Truace
Subscribe

LLM-based analysis of user reviews of reimbursable German digital health applications (DiGAs)

Source article: What Users Say About Reimbursable Digital Therapeutics in Germany: Large-Scale App Store Review Analysis Using a Large Language Model

Background In 2019, Germany introduced a unique regulatory framework for digital therapeutics (DTx), termed digital health applications (DiGAs) in Germany, with the goal of integrating evidence-based DTx into statutory health care. DTx are eligible for reimbursement by statutory health insurance if manufacturers demonstrate positive health care effects, such as improved health status or health literacy, in a controlled study setting. Although regulatory evaluation primarily relies on manufacturer-conducted studi…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1703· cite & verify
What Users Say About Reimbursable Digital Therapeutics in Germany: Large-Scale App Store Review Analysis Using a Large Language Model
MODULE 2 DATAFICATION, DATA PRIVACY AND DIGITAL HEALTH RECORDS by Stiftung Digitale Chancen : Dörte Stahl, Nenja Wolbers Asociacija “Viešieji interneto prieigos taškai”: Monika Arlauskaitė , Laura Grinevičiūtė IASIS NGO : Athanasios Loules , Theodora Alexopoulou, Ilias Michael Rafail Ynte rne t.org : Leonor Afonso , Thanasis Priftis Simbioza Genesis, socialno podjetje : Brigita Dane. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0
Trace impact readingContested
P 68The P score combines the specificity and measured human impact of the grounded problem claim with the strength of this Trace’s cited sources.

Both sides are scored from claims and sources, not community votes.

G 67The G score combines the specificity and measured human impact of the grounded gain claim with the strength of this Trace’s cited sources.

In brief

Researchers collected reviews of 44 reimbursable digital health apps listed in Germany's official BfArM directory from the Apple and Google stores. They used GPT-4o to break each review into statements, label sentiment, and sort them into ten quality themes, then checked the output manually.

The approach suggests that large-scale review mining could add a user-centered layer to postmarket oversight, which currently leans on manufacturer-run studies. What the findings do not show is clinical outcome data, since the results reflect self-selected public reviews and perceived rather than measured effectiveness.

Main points

  1. The dataset covered 44 mobile DiGAs, 4328 reviews and 9439 interpretable statements.
  2. Manual validation found 99% accuracy for sentiment and 95% accuracy for thematic category classification.
  3. Effectiveness (89.5% positive) and overall impression (86.2% positive) received the most favorable feedback.
  4. Login and registration (93.9% negative) and technology aspects (84.3% negative) drew the most criticism.
  5. The authors argue access prerequisites should be treated as essential implementation requirements.

The gain

GPT-4o classified user statements from DiGA app reviews with high accuracy, letting researchers surface user-centered real-world evidence to complement formal manufacturer-led evaluation.

The problem

The same review analysis found that login, registration and technology problems are overwhelmingly negative, pointing to persistent barriers to accessing and using reimbursed DiGAs.

The rundown

Germany introduced the DiGA framework in 2019 so that evidence-based digital therapeutics could be reimbursed by statutory health insurance once manufacturers show positive health care effects in controlled studies. The authors note these studies do not fully capture everyday user experience.

Reviews were extracted from the German Apple App Store and Google Play Store with a tailored Python script. GPT-4o extracted up to five core statements per review, each tagged with a sentiment and one of ten categories derived from DiGA certification criteria, including cost and reimbursement, support and prescription and approval.

Each of the three authors independently reviewed one-third of the model outputs. Overall, feedback was predominantly positive, but recurring complaints about registration, technical reliability and prescription and approval procedures were described as persistent barriers.

Sources

  1. Peer-reviewedJMIR Human Factors2026-10-08

The debate