Over half the mental health apps in a curated database were gone within five years

Design

Five-year longitudinal survival analysis with Cox regression and random forest classification

Population

865 mental health apps listed in the M-Health Index and Navigation Database

Primary outcome

Removal from the database over 2,142 days

Effect

464/865 (53.6%) removed; Android-only apps HR 2.32 for attrition; Cox C-index 0.77, random forest AUC 0.82

Clinicians recommend mental health apps on content and on star ratings. Neither predicts whether the app will still be there next year. This analysis used the M-Health Index and Navigation Database — a curated catalogue, not the open store — and ran a five-year survival analysis on 865 apps, with Cox regression across 54 clinical, technical and commercial predictors and a random forest classifier for two-year survival.

Of 865 apps, 464 (53.6%) were removed over the 2,142-day study period. Longevity differed by clinical target: sleep apps lasted longest, smoking cessation apps were the most likely to go (setting aside a small schizophrenia app cohort). The strongest single predictor of removal was platform exclusivity. Android-only apps carried a hazard ratio of 2.32 for attrition. The Cox model reached a C-index of 0.77 and the random forest an AUC of 0.82.

This changes the recommendation itself, not just the caveat attached to it. An app that disappears mid-treatment is a treatment discontinuation that nobody logs — the patient simply stops, and you find out at the next appointment or not at all. Recommending from a maintained list and checking cross-platform availability turns a coin toss into something closer to a considered referral.

• Recommend from a short maintained list rather than from memory or a store search
• Cross-platform availability is the single best durability signal in this analysis
• Warn patients that apps withdraw without notice and local data goes with them
• In India the Android-only skew matters more, since most patients are on Android
• Ask at follow-up whether the app is still working, the same way you would ask about a medicine

Why it matters

The durability of a digital recommendation has not been part of the decision, and it turns out to be close to a coin toss.

Don't overread it

This measured removal from a curated database, which is a proxy for an app becoming unavailable or obsolete, not a direct measure of patient harm.

The statistics, in plain English

A hazard ratio of 2.32 means Android-only apps were removed at roughly twice the rate over the study period, not that they were twice as bad. A C-index of 0.77 and AUC of 0.82 mean the models rank apps by survival risk reasonably well but are nowhere near deterministic for a single app. And the sample was a curated database, so the open marketplace is likely worse, not better.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.