PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
Synapse
⌘+K
Synapse
May 10, 2026SLEEP

0509 Applying Machine Learning to Predict Obstructive Sleep Apnea Using Electronic Health Records

View Full Paper
Ask AI
Bookmark
Share

Why the study?

Predicting patients at risk of obstructive sleep apnea may be incorporated into clinical screening tools.

Do machine learning models utilizing electronic health record data accurately predict obstructive sleep apnea in adults?

Population

285,292 adults who underwent diagnostic sleep studies with available AHI4% values in Kaiser Permanente Southern California

Comparison

Logistic regression vs random forest vs XGBoost models using EHR data

Design

Retrospective cohort prognostic model development and validation study

Key result

Random forest machine learning models utilizing EHR data demonstrated strong discrimination for predicting OSA (ROC-AUC 0.84) and moderate-severe OSA (ROC-AUC 0.82).

Authors

NHN. HwangMWM Brandon WestoverDMDiego Mazzotti

Discussion

Loading...

Member takes

Overview

May aid OSA detection from routine labs but should not yet change practice; leaves open prospective validation before clinical adoption.**[[1]](https://pmc.ncbi.nlm.nih.

Key Points

  • This study aims to predict obstructive sleep apnea (OSA) using machine learning models based on electronic health record (EHR) data.
  • Developed models using EHR data from adults who underwent sleep studies from 2016-2025.
  • Utilized multiple machine learning algorithms: logistic regression, random forest, and XGBoost.
  • Data was split into training (80%) and testing (20%) sets for model evaluation.
  • Random forest achieved the highest performance for predicting OSA with an ROC-AUC of 0.84 and sensitivity of 0.96.
  • For moderate-severe OSA, random forests again showed strong performance with an ROC-AUC of 0.82 and specificity of 0.81.
  • Key predictors included age, male sex, BMI, cardiometabolic comorbidities, and metabolic laboratory biomarkers.

Study Design

Type

Observational (n=285,292)

Structured PICO

Do machine learning models utilizing electronic health record data accurately predict obstructive sleep apnea in adults?

P
Population
285,292 adults who underwent diagnostic sleep studies (2016-2025) with available AHI4% values in Kaiser Permanente Southern California. Mean age 50.5±16.7 years, 57% male.
I
Intervention
Machine learning models (logistic regression, random forest, and XGBoost) utilizing electronic health record (EHR) data including demographics, comorbidities, vitals, and laboratory values.
O
Outcome
Prediction of OSA (AHI ≥5) and moderate-severe OSA (AHI ≥15)

Main Result

Effect estimate: ROC-AUC 0.84

EHR-based machine learning models, particularly random forests, demonstrate strong potential for predicting obstructive sleep apnea using cardiometabolic risk features.

Limitations

  • Further refinement, model optimization, and real-world validation are needed
  • Further refinement and model optimization needed
  • Lacks real-world validation

Cite This Study

Hwang et al. (2026) conducted an observational in Obstructive sleep apnea (n=285,292). Machine learning models (Random forest, Logistic regression, XGBoost) was evaluated on Prediction of OSA (AHI≥5) and moderate-severe OSA (AHI≥15) (ROC-AUC 0.84). Random forest machine learning models utilizing EHR data demonstrated strong discrimination for predicting OSA (ROC-AUC 0.84) and moderate-severe OSA (ROC-AUC 0.82).

synapsesocial.com/papers/6a002126c8f74e3340f9c0a5https://doi.org/10.1093/sleep/zsag091.0508
View Full Paper
Ask AI
Bookmark
Share