Nurosene's NURO app (nurosene.com) is an innovative smartphone application that gathers and analyzes active self-report metrics from users, empowering them with data-driven health machine intelligence. We present the data collected and analyzed from the initial round of participants who responded to a 12-question survey on their life-style and health status. Exploratory results using a variational autoencoder (VAE) suggested that much of the variability of the 12 dimensional data could be accounted for by two approximately uncorrelated latent variables: one pertaining to stress and sleep, and the other pertaining to exercise and diet. Subsequent modeling of the data using exploratory and confirmatory factor analyses (EFAs and CFAs) found that optimal data fits consisted of four factors, namely exercise, diet, stress, and sleep. Covariance values were high between exercise and diet, and between stress and sleep, but much lower between other pairings of non-identical factors. Both EFAs and CFAs provided extra contexts to and quantified the more preliminary VAE observations. Overall, our results significantly reduce the apparent complexity of the response data. This reduction allows for more efficient future stratification and analyses of participants based on simpler latent variables. Our discovery of novel relationships between stress and sleep, and between exercise and diet suggests the possibility of applying predictive analytics in future efforts.
Keywords: confirmatory factor analyses; digital health technology; exploratory factor analyses; machine learning; mental and physical wellbeing; smartphone application; survey questionnaire; variational auto encoder.
Copyright © 2022 Gallucci, Ho, Geraci, Loren and Pani.