# Characterizing Sleep Disturbance Subgroups and Identifying Associated Factors in Traditional Chinese Medicine Nurses: A Latent Profile Analysis and Explainable Machine Learning Approach

## TL;DR

This study identifies three sleep disturbance subgroups among TCM nurses and uses machine learning to find factors that distinguish them, such as income and resilience.

## Contribution

The novel combination of latent profile analysis and explainable machine learning to classify and interpret sleep disturbance subgroups in TCM nurses.

## Key findings

- Three distinct sleep disturbance subgroups were identified: mild-stable, moderate-fluctuating, and severe-persistent.
- XGBoost achieved the highest classification performance with an AUC of 0.84 in distinguishing subgroups.
- Key factors influencing subgroup classification included monthly income, organizational support, and resilience.

## Abstract

Nurses in traditional Chinese medicine (TCM) departments face significant sleep challenges associated with occupational stressors. However, person‐centered analyses classifying these sleep patterns remain scarce. This study aimed to identify heterogeneous sleep disturbance subgroups via latent profile analysis (LPA) and evaluate the performance of explainable machine learning models in discriminating these subgroups based on demographic and occupational features.

A cross‐sectional survey enrolled 7721 nurses from 130 TCM healthcare institutions in Liaoning Province (December 2024). Data encompassed demographic, occupational, and psychological variables obtained via self‐administered questionnaires, including the Patient‐Reported Outcomes Measurement Information System (PROMIS) Sleep Disturbance short form 8a. LPA was employed to categorize sleep disturbance patterns. Recursive feature elimination with random forest (RFE‐RF) was used to select features associated with subgroup membership for five machine learning models. Models were trained on 70% of the data and evaluated on a 30% independent test set. The optimal classification model (XGBoost) underwent interpretability analysis using Shapley additive explanations (SHAP).

LPA identified three subgroups: mild‐stable (29.8%), moderate‐fluctuating (60%), and severe‐persistent (10.2%). Machine learning models achieved test AUCs of 0.71–0.84, with XGBoost demonstrating the highest discriminatory performance (AUC = 0.84, 95%CI: 0.83–0.85) in classifying subgroups. SHAP analysis indicated that monthly income, organizational support, hospital level, self‐compassion, and resilience were the top five features contributing to the model’s classification output.

This study characterized three distinct sleep disturbance subgroups among TCM nurses, with the majority exhibiting moderate symptoms. The sequential application of LPA and explainable machine learning demonstrated robust performance in distinguishing sleep disturbance patterns. Identifying correlates—such as lower income and resilience—may assist nurse managers in stratifying risk and tailoring interventions for those most likely to fall into the severe subgroup. Future longitudinal studies are required to validate the stability of these subgroups and establish causal relationships.