# Machine learning model based on survey assessment of sleep quality in chronic obstructive pulmonary disease patients

## TL;DR

A machine learning model was developed to predict sleep quality in COPD patients using clinical data and survey responses.

## Contribution

The novel contribution is a machine learning model that identifies clinical factors affecting sleep quality in COPD patients.

## Key findings

- The model achieved 70.99% accuracy in predicting sleep quality.
- High exacerbation rates, EDS, and active smoking were found to negatively affect sleep quality.
- CAD under treatment was associated with better sleep quality.

## Abstract

The aim is to develop a learning model based on clinical and survey data to assess sleep quality and identify determining factors affecting sleep quality in chronic obstructive pulmonary disease (COPD) patients.

The Pittsburgh Sleep Quality Index (PSQI) was administered to stable COPD patients to assess sleep quality. Patients were categorized into two groups: good sleep quality and poor sleep quality. Parameters for the best model were selected from a total of 61 clinical and laboratory parameters using recursive feature elimination (RFE) and the Bayesian Information Criterion (BIC). A logistic regression (LR) model was created. The model was evaluated using nested cross-validation with 5 inner and 5 outer folds, and this process was repeated with 1000 bootstrap iterations. Results were obtained with a 95% CI.

The mean age of the 132 patients was 66.68 ± 8.16 years, with a predominance of males (117, or 88.6%). Of the 132 patients, 68 were in the poor sleep quality group. In this group, the prevalence of dyspnea, snoring, witnessed apneas, and excessive daytime sleepiness (EDS) was higher. The parameters included in the model and occurrence rates in the poor sleep quality group are as follows: annual exacerbation and hospitalization (71.9%), presence of EDS (35.9%), cough (64.1%), active smoking (95.4%), short-acting beta agonist (SABA) requirement (59.4%), pH level, and coronary artery disease (CAD) (20.3%). In our final model, the test set demonstrated a sensitivity, specificity, accuracy, and AUC of 70.21%, 71.76%, 70.99%, and 0.757, respectively.

Our machine learning model, developed using clinical data of COPD patients, can predict their sleep quality. We found that high annual exacerbation and hospitalization rates, the presence of EDS and cough symptoms, active smoking, and regular use of SABA as well as high pH levels, negatively affect sleep quality. Conversely, the presence of CAD under treatment in patients positively affects sleep quality.