JMIR Publications

ABSTRACT

Background:

mHealth App was believed as an effective method to support family caregivers to better care of stroke patients. This study aimed to explore the status and the influencing factors of mHealth App utilization among family caregivers of stroke patients via machine-learning (ML) models.

Objective:

The purpose of this study was to understand the status quo of mHealth App use among community family caregivers of stroke patients and the factors influencing their use behavior. Six machine learning models were used to construct the classifier, and SHAP algorithm is introduced to interpret the best machine learning model.

Methods:

In this cross-sectional study, we included family carers of stroke patients into the study, whose basic profile and mHealth App usage were obtained through face-to-face questionnaires, and for the users, additional measurements of hedonic motivation, usage habits and other information were taken. Six ML algorithms were adopted and a total of 12 models were constructed, of which the top performing Random Forest (RF) and Logistic Regression (LR) models were further analysed for SHAP interpretability to gain a deeper understanding of the key factors affecting the model output.

Results:

A total of 360 family caregivers of stroke patients were included in this study from March to November 2023, of which 206 reported having used the mHealth Apps, with a usage rate of 57.22%. Of the six ML models, the RF model performed the best in terms of whether or not caregivers used the mHealth App, with an area under the curve (AUC) of 0.735 (0.635, 0.822), accuracy of 0.722 (0.630, 0.806), sensitivity of 0.742 (0.635, 0.843), and specificity of 0.696 (0.558, 0.825). SHAP analysis showed that the top five most influencing factors were age, literacy, relationship with the cared-for person, the patient's ability to care for him/herself, and average monthly income. The LR model performed best in terms of use behaviour with an AUC of 0.800 (0.681, 0.904), accuracy of 0.742 (0.629, 0.839), sensitivity of 0.662 (0.503, 0.811) and specificity of 0.828 (0.764, 0.893). SHAP analysis revealed that usage habits, performance expectations, hedonism, effort expectations, and price value were the five most significant influencing factors .

Conclusions:

The results suggest that mHealth Apps software developers and policy makers should take the above influencing factors into consideration when developing and promoting software, focusing on older adults with lower literacy levels, and lowering the threshold and cost of software use. Capture the hedonism and habitual use of users and provide them with more concise and accurate health information， which will increase the popularity and effectiveness of mHealth App.