Summary: In a retrospective secondary analysis of the intervention arm of a randomised controlled trial, app-usage data from 328 pregnant women at risk of gestational diabetes using a coaching app were examined with unsupervised clustering. Three engagement patterns were identified (Averagers, Goalers, Immersers), with 82 percent classified as Averagers. No clinical outcomes or comparison with standard care were assessed in this analysis.
PICO Summary
| Element | Detail |
|---|---|
| Population | 328 pregnant women at risk of gestational diabetes, drawn from the intervention arm of a randomised controlled trial; retrospective secondary analysis of app-usage data (multi-site, Denmark-based study). |
| Intervention | Digital lifestyle coaching delivered through a mobile health app (the Liva app), with coach features and goal-setting features; n=328 (intervention arm only, first study phase). |
| Comparison | No control or standard-care comparator was analysed. The contrast was between data-driven engagement clusters identified within the single intervention group. |
| Outcome | Principal component analysis reduced engagement data to two components; a Gaussian mixture model clustered participants into 3 patterns: Averagers (82 percent; moderate use of both coach and goal features), Goalers (predominantly goal features) and Immersers (high use of both). No effect estimates, 95 percent confidence intervals, p values or absolute risk reductions were reported, as this was a descriptive, exploratory clustering analysis without significance testing of clinical outcomes. |
Expert Commentary
This work should be read as an exploratory, hypothesis-generating description of how pregnant women interact with a coaching app, and not as evidence that the app improves any health outcome. Because the analysis was restricted to the intervention arm of a trial and used unsupervised clustering, no comparison with standard care was possible and no statistical significance was tested. The headline finding is that engagement is heterogeneous even under identical program conditions, with most participants (82 percent) being moderate, balanced users and smaller groups gravitating toward either goal-setting or fuller immersion. That observation is plausible and is reinforced by the large within-arm sample, although clusters derived from principal component analysis and a Gaussian mixture model are sensitive to the variables chosen and may not replicate elsewhere; that is the single most important limitation here. The clustering was data-driven rather than linked to glycaemic or pregnancy outcomes, so the clinical meaning of being an Averager versus an Immerser remains unknown. Can I use this with my patients? Not yet, beyond setting expectations that digital coaching uptake will vary and that some women may need more proactive outreach to stay engaged. Future work should connect these engagement phenotypes to maternal and neonatal outcomes and validate them across diverse settings before tailoring is recommended.
References
Bendsen SB, Skinner TC, O’Reilly SL, Rey Velasco E, Heltberg MS, Laursen DH. Exploring engagement patterns within a mobile health intervention for women at risk of gestational diabetes. Womens Health (Lond). 2025;21:17455057251327510. doi:10.1177/17455057251327510
