within the same timeframe are randomly selected and assigned to the non-MPPS group. Chi-square test and rank-sum test are used to select demographic variables, pelvic floor pressure assessment data variables, and modified Oxford muscle strength grading data for logistic univariate analysis. The selected variables are further subjected to multivariate logistic regression analysis, and a random forest model is also established. The predictive performance of the two models is evaluated by comparing their accuracy, sensitivity, specificity, precision, receiver operating characteristic (ROC) curve, and area under the curve (AUC) area.

RESULTS

Based on a dataset of 1245 cases, we implement the random forest algorithm for the first time in the screening of MPPS. In this investigation, the Logistic regression model forecasts the accuracy, sensitivity, specificity, and precision of MPPS at 69.96 %, 57.46 %, 79.63 %, and 68.57 % respectively, with an AUC of the ROC curve at 0.755. Conversely, the random forest prediction model exhibits accuracy, sensitivity, specificity, and precision rates of 87.11 %, 90.66 %, 90.91 %, and 83.51 % respectively, with an AUC of the ROC curve at 0.942. The random forest model showcases exceptional predictive performance during the initial screening of MPPS.

CONCLUSION

The random forest model has exhibited exceptional predictive performance in the initial screening evaluation of MPPS disease. The development of this predictive framework holds significant importance in refining the precision of MPPS prediction within clinical environments and elevating treatment outcomes. This research carries profound global implications, given the potentially elevated misdiagnosis rates and delayed diagnosis proportions of MPPS on a worldwide scale, coupled with a potential scarcity of seasoned healthcare providers. Moving forward, continual refinement and validation of the model will be imperative to further augment the precision of MPPS risk assessment, thereby furnishing clinicians with more dependable decision-making support in clinical practice.