AI in DiGA: What Does Responsible Innovation Look Like?

Digital health applications are becoming an increasingly important part of modern healthcare. In Germany, Digital Health Applications (DiGA) have created a pathway for evidence-based digital solutions to become part of routine care.

At the same time, artificial intelligence is changing what these applications can do.

AI can personalise interactions, identify patterns across patient-reported data, support self-management and potentially detect changes in a patient's condition earlier. Generative AI can also make digital health applications more conversational and responsive.

But greater capability does not automatically mean better healthcare.

The important question is not simply how AI can be integrated into a DiGA, but how it can be integrated responsibly.

For AI-enabled DiGA, innovation needs to be measured against clinical value, patient safety, evidence, transparency and the realities of healthcare delivery.

From digital tools to intelligent support

Traditional digital health applications often operate through predefined pathways: a patient enters information, follows a programme and receives recommendations based on established rules.

AI creates the possibility of something more adaptive.

A system can potentially recognise patterns in longitudinal patient data, tailor information to individual needs, identify changes that may require attention and support patients between clinical appointments.

This is particularly relevant for chronic conditions, where health does not change only during a doctor's visit. Symptoms, behaviour, adherence and risk can evolve continuously.

AI can help make digital care more responsive to these changes.

But this also introduces a fundamental responsibility: the more adaptive the system becomes, the more carefully its behaviour must be designed, tested and monitored.

Responsible AI starts with the clinical purpose

The starting point should never be the technology itself.

A responsible DiGA does not use AI simply because AI is available. It starts with a clearly defined healthcare problem.

What should the system improve?

Does it help patients understand their condition? Does it support adherence or self-management? Can it identify relevant changes earlier? Can it reduce unnecessary burden for healthcare professionals? Does it contribute to measurable clinical outcomes?

These questions are more important than whether a solution uses a particular AI model.

The objective should be to use AI where it provides meaningful additional value — and not to introduce unnecessary complexity where a simpler, more predictable solution is better.

Evidence must remain at the centre

Healthcare innovation ultimately has to demonstrate that it works.

This is particularly important when AI influences how patients receive information, manage their health or interact with a digital therapeutic.

An impressive demonstration is not the same as clinical evidence.

Responsible AI in DiGA therefore requires systematic evaluation of performance, safety and real-world outcomes. It also requires understanding where a system performs well, where it is less reliable and under which conditions its recommendations should not be used.

For AI systems that evolve or adapt, this becomes an ongoing responsibility rather than a one-time validation exercise.

Personalisation without losing control

One of AI's most promising capabilities in digital health is personalisation.

A patient with a chronic cardiovascular condition may have a very different situation from another patient with the same diagnosis. Their symptoms, lifestyle, treatment plan, risk factors and engagement with care can vary significantly.

AI can potentially use this context to make digital support more relevant.

But personalisation should not mean giving an AI unrestricted authority.

A responsible system needs clearly defined boundaries around what it can recommend, what requires additional information and what should be escalated to a healthcare professional.

The principle is simple:

Personalised does not mean autonomous.

The role of AI should be to strengthen the patient's ability to manage their health and support clinical care — not to replace professional judgement.

Transparency is part of the product

Trust cannot be added after an AI-enabled DiGA has been built.

Patients and healthcare professionals should have a clear understanding of what the system is doing, what information it uses and where its limitations lie.

If AI generates a recommendation, users need appropriate context for interpreting it.

If the system is uncertain, that uncertainty should not be hidden.

If a situation falls outside the system's intended use, the user should not be given the impression that the AI can safely handle it anyway.

This is particularly important for generative AI, where fluent language can create an impression of authority even when an answer may be incomplete or incorrect.

Responsible design therefore includes mechanisms for verification, uncertainty management, escalation and human oversight.

AI should complement healthcare, not isolate the patient

A digital therapeutic should not become a digital silo.

The greatest value of AI-enabled DiGA may come when digital information can contribute to a broader care pathway.

For example, AI could help identify meaningful changes in a patient's condition and support appropriate escalation. It could help structure patient-generated information so that clinicians receive relevant signals rather than additional data noise.

This is where responsible innovation becomes closely connected to interoperability and workflow integration.

AI should not simply generate more information.

It should help make the right information available to the right person at the right time.

Responsible innovation means knowing when not to use AI

Perhaps the most important principle is also the simplest: not every problem requires AI.

If a deterministic rule is safer, easier to validate and clinically sufficient, there may be no reason to replace it with a generative model.

If a decision requires professional clinical judgement, AI should not create the illusion that an automated answer is an adequate substitute.

And if the available evidence is insufficient, the responsible choice may be to ask for more information, escalate to a healthcare professional or not make a recommendation at all.

The goal is not maximum automation.

The goal is maximum meaningful clinical value within clearly defined safety boundaries.

From innovation to responsible implementation

The future of AI-enabled DiGA will not be determined by who integrates the most advanced technology first.

It will be determined by who can demonstrate that AI delivers meaningful value safely and reliably in real healthcare.

That requires collaboration between technology developers, clinicians, researchers, patients, regulators and healthcare organisations.

For Chatdok, this means approaching AI in digital therapeutics as a clinical and healthcare challenge first, and a technology challenge second.

The opportunity is significant: AI can make digital care more personalised, continuous and responsive.

But responsible innovation requires discipline.

AI should not simply make DiGA more intelligent. It should make them more useful, more clinically meaningful and more capable of supporting better outcomes — while keeping safety, evidence and human responsibility at the centre.