Twenty-five years of digital health trials, mapped by strategy

In the early 2000s, a digital health intervention usually meant a text message reminder, a static website with health advice or a telephone follow-up. Today the same label covers smartphone apps that adapt to their user, wearables that stream data continuously and conversational assistants. Most of these tools have been tested at some point in randomised trials, and that body of evidence is now large enough to be studied as a whole.

A review published in the Journal of Medical Internet Research in early 2025 did that. The authors searched PubMed, Web of Science and Scopus for randomised controlled trials of digital health interventions published in English between January 1999 and March 2024. From 4,122 candidate papers, they retained 885 trials involving 954,847 participants in total, and coded every technique the interventions used to change behaviour or improve health. They identified 173 distinct strategies, which they grouped into 19 themes. A Tech Anatomy episode discussed the review shortly after it appeared.

What the review counted

Across the whole period, three themes dominated. "Guide", meaning structured advice, instructions and step-by-step support, appeared in 55.6 percent of trials. "Monitor", the tracking of indicators such as blood glucose, activity or symptoms, appeared in 55.4 percent. "Communication", such as messages and contact with professionals, appeared in 44.3 percent. Most interventions combined several strategies.

Almost all of the interventions, 95.4 percent, were designed for patients or the public, rather than for health professionals or health system managers. Studies aimed at the general population outnumbered those focused on vulnerable groups by about three to one: 26.9 percent of the trials targeted groups such as children, women, older adults or LGBT people. Studies from high-income countries showed a larger number and a wider diversity of strategies than those from low- and middle-income countries.

The 2017 cut-off

To look at change over time, the authors split the trials at 2017. They chose that year because of policy milestones: the FDA launched its Digital Health Innovation Action Plan in 2017, and the World Health Organization began developing its first guideline on digital health interventions, published in 2019. Before 2017, monitoring was the most frequent strategy, used in 51.4 percent of trials. From 2017 onward, guidance led, at 60.8 percent, followed by monitoring and then "engagement", which entered the top three.

Two points about this split are easy to miss. The cut-off was set by the authors in advance, so the review does not show that something changed in 2017 specifically. And the shift is in what trial designers chose to test. The episode attributed it to advances in AI and real-time analytics enabling more personalised guidance, which is a plausible explanation, but the review does not test it.

Where the episode went beyond the data

The episode presented personalisation, behavioural science and peer support as what makes digital interventions effective, and cited cryptocurrency rewards for exercise as an example. Those points come from the broader literature and from the episode's own reading, not from this review, which counted how often strategies were used without measuring how well each one worked. The cryptocurrency example does not appear in it and has been left out here.

The same caution applies to the review's own conclusions. Its authors describe an evolution from simple approaches toward personalised, multi-component interventions with real-time feedback. That describes the design of tested interventions. Whether the more sophisticated designs perform better is a separate question that requires comparing effect sizes, which the review did not do.

Two readings of the same map

One reading sees a field that has matured. Interventions combine more techniques, draw on behaviour change science more explicitly and are tested in randomised trials, which is the standard of evidence medicine asks for. The volume of trials alone shows that digital health is no longer a fringe activity.

A second reading focuses on the gaps. Three quarters of the trials concerned the general population, most came from wealthy countries, and few targeted health professionals or health systems, although many of the barriers to adoption sit there. Research effort, on this view, has followed what is easiest to build and test rather than where needs are greatest.

What the review does not show

The review includes only English-language papers from three databases, which its authors acknowledge may underrepresent studies from low- and middle-income countries and grey literature. Coding depended on how each paper described its intervention, which varied in detail. And because it maps strategies rather than outcomes, it cannot say which strategies work, for whom or for how long. The authors call for longer-term effectiveness studies, and for more attention to vulnerable groups.

What to watch

The natural next step is analyses that link these strategy categories to measured effects, ideally with long follow-up. Two other indicators would show whether the gaps identified here are narrowing: the share of trials run in low- and middle-income countries and in vulnerable populations, and the number of interventions designed for clinicians and health systems rather than only for patients.

Listen to the episode

25 Years of Digital Health: What Works, What Doesn’t?

Listen · 8 min