Objective:

To evaluate the predictive capabilities of the SleepFM model for the onset of various medical conditions based on overnight polysomnography data, highlighting its significance compared to existing predictive tools.

Key Findings:

• SleepFM achieved a C-Index of at least 0.75 for all 130 conditions, with notable scores for dementia (0.85) and myocardial infarction (0.81), significantly outperforming baseline models.
• The model outperformed baseline models, achieving an AUROC of 0.85 for all-cause mortality compared to 0.78 for demographics and end-to-end PSG models, indicating substantial improvements.
• Specific conditions like Parkinson disease and Alzheimer's showed high predictive accuracy (AUROC of 0.93 and C-Index of 0.91, respectively), underscoring the model's effectiveness.

Interpretation:

The findings suggest that a single night of sleep data can provide significant insights into the future onset of various diseases, highlighting the potential of SleepFM as a predictive tool in clinical settings and its implications for early diagnosis.

Limitations:

• The study may have biases due to the specific cohorts used for training and validation, which could affect the generalizability of the results.
• The model's performance may vary across different populations and settings, necessitating further validation.

Conclusion:

SleepFM demonstrates promising capabilities in predicting a wide range of diseases from sleep data, potentially transforming early diagnosis and preventive healthcare strategies.