Predictor Of Mortality In Obstructive Sleep Apnoea: Results of Explainable Deep Learning based Survival Analysis From...

Area of Science:

• Medical research
• Respiratory medicine
• Artificial Intelligence in healthcare

Background:

• Obstructive sleep apnoea (OSA) is a common respiratory disorder associated with numerous comorbidities.
• Continuous positive airway pressure (CPAP) is a primary treatment, but its long-term benefits require further investigation.

Purpose of the Study:

• To investigate the impact of comorbidities on mortality in OSA patients.
• To compare the reliability of survival analysis models enhanced with eXplainable Artificial Intelligence (XAI).

Main Methods:

• Survival analysis combined with eXplainable Artificial Intelligence (XAI) techniques.
• Utilized a dataset of 45 clinical and polysomnography features from 1394 OSA patients followed for 15 years.
• Employed Deep Learning (DL) models and time-dependent XAI for feature analysis.

Main Results:

• Key mortality predictors identified include age, years of CPAP, renal dysfunction, COPD, BMI categories, sex, and anemia.
• Other significant factors were Apnoea-Hypopnoea Index (AHI) and minimum oxygen saturation (SaO2 min).
• Model reliability was assessed using CoxTime and LogHazard approaches.

Conclusions:

• Advanced age, OSA severity, and comorbidities such as chronic kidney disease, COPD, and anemia significantly increase mortality risk.
• The study highlights the importance of managing comorbidities in OSA patients.
• XAI techniques provide valuable insights into survival prediction models for OSA.