Background: Short (<6hrs) and long (>8 hrs) sleep durations are linked to stroke and cardiovascular disease. However, results have relied primarily on regression analysis, which may not be optimal to model associations between sleep and medical outcomes. Big-data and machine learning analyses provide unique opportunities to quantify dynamic interactions between sleep and medical outcomes, adjusting for multiple risk factors.
Method: The current study utilized two types of analyses: logistic regression and Bayesian Belief Network, a form of machine learning analysis, to assess sleep-related stroke risk. We used data from the 2004-2013 National Health Interview Survey, yielding 288,888 cases, to investigate how short (<7hrs.) and long (>8hrs.) sleep durations impact stroke risk. In both analyses, we assessed the contribution of 34 demographic, medical, behavioral, and psychosocial factors. We used SPSS 20 to conduct regression analyses and BayesiaLab’s Tree Augmented Naïve Bayes learning algorithm for complex system analysis. We compared results of both analytic models to determine their ecological and clinical utility.
Results: Forty-eight percent of volunteers were ≤45 yrs.; 77.40% were White; 15.96%, Black/African American; and 45.1% made < $35K annually; 29.55% reported short sleep and 8.9%, long sleep; 61.55% were average sleepers (7-8hrs.). Adjusted regression models indicated that relationships between short sleep and stroke were not significant (OR=.97, 95% CI=0.92-1.02, p=.21); however, long sleep was associated with stroke (OR=1.43, 95% CI= 1.32-1.52, p<.001). Results from Bayesian analysis indicated both short and long sleep were associated with stroke, but that long sleep doubled stroke risk (7.48%) relative to short sleep (3.74%). Regression model had a R2 of 0.24 for short sleep and long sleep, while the R2 for Bayesia was 0.73.
Conclusion: Bayesian Belief Network analysis is superior to regression modeling because it provides ecologically and clinically valid findings. Although both short and long sleep durations are associated with stroke risk, long sleep seems to be a stronger predictor.