AI-driven clinical decision support for early diagnosis and treatment planning in patients with suspected sleep apnea using clinical and demographic data before sleep studies.

OBJECTIVE

This study explored the application of Machine Learning (ML) techniques to cluster patients with suspected sleep apnea (SA), based on clinical-demographic data, with the aim of optimizing diagnostic pathways and enabling more personalized management.

METHODS

A cohort of 5385 patients referred for suspected SA to a Sleep-Disordered Breathing Unit in northwest Spain was analyzed. Demographic, anthropometric, comorbidity, and symptom data were collected. Patients were grouped using the k-prototypes algorithm, with the elbow method determining the optimal number of clusters. These clusters were then correlated with cardiorespiratory polygraphy outcomes and continuous positive airway pressure (CPAP) prescription rates. Finally, we developed an Intelligent Clinical Decision Support System (ICDSS) based on Random Forest to assign new patients to clusters using a reduced set of variables.

RESULTS

Five distinct clusters were identified: one of middle-aged men with low symptom burden; a cluster predominantly comprising symptomatic women with high use of psychotropic drugs; a group mainly of young men with severe daytime sleepiness; a cluster of middle-aged men with moderate symptoms; and a group of older men with high comorbidity yet low subjective symptomatology. Significant differences in apnea-hypopnea index (AHI) distributions and CPAP indications were observed among these clusters. The integration of polygraphic findings, CPAP prescription rates, and the distinct clinical features of each cluster supports the formulation of tailored diagnostic and therapeutic strategies according to the specific clinical profile of each subgroup. Using the ICDSS, we accurately assigned patients to their respective clusters based solely on clinical variables, achieving area under the receiver operating characteristic curve (AUC) values ranging from 0.87 to 0.95, reliably guiding precise diagnostic and therapeutic management.

CONCLUSIONS

ML techniques applied to routine data allow the identification of meaningful clinical clusters in patients with suspected SA. These clusters can guide differential diagnostic testing and personalized treatment strategies. The ICDSS enables early and accurate patient classification, supporting a precision medicine approach in sleep medicine.

Read full article here.