JMIR Publications

ABSTRACT

Background:

Screening for influenza in primary care is challenging due to the low sensitivity of rapid antigen tests and a lack of proper screening tests.

Objective:

We developed a machine learning-based screening tool using patient-generated health data (PGHD) obtained from a mobile application (mHealth app).

Methods:

We trained a deep learning model based on GRU to identify influenza based on the PGHD, using each patient’s fever pattern, drug administration records, app-based surveillance calculated from the number of weekly influenza users reported through the app, and meteorological data. We defined a single episode as the set of consecutive days containing the day the user was diagnosed with influenza or other diseases. Any record a user entered after 24 hours from his or her last record was considered as belong to a new episode. Each episode must contain user’s age, gender, weight, and at least one body temperature records. The total number of our dataset was 6,657, of which 3,189 were diagnosed with influenza.

Results:

We achieved reliable performance with an accuracy of 82%, sensitivity of 84%, and specificity of 80% in test set. To evaluate the effect of each input variable, we conducted two experiments. One is removing a variable one by one and observe the change of performance, Another is adding the variable one by one to the base features and observe the change of performance. As a result, app-based surveillance turned out to be most influential variable. We also looked at the correlation between the duration of input data and performance. The Spearman’s rank correlation coefficient was 0.09162, which means the association was not significant.

Conclusions:

These findings suggest that PGHD from a mHealth app could be a complementary tool for influenza screening. Especially, it could be good screening method for infectious disease. In addition, PGHD, along with traditional clinical data, could be used to help improve health conditions.