Digital health for health outcomes: an umbrella review of meta-analyses of randomised controlled trials

Abstract Background Digital health interventions offer unique strengths including wide accessibility, scalability, lower workforce dependence, and potential cost savings. Yet evidence on their effectiveness remains inconsistent, and existing evidence syntheses are fragmented, limiting clinical and policy guidance. This umbrella review aimed to address this gap. Methods We searched six databases from inception to April 2025 for meta-analyses of randomised controlled trials comparing digital health interventions with either inactive controls (eg, usual care, attention placebo or waitlist) or standard therapies (eg, face-to-face treatment). Methodological quality, evidence certainty, and credibility were assessed using AMSTAR 2, GRADE, and evidence classification criteria, respectively. Random-effects reanalyses used HKSJ correction, with effect sizes standardised so positive values indicate benefit. Results We included 232 publications encompassing 1133 unique outcomes across 12 condition categories and 559792 participants. Methodological quality was high in 37%, moderate in 12%, low in 32%, and critically low in 19%. Overall, 53 findings had high-certainty and 84 moderate-certainty evidence. Against inactive controls, telemonitoring reduced systolic blood pressure in hypertension (equivalent Hedges’ g 0.55, 95% CI 0.31–0.78). Other high-certainty findings showed telephone-based interventions reduced diastolic blood pressure (0.17, 0.09–0.24), Internet-based cognitive behavioural therapy reduced depression in insomnia (0.38, 0.23–0.54), and telemedicine reduced weight in type 2 diabetes (0.12, 0.06–0.17). Against standard therapies, specific virtual reality improved upper limb function after stroke (0.23, 0.06–0.40), whereas telemonitoring showed no statistically significant differences in hospitalisation, all-cause mortality, or stroke among patients with pacemakers; synchronous virtual consultations showed no difference in HbA1c in type 2 diabetes; and tele-exercise showed no difference in physical function in knee osteoarthritis. All had high-certainty evidence. Additional moderate-certainty findings concerned stroke, Parkinson’s disease, and musculoskeletal conditions. Surgical, maternal-neonatal, and older adult care outcomes included high and moderate certainty findings. Conclusions Digital health interventions are effective but show no superiority over standard therapies in cardiovascular disease, type 2 diabetes, stroke and Parkinson’s disease, mental health, musculoskeletal conditions, postoperative care, maternal and neonatal health, and care for older adults, though effects vary by outcome. Future work should prioritise standardisation, optimisation and inclusive implementation to realise the equity and resilience potential of digital health. Trial registration PROSPERO, CRD42025645564.