Understanding Barriers to Diabetes Self-Management Using Momentary Assessment and Machine Learning (Preprint)

Abstract

For adolescents with type 1 diabetes (T1D), completion of multiple daily self-management tasks such as monitoring blood glucose and administering insulin is challenging due to psychosocial and contextual barriers. Those barriers are difficult to assess accurately and specifically using traditional patient recall. Ecological momentary assessment (EMA) uses mobile technologies to assesses the contexts, subjective experiences, and psychosocial processes that surround self-management decision-making in daily life. However, the rich data generated via EMA has not been frequently examined in T1D or integrated with machine learning (ML) analytic approaches.

The goal of this study was to identify patterns of psychosocial and contextual factors that may impact diabetes self-management assessed by EMA using ML. To achieve this goal, we trained and compared a number of ML models through a learned filtering architecture (LFA) to identify types of barriers that are related to two self-management behaviors: missed mealtime self-monitoring of blood glucose (SMBG) and insulin administration.

We analyzed data from a randomized controlled pilot study using ML-based filtering architecture to investigate whether novel information related to contextual, psychosocial and time-related factors (i.e. time of day) relate to self-management. We combined EMA-collected variables via the MyDay mobile app with Bluetooth blood glucose data to construct ML classifiers in order to predict the two self-management behaviors of interest.

Using 1,244 data points collected from 45 participants, demographic variables and time-related variables had 75.6+% and ~50% accuracy for predicting missed SMBG respectively. For the 1,855 data points derived from 31 participants’ app-based EMA data mood, stress, and fatigue levels and psychosocial barriers were associated with insulin administration and SMBG behaviors, with an average prediction accuracy of ~74%.

Combining EMA data with ML methods may result in enhanced clinical decision-making and just-in-time patient support and can potentially advance personalized behavioral medicine targeting self-management in T1D. Improvements in self-management insights and predictions may result from sub-group analyses and individual behavioral phenotyping.