Related Experiment Videos
Study on diagnostic models for insomnia and gastralgia with Liver-Spleen Disharmony Syndrome based on machine
Enshi Lu1, Xiaoliang Zhao1, Hongjiao Li1
1Institute of Basic Research in Clinical Medicine, China Academy of Chinese Medical Sciences, Beijing, China.
Background:
The prevalence of insomnia and gastralgia has recently increased. Traditional Chinese Medicine (TCM) offers unique advantages in treating these conditions. Accurate syndrome differentiation is essential for effective treatment. However, as a common syndrome in TCM for these conditions, the diagnostic criteria for Liver-Spleen Disharmony Syndrome (LSDS) currently lack uniformity.
Objective:
To develop a machine learning-based diagnostic model for LSDS in patients with insomnia and/or gastralgia.
Methods:
In this prospective observational study, clinical data from 575 patients with insomnia and/or gastralgia were collected using structured scales. The data included demographic information and clinical symptoms related to LSDS. Six machine learning algorithms-Chi-square Automatic Interaction Detector (CHAID), C5.0 decision tree, Back-Propagation Neural Network (BPNN), Radial Basis Function (RBF) network, Bayesian Network (BN), and Binomial Logistic Regression Analysis (BLRA)-were employed to construct diagnostic models for LSDS. Model performance was evaluated using receiver operating characteristic (ROC) curves, confusion matrices, and precision-recall (P-R) curves.
Results:
The CHAID model outperformed all others, achieving an area under the ROC curve (AUC) of 0.889 (95% CI: 0.861-0.914), accuracy (AC) of 0.960, sensitivity (SE) of 0.972, specificity (SP) of 0.977, and an F1 score of 0.955. Prediction of variable importance in the CHAID model indicated that depression or irritability was the most important symptom variable for LSDS, followed by epigastric fullness, distending pain in hypochondrium, poor excretion of stool, diarrhea with abdominal pain, and excessive flatus. The C5.0 model ranked second (AUC = 0.797, accuracy = 0.958, F1 = 0.953). Other models showed progressively lower performance: BLRA (AUC = 0.648, accuracy = 0.947, F1 = 0.942), BPNN (AUC = 0.697, accuracy = 0.913, F1 = 0.901), BN (AUC = 0.538, accuracy = 0.944, F1 = 0.935), and RBF (AUC = 0.569, accuracy = 0.824, F1 = 0.810).
Conclusion:
We successfully developed a machine learning-based diagnostic model for Liver-Spleen Disharmony Syndrome (LSDS) in patients with insomnia and gastralgia. The CHAID model demonstrated superior performance and shows promise as an exploratory tool to assist clinicians; however, independent external validation is required before it can be considered for clinical application.
