What makes Internet‐based interventions work?

Numerous studies, including controlled trials and ordinary clinical investigations, document that clinician‐supported psychological treatments can be delivered online in a course format, not requiring scheduled real‐time therapy sessions ¹ . In fact, available research suggests that guided Internet‐based psychological treatments can be as effective as face‐to‐face psychotherapies ² . Not only Internet‐delivered cognitive behavior therapy has been found to work, but also other psychotherapies, such as psychodynamic interventions ³ .

Different technologies have been used and are currently being developed – as reviewed by Torous et al in this issue of the journal ⁴ – but their added value against text and video content is still uncertain. In fact, it is even not obvious how much technology adds in relation to providing treatment via text in a book format, as old‐school therapist‐supported bibliotherapy often has about the same effects as digital treatments, although with much smaller studies and less implementation ¹ . Where more recent advancements will lead us, when artificial intelligence (AI) and digital phenotyping become increasingly used in research ⁴ , remains an open question.

The way digital treatments should be presented is far from certain. First, it is tempting to shorten treatments, sometimes even presenting them in a single session format ⁵ . While such condensing of information may be useful for preventive purposes, it needs to be acknowledged that very brief treatments may lead to short processing time. From cognitive psychology we know that deep learning is less likely to occur if material is just briefly processed at a surface level.

Second, much effort has been placed on making treatments more attractive and persuasive. This can pay off, if it does not come at the expense of missing useful change strategies and intervention components. Game features can be added, but should not distract. Instead, educational aspects can make a difference: by focusing on strengthening learning of treatment content, both interactive features and gamification can be helpful ⁶ .

Third, a long‐standing discussion in the field of digital treatments concerns the role of clinicians. This discussion involves not only guided treatments often being superior to purely self‐guided interventions, but also patient preferences. Now we have the possibility of AI‐therapists ⁴ : this is currently being investigated, as it is over and beyond what has previously been done with automated conversational agents, but challenges remain ⁴ . The question of patient preferences is relevant again in this respect, as it may be that at least a proportion of users prefer a “real person” and are sensitive to any cues suggesting that the AI‐therapist is fake. On the other hand, with informed consent and well‐trained models, it is also likely that, at least for some patients and conditions (for example insomnia), this will be a useful addition to our clinical services.

But what works for whom? This is the eternal question in all clinical and research settings, also from the user perspective, as patients make decisions informed by what they hear from their networks and also by research (albeit often indirectly through the press). Numerous studies on digital treatments have explored predictors of outcome. As with psychotherapies in general, there is limited evidence for most candidates (including also moderators and mediators of outcome). Biological markers have also been studied, but the relevant state of knowledge is far from being sufficiently robust ¹ . A systematic review focusing on Internet interventions included 80 studies and a total of 88 predictors ⁷ . The authors reported that better adherence, credibility of the treatment, and patient‐rated therapeutic working alliance did predict the outcome, which is pretty much in line with research on face‐to‐face psychotherapies. They also reported the fairly obvious correlation between high pre‐treatment scores and more change, which is a natural consequence of data structure and not much informative for treatment planning.

Having been involved in many controlled trials, it is pretty obvious for me why we fail to find predictors. The reason is that we remove them from the studies by using inclusion and exclusion criteria. While the situation is somewhat better in regular clinical settings, the problem is still there, because of referral patterns and intake procedures which for very good reasons result in unsuitable patients not entering therapy. Added to that we have the self‐selection bias, with more difficult patients not seeking treatment and not accepting the offer to receive treatment. Yet another problem is dropout ⁸ : while missing data can be handled by statistical models, they still make prediction models less stable, as it is the post‐treatment outcome that we want to predict.

Despite this knowledge gap when it comes to prediction, the development and implementation of Internet‐based cognitive behavior therapy is a success story ⁹ . It may for some readers be a surprising fact that we now have more evidence for Internet treatments than for regular face‐to‐face therapy for many conditions and disorders ¹ , at least when it comes to study size and quality of data. Yet, implementation lags behind, as it is still rather rare that Internet treatments are provided on a regular basis in clinical service settings worldwide. When it comes to other digital alternatives, such as mobile phone apps and virtual reality, they are often not evidence‐based and seldom implemented in clinical settings. Use of technology in a blended format is fairly common when it comes to administration of services, but still rather rare in the form of combined technology and face‐to‐face treatments.

With a 25 year history of Internet‐based cognitive behavior therapy, there are lessons learned that can help researchers and clinicians to achieve better outcomes. The top tips for successful Internet interventions are the following:

• Start with a proper full assessment. If a patient cannot complete the full pre‐treatment questionnaires, he/she is very unlikely to be able to engage in and complete the treatment.

• Move fast. If a patient applies for treatment online, schedule him/her for interview, check him/her for suitability, and inform him/her about inclusion and start of treatment as soon as possible.

• Have a clear deadline for post‐treatment assessment regardless of treatment completion and adherence. Schedule a phone meeting that will serve as a deadline.

• Technology should be user friendly and not crash. A good system is crucial and should work for different platforms (e.g., computer and phone).

In the future, more work will have to be done on global mental health and delivery of interventions in non‐Western languages. AI will have to be fully investigated. Understudied and marginalized groups will have to come to focus. And, crucially, the treatment‐demand gap will have to be reduced.