**Background:** Childhood obesity is a major public health problem with rising prevalence of severe obesity and associated comorbidities. Family-based lifestyle interventions are first-line treatment, but digital interventions offer scalable alternatives. The ENDORSE platform is an AI-based ecosystem designed to support weight management in children aged 6–14 years through mobile apps, activity trackers, serious games, and personalized recommendations. This study evaluates the impact of the ENDORSE program on metabolic parameters and food parenting practices.
**Methods:** The ENDORSE pre-post intervention study included 50 mother-child dyads recruited from an obesity clinic in Athens, Greece, from March 2021 to May 2022. Children aged 6–14 years with BMI >85th centile were eligible. Exclusion criteria included secondary causes of obesity, genetic syndromes, and severe neurodevelopmental or psychiatric disorders. The intervention lasted 12 weeks and consisted of the ENDORSE platform components: a physical activity tracker, mobile applications for parents and health professionals, a mobile serious game, and an AI-based recommendation system. The study had three phases: a pre-pilot group (n=20), a pilot active control group (n=15, without the game and recommendation system), and a pilot intervention group (n=15, full platform). The present analysis focuses on the 45 children (90% of total) who completed the post-intervention assessment. Children were categorized into three groups based on BMI z-score change: group 1 (clinically significant decrease ≥0.2, n=21), group 2 (decrease/stable 0.00–0.20, n=16), and group 3 (increase >0.00–0.25, n=8). Clinical assessments included anthropometrics, blood pressure, and pubertal staging. Fasting blood samples were analyzed for glucose, insulin, HbA1c, lipids, liver enzymes (SGOT, SGPT), cortisol, and ACTH. HOMA-IR was calculated. Food parenting practices were assessed using the Comprehensive Feeding Practices Questionnaire (Greek version) completed by mothers. Statistical analysis included paired t-tests, Wilcoxon tests, and linear regression with covariates (sex, pubertal stage, age, baseline BMI z-score for metabolic outcomes; maternal education and marital status for feeding practices).
**Key Results:** Overall, there was a statistically significant reduction in glycated hemoglobin (mean change = −0.10, p = 0.013), SGOT (mean change = −1.84, p = 0.011), and SGPT (mean change = −2.95, p = 0.022). Between groups, significant differences were observed for HbA1c (greater reduction in group 2, p=0.026), total cholesterol (decrease in groups 1 and 2, increase in group 3, p=0.009), LDL-cholesterol (decrease in group 1, increase in groups 2 and 3, p=0.003), and cortisol (reduction in group 1, increase in groups 2 and 3, p=0.037). Linear regression analysis revealed that BMI z-score change had a robust and significant correlation with HOMA-IR change (beta coefficient = 3.60, p-value = 0.046), SGPT change (beta coefficient = 11.90, p-value = 0.037), and cortisol change (beta coefficient = 9.96, p-value = 0.008) after adjusting for sex, pubertal stage, age, and baseline BMI z-score. Regarding food parenting practices, emotional feeding/food as reward decreased overall (mean change −0.21, p = 0.007) and healthy eating guidance increased (mean change = 0.11, p = 0.051). Between groups, healthy eating guidance increased in groups 1 and 2 but decreased in group 3 (p=0.013), and child control decreased in groups 1 and 2 but increased in group 3 (p=0.037). Multivariate regression showed that healthy eating guidance change had a significant negative relationship with BMI z-score change (beta coefficient = −0.29, p-value = 0.007) after adjusting for maternal education and marital status.
**Clinical Implications:** The ENDORSE digital weight management program improved several metabolic parameters, including HbA1c, liver enzymes, and insulin resistance, and promoted healthier food parenting practices such as reduced emotional feeding and increased healthy eating guidance. These findings support the potential of digital interventions to address pediatric obesity and its comorbidities, especially in settings where access to in-person programs is limited. The robust association between BMI z-score reduction and improvements in HOMA-IR, SGPT, and cortisol highlights the metabolic benefits of even modest weight loss. The improvement in healthy eating guidance and its negative correlation with BMI z-score change underscores the importance of targeting parental feeding practices in digital interventions. Limitations include the small sample size, lack of a true control group, high proportion of children with severe obesity, and absence of long-term follow-up. Future large-scale randomized controlled trials are needed to confirm these findings and assess sustainability.