View Abstract

A2637 - Developing Multivariable Prediction Models of Asthma Control Components Using Fitness Tracker Sleep Patterns in Women
Author Block: J. Castner1, C. Jungquist2, M. J. Mammen3, J. J. Pender2, O. Licata4, G. E. Wilding5, S. Sethi6; 1On file, Buffalo, NY, United States, 2School of Nursing, University at Buffalo, Buffalo, NY, United States, 3Department of Medicine, SUNY at Buffalo, Buffalo, NY, United States, 4School of Engineering and Applied Sciences, University at Buffalo, Buffalo, NY, United States, 5School of Public Health and Health Professions Biostatistics, University at Buffalo, Buffalo, NY, United States, 6Medicine, University at Buffalo, Buffalo, NY, United States.
Rationale The objective of this study was to use fitness tracker sleep patterns to develop predictive models of daily disease control-related asthma-specific wakening and airway obstruction in working-aged women with poorly controlled asthma. Night-time wakening with asthma symptoms are an important indicator of disease control and severity. The ubiquitous nature of inexpensive fitness trackers (e.g. Fitbit) is promising for longer-term, objective assessment of sleep patterns in the individual’s natural home environment. However, there is a paucity of evidence on the potential of fitness-tracker sleep disruption, and more specifically wake counts, as a predictive tool for asthma management and control in women. Methods A repeated measures panel design was used with data from 43 women with data collected for at least 14 consecutive days in their own homes from February to September of 2016. The primary outcomes, measured daily, were 1) self-reported asthma-specific wakening and 2) difference form predicted for self-administered spirometry forced exhaled volume in one second (FEV1). The primary predictor was the nightly wake count, measured by the Fitbit Charge fitness tracker. Additional candidate predictors were collected using an electronic symptom diary, medical record review, and an interview-assisted survey using validated instruments. The model was developed using random-effects and nested multivariate logistic regression. Predictive accuracy was assessed using the area under the receiver operating characteristic curve (AUC). Results Our model demonstrated good predictive value (AUC=.81) for asthma-specific night time waking using fitness tracker wake counts, history of sleep apnea, fatigue, asthma-specific quality of life, dyspnea, and alcohol consumption scores. Our model demonstrated fair predictive value (AUC=.77) for daily FEV1 based on fitness tracker wake counts, exposure to cigarette smoke in the home in the last week, and daily symptoms of wheezing. The fitbit-measured wake count variable was not a statistically significant predictor of FEV1 in our adjusted analysis. Conclusions The fitness tracker wake counts can act as a predictor for, or proxy of, asthma-specific night-time wakening. While not a significant variable in the final model, the fitness tracker wake counts improved the overall predictive model for FEV1, demonstrating that utilizing the fitness tracker results along with electronic daily symptom diaries for wheezing holds ongoing promise for mhealth applications. Our model can be used to inform ongoing informatics development and validation for real-time surveillance, self-care mobile applications, and electronic medical record decision support incorporating real-time fitness tracker data for monitoring of asthma.