Development and internal validation of a LASSO-based nomogram for predicting SAS-defined anxiety symptoms in obstructive sleep apnea.

BACKGROUND

Anxiety symptoms are common but often under-recognized in patients with obstructive sleep apnea (OSA). Conventional polysomnographic severity indices may not fully capture the psychological burden of OSA. This study aimed to develop and internally validate a nomogram for clinically significant anxiety symptoms in patients with OSA.

METHODS

A total of 787 patients with OSA were randomly divided into a training cohort (n=551) and a validation cohort (n=236) at a 7:3 ratio. Demographic characteristics, comorbidities, laboratory parameters, patient-reported scales, and polysomnographic variables were collected. Ten-fold cross-validated LASSO logistic regression was used to select predictors from 36 candidate variables. A post-LASSO multivariable logistic model was constructed to develop the nomogram. Discrimination, calibration, and clinical utility were assessed using ROC analysis, bootstrap calibration, Hosmer-Lemeshow testing, and decision curve analysis. Restricted cubic spline analysis examined the association between longest obstructive apnea duration and anxiety symptoms.

RESULTS

Among the 787 included patients, 71 patients (9.0%) had clinically significant anxiety symptoms. Six predictors were retained: gender, cardiovascular disease, hemoglobin, neutrophil-to-lymphocyte ratio, body mass index, and longest obstructive apnea duration. Patients with anxiety symptoms had poorer subjective sleep quality but relatively milder OSA-related parameters, including lower AHI and ODI, shorter longest obstructive apnea duration, and higher minimum SpO2. Restricted cubic spline analysis showed a significant overall association between longest obstructive apnea duration and anxiety symptoms (Wald χ²=12.21, df=3, P = 0.0067), with borderline non-linearity (P = 0.0583). Compared with the 60-100-second group, patients with a duration <60 seconds had a higher anxiety risk (OR = 2.84, 95% CI 1.43-5.62, P = 0.0028). The nomogram showed acceptable discrimination, with AUCs of 0.735 and 0.751 in the training and validation cohorts, respectively. Calibration was acceptable, with non-significant Hosmer-Lemeshow tests (P = 0.1056 and 0.4948). Decision curve analysis indicated positive net benefit across broad threshold ranges.

CONCLUSION

This LASSO-based nomogram showed acceptable performance for predicting SAS-defined clinically significant anxiety symptoms in patients with OSA. Anxiety risk may not be fully captured by conventional respiratory severity indices alone and may reflect a multifactorial risk profile. External validation is warranted.