Minute‐level ECG‐derived autonomic dynamics around obstructive apnea events: A peri‐event analysis of the PhysioNet apnea‐ECG database
Pratyush Kumar, Kumar Abhishek, Pankti Shah, Munaf Siyamwala, Michael J. WaxmanAbstract
Background
Obstructive sleep apnea (OSA) is linked to arrhythmia and cardiovascular morbidity, but the cardiac response to individual obstructive events is incompletely characterised at minute‐scale resolution. We aimed to determine whether ECG‐derived autonomic features distinguish apnea from nonapnea minutes and whether these features change before scored event onset.
Methods
We performed a secondary analysis of the PhysioNet Apnea‐ECG learning set (34 subjects after exclusion of one duplicated recording; 30 male, 4 female; mean age 47.0 ± 9.7 years; AHI 0–82.4 events/h). For each 1‐min segment, we computed mean heart rate, SDNN, RMSSD, pNN50 and RR irregularity rate from QRS‐derived RR intervals. Minutes were classified as baseline, preapnea, apnea, or post‐apnea and associations tested using mixed‐effects regression with subject random intercepts.
Results
The analytic dataset comprised 17,004 min; 38.2% were apnea minutes. Apnea minutes showed a heart rate reduction of −2.66 bpm (95% CI: −2.92 to −2.41, p < 0.001) and an RMSSD increase of +19 ms (95% CI: 15–23, p < 0.001) versus baseline. In multivariable modelling, SDNN was the dominant feature ( β = +0.472); RMSSD showed a sign reversal ( β = −0.416) and reciprocal logistic odds ratios (SDNN OR 134.9 vs. RMSSD OR 0.034) consistent with substantial collinearity. Preapnea minutes showed directionally similar but smaller changes; post‐apnea minutes showed partial normalisation.
Conclusion
Obstructive apnea is associated with a reproducible minute‐level autonomic signature of bradycardia and increased RR variability, with detectable shifts preceding scored event onset. Variability metrics, particularly SDNN, contribute more than mean heart rate. Findings require validation in cohorts with linked cardiovascular outcomes.