Sequential Associations of Sleep Architecture, Nocturnal Oxygenation, and Respiratory Disturbance With a Composite PSG‐Derived Rhythm‐Event Index

BACKGROUND: Sleep-disordered breathing (SDB), particularly obstructive sleep apnea (OSA), has been associated with cardiovascular complications including cardiac arrhythmias. However, the relative contributions of respiratory, oxygenation, and sleep architecture indices to composite PSG-derived rhythm-event index remain incompletely understood. OBJECTIVE: To investigate the sequential associations of sleep architecture, nocturnal oxygenation, and respiratory disturbance indices with the composite PSG-derived rhythm-event index in patients undergoing polysomnography (PSG). METHODS: Demographic, respiratory, oxygenation and sleep-architecture variables were extracted from PSG records. The primary outcome was a composite PSG-derived rhythm-event count, estimated as round(arrhythmia index × TST_hours) from the PSG-exported arrhythmia index, which was calculated during total sleep time only. Negative binomial regression used an offset of log(TST_hours). Because the outcome demonstrated overdispersion, negative binomial regression with an offset of log(TST hours) was used. Because the original 14-predictor model was over-parameterized for the sample size, a six-predictor model (age, sex, BMI, AHI, T90, and sleep efficiency) was chosen post hoc. This reduced model was not prespecified before the original analysis. Sleep efficiency was retained because it showed a strong signal in the exploratory 14-predictor model. The resulting p-values should therefore be interpreted as optimistic and exploratory. A sequential model-building approach (Model 1: age, sex, BMI, AHI; Model 2: plus T90; Model 3: plus sleep efficiency) was used to document the incremental contribution of each domain. A fully adjusted 14-predictor model was fitted only as a supplementary analysis. Wald 95% CIs and bootstrap percentile 95% CIs (2000 resamples) are reported. RESULTS: ). Severe OSA (AHI ≥ 30 events/hour) was present in 56.1%. Median arrhythmia index was 3.2 events/hour, with marked overdispersion. In the six-predictor model, only higher sleep efficiency was associated with lower arrhythmia rate (IRR = 0.946 per 1 percentage-point increase, 95% CI 0.921-0.972, p < 0.001; bootstrap 95% CI 0.888-0.984). T90 was associated in the Wald inference (IRR = 1.034, 95% CI 1.006-1.063, p = 0.016) but this association was not robust under bootstrap resampling (95% CI 0.979-1.067). AHI was not associated with arrhythmia rate (IRR = 1.000, 95% CI 0.987-1.014, p = 0.954). Age, sex and BMI were also not significant. A 14-predictor exploratory model had a marginally lower AIC (411.4) but a substantially higher BIC (438.8), consistent with overfitting. CONCLUSION: In this exploratory selected sleep-center cohort, AHI and other respiratory-event indices were not independently associated with the composite PSG-derived rhythm-event index. Only higher sleep efficiency showed a robust association with lower arrhythmia rate, but this finding is hypothesis-generating and may reflect unmeasured confounding. These results require confirmation in larger prospective studies and should not guide clinical risk stratification.