Explainable Machine Learning Assists in Revealing Associations Between Polysomnographic Biomarkers and Incident Type 2 Diabetes in Men

Abstract

Introduction: Type 2 diabetes (T2D) shows bidirectional relationships with polysomnographic measures. However, no studies have searched systematically for novel polysomnographic biomarkers of T2D. We therefore investigated if state-of-the-art explainable machine learning (ML) models could identify new polysomnographic biomarkers predictive of incident T2D.

Methods: We applied explainable ML models to longitudinal cohort study data from 536 males who were free of T2D at baseline and identified 52 cases of T2D at follow-up (mean 8.3, range 3.5-10.5 years). Beyond ranking biomarker importance, we explored how the explainable ML model approach can identify novel relationships, assist in hypothesis testing, and provide insights into risk factors.

Results: The top five most predictive biomarkers included waist circumference, glucose, and three novel sleep biomarkers: the number of 3% desaturations in non-supine sleep, mean heart rate in supine sleep, and mean hypopnea duration. Explainable machine learning identified a significant association between the number of non-supine desaturation events (threshold of 19 events) and incident T2D (Odds ratio = 2.4 [95% CI 1.2-4.8], P = 0.013). No significant associations were found using continuous or quartiled versions of non-supine desaturation. Additionally, the model provided an individualized risk factor breakdown, supporting a more personalized approach to precision sleep medicine.

Conclusion: Explainable ML supports the role of established biomarkers and reveals novel biomarkers of T2D likely to help guide further hypothesis testing and validation of more robust and clinically useful biomarkers. Although further validation is needed, these proof-of-concept data support the benefits of explainable ML in prospective data analysis.

Keywords: explainable machine learning; obstructive sleep apnoea; polysomnographic biomarkers; type 2 diabetes.

Figure 1

Schematic demonstrating data-driven approaches for exploring the association between sleep biomarkers and type 2 diabetes. (a) Flow diagram of the study and follow-up assessments. (b) Biomarkers and features from different domains including sleep, breathing disorders, and demographics were extracted to input into the XGBoost machine learning model. (c) The trained model was passed through the AI Explainer framework for estimating the relative importance of biomarkers based on Shapley values. (d) The least important biomarker was removed from the data, and this process repeated until the top 15 most important biomarkers were found (d and e). (e) The output of the explainable ML model was used to reveal novel biomarkers (f), assist hypothesis testing (g), and gain insights into risk factors and assist treatment approaches (h).

Figure 2

Explainable ML reveals novel biomarkers associated with diabetes. (a) Top 15 strongest biomarkers associated with incident T2D. The ranking is based on the mean absolute Shapley values (see Methods for details). Colors indicate the values of biomarkers categorized into four quartiles. A positive change in log odds indicates a higher risk of T2D. (b) The relationship between the top 15 biomarkers and the risk of T2D. The change in log odds is indicated on the y-axis, while the change in biomarker values is shown on the x-axis. The black dashed lines represent smooth curves across the data points to illustrate the overall shape of the relationship. Dot points indicate data from 536 participants.

Figure 3

Comparison of typical scenarios leading to different outcomes at follow-up assessment. (a and b) Comparison of two participants with serum low glucose concentrations at baseline. (c and d) Comparison of two participants with high serum glucose concentrations at baseline. Red and blue colors indicate increased or decreased risk, respectively. The baseline risk for all participants was calculated as the mean of the model output, indicated by green dashed lines (log odds = −0.193), while the final risk outcome is indicated by red dashed lines after adding the additive contributions of all other risk factors.