JMIR Publications

ABSTRACT

Background:

Currently there are around 20,000 mental health apps available in the app stores. With an increase in mental health related problems and easy access to health apps, it could be assumed that the use of mental health apps will continue to increase in the near future. The Organisation for the Review of Care and Health Apps (ORCHA), a United Kingdom digital health compliance company, has assessed digital mental health apps with regards to their quality, their professional & clinical assurance, data privacy and user experience.

Objective:

To examine the characteristics of mental health apps regarding their quality, underpinning evidence, and data privacy.

Methods:

A dataset comprising of ORCHA Baseline Review (OBR) assessments of over 2000 digital health apps including 436 mental health apps that have been used for this study. R language and R studio were used to perform exploratory data analysis/visualisations, statistics and machine learning to gain insight into the quality and characteristics of mental health apps. Mental health apps have been examined across 15 different target user groups (e.g. adults, and teens). Features’ pairwise correlation has been calculated. Wilcoxon two sample rank sum test has been used to compare distribution of data. With p-value of 0.05 considered as statistically significant. Bonferroni corrected alpha value has been calculated when multiple tests had been conducted on the same data. Descriptive statistics had been conducted on the evidence provided to support mental health apps’ content and data collected by mental health apps. Association rule mining (apriori algorithm) was used to discover associations between the types of data that are collected by mental health apps (e.g. usage data, and emails). Other machine learning techniques (k-modes, random forest) were also used to gain insight into mental health app characteristics.

Results:

Information provision, data capture and data sharing were the most common features within mental health apps. The examined apps primarily targeted the following groups: adults (52.5%), everyone (42.2%), and teens (31.0%). The cost of apps have not been linked to the quality of mental health apps, although paid apps or apps with in-app purchases may include additional services. Indicated user acceptance/benefit is the most common type of evidence provided by these mental health apps. 241 out of 436 (55.3%) apps included a qualified professional in the app development and 251 out of 436 (57.6%) apps provided evidence within the app that the developer validated any guidance with relevant reliable information sources or references. Usage data and emails were the most collected data. Association rule mining showed that Email, IP address, Name and Usage data are often co-collected by the same apps. K-modes cluster analysis showed that mental health apps can be categorised into two clusters where one cluster of apps, 182 out of 436 (45.6%), collected more data than the other.

Conclusions:

Mental health apps are commonly broadly targeted for everyone to use, but many apps are targeted towards either teens or adults. Our study suggests that many of the publicly available mental health apps did not take the precautions (such as the involvement of appropriate health professionals, literature references or conducting tests) to ensure that their content is valid, and research based. Greater effort on behalf of mental health app developers is needed to ensure that the public is provided with high-quality apps. Moreover, our study indicates that the mental health apps that collect more data score better on the ORCHA assessment.