Prediction models for sleep quality among frontline medical personnel during the COVID-19 pandemic: cross-sectional study based on internet new media

TL;DR

This study explores factors affecting sleep quality among frontline medical workers during the pandemic and finds that a deep learning model best predicts sleep issues.

Contribution

The study introduces a deep learning model as the most effective predictor of sleep quality among frontline medical personnel during the pandemic.

Findings

Weight, job title, and tea consumption were identified as main factors influencing sleep quality.

The deep learning model showed the best prediction performance with an AUC of 0.656.

75.8% of participants were female, and most were under 35 years old.

Abstract

The factors associated with sleep quality among medical personnel providing support on the frontline during the height of the COVID-19 pandemic remain unclear, and appropriate predictive and screening tools are lacking. This study was designed and conducted to investigate whether factors such as weight change, job title, and tea consumption influence the sleep quality of these workers. Additionally, the study aims to develop predictive models to analyze the sleep problems experienced by healthcare workers during periods of epidemic instability, and to provide relevant data and tools to support effective intervention and prevention strategies.

A cross-sectional study was conducted from June 25 to July 14, 2022, using a self-administered general information questionnaire and the Pittsburgh Sleep Quality Index (PSQI) to investigate the sleep quality of medical personnel providing aid in Shanghai. The relevant influencing factors were obtained via univariate analysis and multivariate stepwise logistic regression analysis, and 80% of the data were used in the training-test set (n = 1,060) and 20% were used in the independent validation set (n = 266). We used snowball sampling to establish the six models of logistics (LG), deep learning (DL), naïve Bayes (NB), artificial neural networks (ANN), random forest (RF), and gradient-boosted trees (GBT) and perform model testing.

Among the participants, 75.8% were female. Those under 35 years of age comprised 53.7% of the medical staff, while those over 35 years accounted for 46.3%. The educational background of the participants included 402 individuals with an associate degree (30.3%), 713 with a bachelor’s degree (53.8%), and 211 with a master’s degree or higher (15.9%).Weight, job title, and tea consumption during the aid period were the main factors influencing the sleep quality of medical personnel during the aid period. The areas under the curve (AUC) of LG, DL, NB, ANN, RF, and GBT were 0.645, 0.656, 0.626, 0.640, 0.551, and 0.582, respectively. The DL model has the best prediction performance (specificity = 86.1%, sensitivity = 45.5%) of all the models.

During the height of the COVID-19 pandemic, the sleep quality of frontline medical personnel providing aid in Shanghai was influenced by multiple factors, and the DL model was found to have the strongest overall predictive efficacy for sleep quality.

Figures