JMIR Publications

ABSTRACT

Background:

Mobile health (mHealth) interventions are increasingly used to support personalized self-management in type 2 diabetes (T2D), particularly for promoting physical activity. Mobile apps enable self-monitoring of various indicators (eg, blood glucose, physical activity), supporting the delivery of personalized feedback and guidance. Such guidance may be provided either by health professionals (human-delivered) or automatically generated by algorithms (automated). While both approaches are promising, evidence comparing human with automated guidance remains limited. Addressing this gap is important for understanding how different approaches to personalized guidance may influence intervention effectiveness as digital health technologies are increasingly integrated into T2D care.

Objective:

To compare the effectiveness of human-delivered and automated personalized guidance in mHealth-based physical activity interventions for adults with T2D, and to examine physical activity outcomes, intervention engagement, and intervention characteristics.

Methods:

We conducted a systematic review in accordance with PRISMA guidelines. We searched four databases (MEDLINE, PsycINFO, CENTRAL, and Scopus) and two trial registries (ClinicalTrials.gov and WHO ICTRP) up to April 2026. We included randomized controlled trials (RCTs) evaluating smartphone-accessible physical activity interventions in adults with T2D that used either human-delivered or automated personalized guidance. We pooled HbA1c, the primary outcome, using random-effects meta-analysis with a generic inverse-variance method. Physical activity outcomes and intervention engagement were narratively synthesized because meta-analysis was not feasible.

Results:

Twenty-two RCTs (Human-delivered guidance: 10, n = 1,369; Automated guidance: 12, n = 3,135) were included. Human-delivered guidance showed a small pooled reduction in HbA1c compared with control conditions (MD -0.14%, 95% CI -0.29 to -0.00, I2=23%), which was attenuated and no longer statistically significant after excluding studies at high risk of bias. Automated guidance showed a larger pooled reduction (MD -0.31%, 95% CI -0.48 to -0.13, I2=79%) that remained statistically significant in sensitivity analysis. The certainty of evidence was low for both human-delivered and automated interventions. Evidence for physical activity outcomes remained inconclusive, with mixed findings across studies.

Conclusions:

Human-delivered guidance showed a smaller and less robust HbA1c effect, whereas automated guidance showed modest and more consistent improvements. The available evidence did not indicated that human-delivered guidance consistently conferred greater glycemic benefit than automated guidance. Guidance source should therefore be considered alongside delivery strategy and broader intervention content. Clinical Trial: PROSPERO: CRD420251039711