JMIR Publications

ABSTRACT

Background:

Early surveillance to prevent the spread of influenza is a major public health concern. If a prediction model can be structured using a mobile app data, it may be possible to forecast influenza earlier and more easily.

Objective:

We aimed to develop and validate a prediction model for seasonal influenza using the frequency of mobile app use among children in Osaka Prefecture, Japan.

Methods:

This was a retrospective observational study that was performed over a three-year period from January 2017 to December 2019. Using a linear regression model, we calculated the R-squared value of the regression model to evaluate the relationship between the number of “fever” events selected in the mobile app and the number of influenza patients of ≤14 years of age. We conducted a three-fold cross validation using data from two years as the training dataset and the data of the remaining year as the test dataset to evaluate the validity of the regression model. And we calculated Spearman’s correlation coefficients between the predicted number of influenza patients estimated using the regression model and the number of influenza patients, limited to the period from December to April, when influenza is prevalent in Japan.

Results:

We included 29,392 mobile app users. The R-squared value for the linear regression model was 0.944, and the adjusted R-squared value was 0.915. The mean Spearman’s correlation coefficient for the three regression models was 0.804. In December–April, Spearman’s correlation coefficient between the number of influenza patients and the predicted number estimated using the linear regression model was 0.946 (p<0.001).

Conclusions:

In this study, the number of times that mobile apps were used was positively associated with the number of influenza patients, and the predictive performance of the linear regression model was good. In particular, the predictive performance of the regression model was good during the influenza epidemic season.