Research on predicting the sleep status of orthopedic pain patients based on machine learning

Machine learning (ML) methodologies offer significant flexibility and scalability, driving their adoption in medical research, including pain studies. However, their potential in analyzing the interplay between pain factors and sleep remains underexplored. This study aimed to investigate the effects of pain-related factors on deep sleep duration in patients and to develop predictive models using ML techniques. Six machine learning models were employed to analyze pain-related factors affecting sleep, with the most stable and accurate model selected for further analysis. The Gradient Boosted Decision Tree (GBDT) model demonstrated superior performance, achieving an internal validation accuracy of 0.897 and an external validation accuracy of 0.885, with area under the curve (AUC) values exceeding 0.6 in both validations. Importance analysis using the GBDT model identified key factors influencing sleep duration. Trauma-related pain was identified as the most significant factor affecting deep sleep duration (Importance score: 0.16), followed closely by the use of alternative medical therapies (Importance score: 0.14). Among the 385 patients included, 203 reported utilizing alternative medical therapies, such as acupuncture, moxibustion, and herbal remedies. Patients using these therapies experienced a reduced risk of disruptions in deep sleep duration, highlighting the potential analgesic and therapeutic effects of these interventions. ML models effectively identify factors influencing sleep in pain patients. Alternative therapies may mitigate deep sleep disturbances, highlighting their potential role in pain management.