Health apps move blood pressure and glucose, not mortality

Non-communicable diseases are the leading cause of death globally, and health systems everywhere are looking for scalable, low-cost ways to shift the behaviours behind them. Smartphone apps are the obvious candidate: nearly everyone carries one, and thousands of products promise better glucose control, weight loss or calmer minds. The evidence base, though, is scattered across hundreds of individual trials and dozens of overlapping reviews of wildly varying quality. Someone needed to gather all of it in one place and ask, outcome by outcome, where apps actually move the needle and where the promise outruns the data.

The researchers ran an umbrella review with meta-analysis, searching eight databases for systematic reviews with meta-analysis published between January 2013 and January 2024, with no restriction by location or language. Eligible reviews covered randomised or controlled trials of app-based interventions in adults aged 18 or over with, or at risk of, non-communicable diseases. From 6951 unique records, 383 went to full-text review and 78 systematic reviews were included, spanning 31 outcomes, 496 primary studies and 177 373 participants. Effects were pooled as standardised mean differences using random-effects models, with review quality judged by AMSTAR 2, certainty by GRADE and the Ioannidis criteria, and heterogeneity by I².

Apps produced significant improvements across a wide spread of outcomes. The largest effects were for glucose measures: 2 h postprandial glucose, fasting blood glucose and glycated haemoglobin all fell, alongside systolic and diastolic blood pressure. Bodyweight, BMI, waist circumference and sedentary time all decreased, while diet quality, exercise capacity, moderate-to-vigorous physical activity, daily steps, mindfulness, wellbeing, quality of life and medication adherence all improved. Anxiety, depression and stress showed smaller but significant reductions. Crucially, no significant effect appeared for cardiovascular mortality, HDL, LDL, total cholesterol or triglycerides, body fat, distress, fruit and vegetable intake, hospitalisation or smoking abstinence. Only five outcomes reached highly suggestive evidence, with seven categorised as non-significant.

This is the most comprehensive synthesis to date of a field previously assessed one condition at a time, and it broadly confirms that digital behaviour-change tools work for intermediate risk markers. The pattern is instructive: apps shift the things that respond to daily self-monitoring and prompting — glucose, blood pressure, steps, adherence — but not hard endpoints or blood lipids. The overriding limitation is that the underlying reviews were mostly poor: just one was rated high quality against 57 rated critically low, and heterogeneity ran above 90% for many outcomes. Whether the improvements persist after the novelty fades, and whether they translate into fewer events, remains unanswered.

This is a reason to think a well-designed health app is a genuine tool rather than a gimmick, particularly if the target is glucose control, blood pressure, activity levels or sticking with medication. The honest framing is that apps appear good at nudging measurable daily behaviours and the numbers closely tied to them. They did not show an effect on cholesterol, body fat, hospitalisation, giving up smoking or dying of cardiovascular disease, so expectations should be calibrated accordingly. The very low quality of most underlying reviews also means the effect sizes here should be read as directionally encouraging rather than precise.