JMIR Publications

ABSTRACT

Background:

Diabetes mellitus, which causes dysregulation of blood glucose in humans, is a major public health challenge. Patients with diabetes have to monitor their glycemic levels in order to keep them in a healthy range. This task is made easier by using continuous glucose monitoring devices (CGMs) and relaying their output to modern smartphone apps, thus providing users with real-time information on their glycemic fluctuations, possibly along with added statistical data and predictions of future trends.

Objective:

The present study discusses various challenges of real-time predictive monitoring of glycemia and examines accuracy and blood glucose control effectiveness of Diabits, a smartphone application that helps diabetic patients monitor and manage their blood glucose levels in real time.

Methods:

Using data from CGMs and user input regarding meals, exercise, and physical activity, Diabits applies modern machine learning techniques to create personalized patient models and predict future blood glucose fluctuations up to 60 minutes in advance. These predictions are meant to give patients an opportunity to take preemptive action in order to keep their blood glucose values within normal range. In this retrospective observational cohort study, the predictive accuracy of Diabits and the correlation between daily use of the app and standard metrics of blood glucose control were examined based on a large amount of data obtained from the most frequent users of the app. In addition, the accuracy of predictions on the 2018 Ohio T1DM Dataset was calculated and compared against other published results for this dataset.

Results:

Based on over 6.8 million actual in-app predictions for free-living users, the accuracy of Diabits predictions, evaluated using Parkes (Consensus) Error Grid, was found to be 86.89% (5963930/6864130) clinically accurate (zone A) and 99.56% (6833625/6864130) clinically acceptable (zones A and B) for 30-minute predictions, while the results of 60-minute predictions were 70.56% (4843605/6864130) clinically accurate and 97.49% (6692165/6864130) clinically acceptable. By analyzing daily use statistics and corresponding CGM data for the 280 most long-standing users of Diabits, it was established that, under free-living conditions, many common blood glucose control metrics improved with increased frequency of app use. For instance, the average blood glucose for the days these users did not communicate with the app was 154.0 mg/dL, with 67.52% of the time spent in the healthy 70-180 mg/dL range. For days with 10 or more Diabits sessions, the average blood glucose decreased to 141.6 mg/dL (P < .001), while the time in range (TIR) increased to 74.28% (P < .001). On the Ohio T1DM dataset of 6 type 1 diabetic patients that was used in the 2018 Blood Glucose Prediction Challenge, 30-minute predictions of the base Diabits model had an average root mean squared error (RMSE) of 18.68 mg/dL, which is an improvement over published state-of-the-art results for this dataset.

Conclusions:

The obtained results show that Diabits accurately predicts future glycemic fluctuations, potentially making it easier for diabetic patients to keep their blood glucose in the normal range. Furthermore, an improvement in glucose control was observed for app users on the days with higher frequency of Diabits use.