Home > Pulmonology > ATS 2024 > Obstructive Sleep Apnea > Is the Apnea Hypopnea Index ready to be replaced?

Is the Apnea Hypopnea Index ready to be replaced?

Presented by
Dr Sajila Wickramaratne, Icahn School of Medicine at Mount Sinai, NY, USA
ATS 2024
An AI-guided model may be better than an Apnea Hypopnea Index (AHI) model to predict survival in patients with obstructive sleep apnea (OSA). Although the findings of the current study [1] need to be validated and model performance can be improved, AI-guided models appear promising to assess disease severity in this population. As a consequence, physicians may finetune patient management based on the outcomes of these models.

“The commonly used AHI is a tool for physicians to diagnose OSA and to evaluate the severity of this condition,” explained Dr Sajila Wickramaratne (Icahn School of Medicine at Mount Sinai, NY, USA). “However, the instrument, that uses a fixed combination of ventilatory, hypoxic, and arousal domains, fails to capture the depth and breadth of sleep apnea across the domains.” She outlined that ventilatory features, such as ‘event duration’ and ‘% of small breaths’ and arousal features such as ‘arousal intensity’ can characterise the disease burden along each domain. The current study aimed to assess whether a machine learning model, trained with feature from ventilatory, hypoxic, and arousal domains, is better than the AHI at generating survival profiles of patients. Dr Wickramaratne and colleagues used data from individuals who participated in the Sleep Heart Health study and had valid polysomnographs (n=5,074). Ventilatory distribution, hypoxic distribution, and arousal distribution data were created and analysed to generate survival outcomes in the AI model.

The mean follow-up duration was 8.2 years. The AI-model appeared to be slightly better at predicting sleep apnea-related mortality with a C-index of 0.63 compared to a C-index of 0.55 of the AHI-model. When the authors corrected for demographic features, like age, sex, race, BMI, prior history of cardiovascular disease, and smoking status, the AI-model and AHI-model had C-indexes of 0.79 and 0.76, respectively. “Looking at the individual facets of the respective domains, ‘full wakefulness’ appeared to be the most influential factor in this population,” added Dr Wickramaratne.

  1. Wickramaratne SD, et al. Physiology guided AI to predict survival curves for mortality in obstructive sleep apnea. Sleep apnea uplugged: navigating a myriad of health outcomes.

Medical writing support was provided by Robert van den Heuvel.

Copyright ©2024 Medicom Medical Publishers

Posted on