Artificial Intelligence/Machine Learning-Enabled Precision Nutrition Modestly Improves Anthropometric Outcomes and Systolic Blood Pressure: A Systematic Review and Meta-analysis.
Janet Antwi, Obed Akwaa Harrison, Niki Hayatbini
✦ AI-curated · Sources linked
In 30 seconds
This PRISMA-guided systematic review and meta-analysis assessed AI/ML-enabled and other technology-supported precision nutrition across the lifespan. It included 262 studies published from January 2015 to December 2025, combining meta-analysis for comparable outcomes with qualitative synthesis for others. Pooled estimates showed modest reductions in body weight (-1.97 kg), BMI (-0.57 kg/m 2 ), waist circumference (-2.41 cm), and systolic blood pressure (-4.05 mmHg). No significant pooled effects were found for lipid outcomes or glycemic markers. The authors judged the evidence generally low to very low certainty, with substantial heterogeneity and prediction intervals crossing the null for several significant outcomes.
Key findings
- Pooled body weight reduction was -1.97 kg (95% CI, -3.02 to -0.92; 95% prediction interval, -6.69 to 2.75 kg).
- Pooled reductions were -0.57 kg/m 2 for BMI and -2.41 cm for waist circumference; prediction intervals crossed the null for both.
- Systolic blood pressure decreased by -4.05 mmHg (95% CI, -6.90 to -1.19; 95% prediction interval, -13.24 to 5.15 mmHg).
- No significant pooled effects were observed for lipid outcomes or glycemic markers; dietary intake findings were generally small and inconsistent.
Why it matters
Technology-supported personalization is being used to tailor dietary guidance, and this synthesis suggests possible modest benefits for weight-related measures and systolic blood pressure. However, the broad uncertainty and lack of pooled benefit for lipid and glycemic outcomes mean these findings do not establish consistent clinical impact. The review also highlights the need to assess implementation and equity alongside effectiveness.
What to keep in mind
Certainty was generally low to very low because of risk-of-bias concerns, imprecision, and substantial heterogeneity. Prediction intervals crossed the null for several significant pooled effects, and subgroup analyses were exploratory and underpowered.