<img height="1" width="1" alt="" style="display:none" src="https://www.facebook.com/tr?id=648880075207035&amp;ev=NoScript">
3rd Annual BayesiaLab Conference

Presentation on October 7, 2015, at the 3rd Annual BayesiaLab Conference:

Using Machine Learning to Determine effects of Sleep Duration and Physical Activity on Stroke Risk: Analysis of the National Health Interview Survey




Azizi Seixas, Ph.D.
NYU School of Medicine

Background:  Big data and complex system analyses provide unique opportunities to quantify dynamic omnidirectional interactions among multiple factors that impact diseases and health outcomes. Applying this type of analysis to sleep data is crucial, as sleep is linked to a host of chronic medical conditions.

Method: Analysis was based on the 2004-2013 National Health Interview Survey (N=288,888). We employed a machine-learning Bayesian Belief Network (BBN) to model the probabilistic relationships (independent and additive) of sleep duration and physical activity to stroke risk. Factors considered included demographic, behavioral, health/medical, and psychosocial as well as sleep duration (short, average, and long), and physical activity (leisurely walking/bicycling, slow swimming/dancing, and simple gardening activities). 

Results: Of the sample, 48.1% were ≤45 years; 77.4% were White; 15.9%, Black/African American; and 45.1% reported less than $35K annually. Overall, the model had a precision index of 95.84%. Average sleepers (7-8hours) were 25% (2.3% to 3%) less likely to experience a stroke. Respectively, long sleepers (>8hours) were 146% (3% to 7.5%) and short sleepers (<7 hours) 22% (3%-3.74%) more likely to report a stroke.  A model-based adaptive method evidenced that the combined effect of health status, hypertension, heart condition, income, and alcohol consumption increased the likelihood of stroke from 3% to 90%. Healthy sleep (7-8 hours) and vigorous leisurely activity (30-60 minutes) three to six times per week significantly decreased stroke risk.  Using the observational inference technique, we developed idiosyncratic profiles of protective behaviors (i.e. minutes and frequency of moderate or vigorous exercise per week and short, average or long sleep) that reduced stroke risk.

Conclusion: Utilization of BBN analysis is important, as it provides a more dynamic risk stratification system.  Our findings revealed healthy sleep and exercise routines reduced stroke risk, based on systematic iterations using multiple demographic, behavioral, health/medical, and psychosocial conditions and factors.