Why Clinicians Adopt Digital Tools — and What This Meta-Analysis Can and Cannot Tell Us

One number in this paper is easy to miss and worth pausing on. Of all the relationships the authors pooled, the link from a clinician's stated intention to use a tool to their actual use of it is among the weakest: a path coefficient of 0.199. Intention barely predicts behaviour. That single figure sets the terms for how to read everything else here — including the headline findings that are genuinely useful.

The study in question is a systematic review and meta-analysis by Thanthrige and colleagues, published on 30 December 2025 in the Journal of Medical Internet Research. It pooled 50 studies covering 24,764 health professionals — work on electronic health records, telemedicine, mobile health apps and wearables across many countries — and combined two long-established frameworks: the Unified Theory of Acceptance and Use of Technology (UTAUT) and the Task-Technology Fit (TTF) model. The scale is the draw. What the numbers actually measure is the catch.

The design, and why it constrains the reading

Thirty-five of the studies used UTAUT (20,723 participants) and 15 used TTF (4,041). The authors extracted the reported path coefficients from each and pooled them. A path coefficient (β) here is a standardised measure of how strongly one factor — say, how useful clinicians expect a tool to be — predicts another, usually their stated intention to use it. Crucially, this is a meta-analysis of correlational survey data, not of trials. No one was randomised to anything. The studies recorded what people said about technology they had met, mostly at a single point in time, and this paper averaged those statements.

That design fixes two limits before any result is read. The outcome is almost always intention, not measured use — and as the 0.199 coefficient above shows, intention and use are loosely coupled at best. And because everything is measured together at one moment, the analysis can describe what travels with what, never what causes what.

What the evidence supports

Two findings are robust and worth carrying out of the room. Within UTAUT, Performance Expectancy — whether the tool is expected to improve performance — was the strongest predictor of usage intention (β = 0.304, p < .001), comfortably ahead of Effort Expectancy, or ease of use (β = 0.177), Social Influence (β = 0.167) and Facilitating Conditions such as support and infrastructure (β = 0.105 toward intention; 0.155 toward behaviour). Health professionals are task-driven: the dominant question is not whether a tool is modern or pleasant, but whether it does the job.

The TTF side sharpens this. The single strongest coefficient in the whole analysis was Technology Characteristics → Task-Technology Fit (β = 0.445, p < .001) — system reliability, speed and how deeply the tool integrates with existing work — ahead of Task Characteristics (β = 0.263), with the resulting fit then predicting intention (β = 0.271). In plain terms: a tool that is slow, flaky or bolted on beside the record system will be judged a poor fit for the clinical task however well it demos, and ease of use will not rescue it. None of this is surprising to anyone who has watched a launch fail. The value is that the pattern now carries pooled numbers rather than anecdote.

Social Influence is the one factor where the paper lets you see the spread behind the average, and it is instructive. Across the 33 studies that examined the link from social influence to intention, 64 percent (21 studies) found a positive, statistically significant effect, 24 percent (8) a positive but non-significant one, and 12 percent (4) a negative, non-significant one; none found a significant negative effect. So peer and managerial pressure tends to push the same way, but its force is modest and inconsistent — a pooled β of 0.167 that, study by study, often does not reach significance at all. Whatever moves clinicians, it is mostly not what their colleagues are doing.

What it does not support

Return to the opening number. A meta-analysis of cross-sectional acceptance surveys can tell you what correlates with people saying yes; it cannot tell you what sustains them in actually using a tool over months, and certainly not whether use changed anything for a patient. The tool nobody opens after week three is a story about behaviour over time, and this design cannot see that arc. The weak intention-to-behaviour link is the quantitative trace of exactly that gap.

The point-in-time design also rules out causation. A strong coefficient for technical quality does not prove that improving reliability will lift real-world usage — only that the two travel together in the survey data. And the studies disagree sharply with one another: heterogeneity ran from an I² of 81.9 percent to 94.87 percent, meaning most of the variation between studies is genuine difference, not chance, which the authors attribute to regional, cultural and technological differences. A pooled average across such heterogeneous settings is a useful orientation, not a transportable coefficient you can apply to your own ward.

The authors are candid about the rest. The review was limited to English-language publications in selected databases; it deliberately analysed only the core UTAUT and TTF constructs and so left out trust, perceived risk, security and organisational culture — plausibly among the most important factors in a hospital. It excluded qualitative work, performed no formal risk-of-bias assessment with an established tool such as Joanna Briggs or GRADE, and did not test for publication bias with funnel plots or an Egger test. None of this sinks the paper. It means the paper maps the well-lit part of the question. (The authors declare no external funding and no competing interests.)

Why it matters

For anyone choosing or building clinical software in a European system, the defensible reading is narrow and useful. Perceived usefulness and genuine technical quality — reliability, speed, integration depth into the record system — are the factors most consistently associated with clinicians' willingness to adopt, more so than surface ease of use or social pressure, and a tool weak on those will struggle whatever the training budget. That is worth knowing before procurement. What this evidence cannot do is promise that getting those factors right will produce sustained use, let alone better care; those remain questions for prospective evaluation, measured in behaviour and outcomes rather than in survey intentions.

Source: Thanthrige A, Lu B, Sako Z, Wickramasinghe N. Determinants of Health Care Technology Adoption Using an Integrated Unified Theory of Acceptance and Use of Technology and Task Technology Fit Model: Systematic Review and Meta-Analysis. J Med Internet Res 2025;27:e64524. A meta-analysis of 50 cross-sectional, correlational studies whose outcome is largely stated intention to use rather than measured use or patient outcomes, with high between-study heterogeneity and no formal bias assessment.