A new method for assessing chronic insomnia: Machine learning-based fusion of TCM observation digital features

Area of Science:

• Integrative medicine
• Medical informatics
• Traditional Chinese Medicine

Background:

• Clinical assessment of chronic insomnia faces challenges with costly and complex objective methods like polysomnography.
• There is a need for objective, varied, and accessible methods for chronic insomnia assessment.

Purpose of the Study:

• To develop and validate a non-invasive classification model for assessing chronic insomnia severity.
• To explore the utility of facial and tongue features in chronic insomnia diagnosis using machine learning.

Main Methods:

• Collected clinical data from 594 chronic insomnia patients, including facial and tongue features analyzed via the tongue face diagnosis analysis-1 instrument.
• Employed stepwise-regression, principal component analysis, and zero-inflated negative binomial (ZINB) regression for variable screening.
• Constructed a classification model using six supervised machine learning methods, evaluated with sensitivity, specificity, F1 score, precision, and accuracy, and visualized using SHAP and DCA.

Main Results:

• Model 4, integrating baseline data, sleep symptoms, and facial features, demonstrated superior performance with a receiver operating characteristic curve value of 0.822.
• Decision curve analysis (DCA) confirmed significant clinical utility for Model 4.
• Shapley Additive exPlanations (SHAP) highlighted the prominence of specific traditional Chinese medicine features (Ch-Y, CH-G, Ch-R) in predicting insomnia severity compared to conventional scales (PSQI, SAS).

Conclusions:

• This study successfully developed a convenient and non-invasive classification model for chronic insomnia severity assessment.
• The integration of facial and tongue features offers a promising advancement for diagnosing and managing chronic insomnia.
• The findings support the use of traditional Chinese medicine-based features in modern machine learning models for improved clinical utility.