For millions of people living with obstructive sleep apnea, the standard of care has long been a mask and a machine. Continuous positive airway pressure, or CPAP, is widely regarded as the most effective way to retain airways open during sleep, yet the medical community has struggled with a persistent paradox: while the therapy treats the breathing disorder, large-scale studies have not consistently shown that it lowers the risk of heart attack or stroke.
New research from Mount Sinai suggests the problem may not be the therapy itself, but a “one-size-fits-all” approach to prescribing it. Researchers have developed a machine learning model to predict how CPAP affects cardiovascular disease risk, allowing clinicians to identify which patients are likely to benefit from the treatment and, crucially, which patients might actually be harmed by it.
The findings, published in Communications Medicine, mark a shift toward precision medicine for a condition that affects an estimated 25 million people in the United States. By analyzing a vast array of health markers, the AI tool can estimate an individual’s cardiovascular risk response to CPAP, moving beyond general averages to personalized predictions.
The implications are stark. The model identified a specific subgroup of patients who saw a 100-fold improvement in future cardiac risk when using CPAP compared to those receiving usual care. Conversely, the AI flagged another subgroup for whom the therapy was associated with a greater than 100-fold increase in adverse cardiovascular outcomes, including recurrent strokes and heart attacks.
Decoding the “SAVE” Trial Data
To build the model, the Mount Sinai team leveraged data from the Sleep Apnea Cardiovascular Endpoints (SAVE) trial. This represents the largest clinical cohort of its kind, involving more than 2,600 participants across 89 sites in seven different countries. The goal was to move away from simple pattern recognition and toward “causal reasoning”—understanding not just that a patient has sleep apnea, but how a specific intervention changes their biological trajectory.
The researchers didn’t rely on a few basic metrics. They began by screening more than 100 different predictors derived from sleep and health records. From this massive dataset, they distilled 23 key baseline features—including smoking status and prior medical conditions—to create the final analysis model.
“Our findings represent a significant advancement in personalized medicine, moving away from a one-size-fits-all strategy in the treatment of obstructive sleep apnea,” said co-corresponding author Neomi A. Shah, MD, MPH, MSC, Professor of Medicine and Artificial Intelligence and Human Health at the Icahn School of Medicine at Mount Sinai. According to Dr. Shah, these data-driven approaches can assist clinicians in making more informed, individualized treatment recommendations.
The Shift Toward Predictive AI in Cardiology
The integration of artificial intelligence into sleep medicine is an attempt to solve a long-standing clinical frustration. For years, physicians have known that obstructive sleep apnea—a condition where breathing repeatedly stops and starts—is linked to elevated risks of stroke and heart disease. However, the lack of clear, universal evidence that CPAP prevents these events has left some providers uncertain about the long-term cardiovascular utility of the device.
By using machine learning, the researchers were able to uncover the hidden variance in how patients respond. The model acts as a decision-support tool, estimating “individualized treatment effect scores” to determine if the therapy is likely to be a lifesaver or a risk factor for a specific person.
However, the researchers are cautious about immediate implementation. Co-primary author Oren Cohen, MD, Assistant Professor of Medicine at the Icahn School of Medicine, noted that while these results demonstrate the power of machine learning, such models require “careful validation to prove their utility in clinical practice.”
Understanding the Risk Profiles
The study highlights a critical divide in patient response that traditional clinical trials often obscure by averaging results across a whole group:
- The High-Benefit Group: Patients whose health markers aligned with the model’s “benefit” prediction saw a massive reduction in cardiac risk when using CPAP.
- The High-Risk Group: Patients predicted to be harmed experienced a significant increase in cardiovascular events, such as heart attacks, when using the therapy.
- The Neutral Group: Patients for whom the therapy may treat the sleep disorder but provide little to no change in overall cardiovascular risk.
Mayte Suarez-Farinas, PhD, Co-Director for the Division of Biostatistics and Data Science at the Icahn School of Medicine, emphasized that the goal is to move predictive AI toward tools “grounded in causality.” This ensures that the AI isn’t just guessing based on a trend, but is informing real-world treatment decisions that can actually improve patient outcomes.
What This Means for Patients
For the average patient, this research does not mean abandoning CPAP, but rather advocating for a more nuanced approach to its prescription. The study was a collaborative effort involving global institutions, including The George Institute for Global Health and the University of New South Wales in Australia, underscoring the global nature of the sleep apnea epidemic.
| Factor | Traditional Approach | AI-Driven Approach |
|---|---|---|
| Treatment Strategy | One-size-fits-all CPAP | Personalized risk-stratification |
| Data Inputs | Basic sleep study results | 23 key baseline health features |
| Primary Goal | Stop breathing interruptions | Reduce cardiovascular disease risk |
| Outcome Prediction | Average cohort response | Individualized treatment effect scores |
The research was supported by funding from the National Heart, Lung, and Blood Institute at the National Institutes of the Health, the American Academy of Sleep Medicine Foundation, and the Stony-Wold Herbert Fund.
Disclaimer: This article is for informational purposes only and does not constitute medical advice. Patients should consult with their healthcare provider before making any changes to their prescribed treatment or CPAP therapy.
The next phase for this technology involves further clinical validation to ensure the model’s predictions hold up across diverse, real-world patient populations before it can be integrated into standard electronic health records. Future updates from the Icahn School of Medicine are expected to clarify how these tools will be deployed in clinical settings.
Do you have experience with CPAP therapy or sleep apnea? We invite you to share your thoughts and comments in the section below.
