JMIR Publications

ABSTRACT

Background:

Type 1 diabetes mellitus (T1DM) in children requires sustained self-management to achieve glycemic targets and prevent long-term complications. Continuous glucose monitoring (CGM) has transformed pediatric diabetes care, yet adherence to device wear remains inconsistent. In May 2024, the Sultanate of Oman launched a national initiative that distributed CGMs to children with T1DM across all governorates, creating an unprecedented real-world opportunity to study adherence determinants and to develop a culturally adapted, locally validated artificial intelligence (AI)-assisted predictive tool.

Objective:

This is a multi-phase translational research project that aims to (1) characterize the population of Omani children with T1DM who received CGM; (2) identify demographic, psychosocial, dietary, and physical activity correlates of optimal CGM use; (3) develop, train, and validate an AI-assisted behavioral predictive tool “OMNIdiasense” to forecast CGM adherence prior to device dispensing; and (4) pilot test the OMNIdiasense tool.

Methods:

Three sequential, interlinked sub-studies will be conducted. Sub-study 1 is a retrospective cohort analysis of routinely collected Al Shifa data for all children who received CGMs between July 2024 and February 2025, with glycemic, anthropometric, and laboratory outcomes compared at baseline and at ≥3 months. Outputs on adherence prevalence, clinical predictors, and the de-identified analytic dataset becomes the structured input layer for the AI model in sub-study 3. Sub-study 2 is a cross-sectional, mixed methods study using face-to-face structured interviews with a randomly selected sample of children aged 10-18 years drawn from sub-study 1, classified as “CGM Optimizers” (≥6 days/week) or “CGM Sub-users” (<6 days/week or discontinued); responses across validated behavioural, stress, and dietary instruments are compared. Outputs are the psychosocial and behavioral feature set, qualitative themes, and effect sizes that drive feature selection for the AI model. Sub-study 3 develops the AI tool (OMNIdiasense), comprising (a) a quasi-experimental single-arm pilot among 100 existing CGM sub-users and (b) a parallel pilot randomized controlled trial (n=50; 25 intervention, 25 control) among newly diagnosed children, with assessments at baseline, 3, 6, and 12 months. The primary outcome of sub-study 3 is between-group difference in CGM adherence; secondary outcomes include HbA1c, anthropometric, cardiovascular, and laboratory measures, and barriers to use. Reporting will follow SPIRIT 2025 (interventional component), STROBE (observational component), the CONSORT 2025 extension for pilot and feasibility trials, COREQ (qualitative component), and TRIPOD+AI (predictive model).

Results:

The proposed AI tool is intended as a decision-support adjunct and not a gatekeeping mechanism for CGM access.

Conclusions:

This translational program, to our knowledge, is the first in the Gulf Cooperation Council region to combine retrospective real-world CGM data, validated psychosocial measurement, and locally developed AI to predict pediatric CGM adherence prior to device dispensing. By targeting behaviorally vulnerable patients before sensor distribution, OMNIdiasense is expected to support clinical benefit and reduce financial waste. Clinical Trial: ISRCTN15827616