# Development and Validation of Prediction Models for Severe Obstructive Sleep Apnea Based on Periodic Health Examinations

## TL;DR

This study creates models to predict severe sleep apnea using health exam data, helping identify high-risk individuals without relying on questionnaires.

## Contribution

New models for predicting severe OSA using objective health examination data, independent of subjective questionnaires.

## Key findings

- Two models were developed with acceptable internal and external validation performance.
- The AHI-based model showed high sensitivity and specificity in external validation.
- The models can aid in early detection of severe OSA in occupational health settings.

## Abstract

Obstructive sleep apnea (OSA) is not only associated with reduced work efficiency and an elevated risk of occupational accidents but also with hypertension, diabetes, and other lifestyle‐related diseases, making it an important occupational health concern. Conventional questionnaire–based screening may fail to detect OSA because it frequently lacks subjective symptoms. Herein, we aimed to develop and validate a simple objective, questionnaire‐independent prediction model for severe OSA using periodic health examination (PHE) data.

Following the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD), we analyzed the data of 671 patients who underwent overnight polysomnography (PSG) at Tohoku University Hospital. Eight predictors—age group, sex, obesity, hypertension, diabetes mellitus, dyslipidemia, polycythemia, and liver dysfunction—derived from routine PHE items—were included in logistic regression models to predict severe OSA, defined as an apnea–hypopnea index (AHI) ≥ 30 or a 3% oxygen desaturation index (ODI) ≥ 30. Internal validity was assessed using bootstrap samples. External validation was performed using overnight percutaneous oxygen saturation data of 100 university employees.

The areas under the receiver operating characteristic curve were 0.67 and 0.72 for the AHI‐ and ODI‐based models, respectively. The internal validity was generally acceptable. In external validation, the AHI model had a sensitivity and specificity of 1.00 and 0.95, respectively, while the ODI model exhibited values of 0.50 and 0.97, respectively.

We developed and validated two predictive models for severe OSA using the PHE data. These models could be used for screening by occupational physicians and clinicians.

This study developed and validated two prediction models for severe obstructive sleep apnea using data from periodic health examinations. These models enable early detection of high‐risk individuals and may facilitate timely referral and intervention in occupational and clinical settings.