Effects of a personalized nutrition program on cardiometabolic health: a randomized controlled trial

The study addresses whether a personalized dietary program can improve cardiometabolic health compared with generalized dietary guidance. Despite strong evidence that diet and lifestyle affect chronic disease risk, adherence to population-level dietary guidelines is poor, and there is substantial inter- and intra-individual variability in metabolic responses to foods. This variability suggests a single dietary approach may not suit everyone. Personalized nutrition that integrates biological, phenotypic, lifestyle and microbiome data could tailor diets to individual responses and potentially enhance adherence and outcomes. The trial tests an app-based personalized program incorporating postprandial glucose and triglyceride responses, microbiome profiles and health history against standard USDA dietary advice in generally healthy middle-aged and older adults in the US.

The authors reference prior work demonstrating large variability in postprandial responses and the potential for precision nutrition, as well as low adherence to population dietary guidelines in the US and UK. Several personalized nutrition trials have shown improved dietary behaviors but mixed effects on weight and cardiometabolic markers: Food4Me showed improved diet quality without weight differences; Ben-Yacov et al. reported comparable weight loss between a personalized postprandial-targeting diet and a Mediterranean diet; the Personal Diet Study and PREVENTOMICS did not demonstrate additional benefits for weight or fat mass with personalization over controls. The study builds on machine-learning approaches linking diet, microbiome and metabolic responses, and on evidence for microbiome–diet interactions in cardiometabolic health.

Design: The ZOE METHOD study was an 18-week, parallel-group, randomized controlled trial (remote, US). ClinicalTrials.gov: NCT05273268. Ethical approval: Advarra IRB (IRB no. 00000971; protocol no. 00044316). Participants provided written informed consent. Participants: Adults aged 40–70 years living in the US, reflective of the average US population, with waist circumference above ethnicity- and sex-specific 25th percentile, and fruit/vegetable intake <450 g/day. Key exclusions: prior participation in ZOE/PREDICT studies; incompatible devices; lipid-lowering or antidiabetic medications (unless safely discontinued before and during the study); certain inflammatory or gastrointestinal diseases; recent antibiotics/immunosuppressants; acute psychiatric disorders; recent MI/stroke; pregnancy/lactation; vegan diet; eating disorders; food restrictions incompatible with study; adhesives allergy precluding CGM. Recruitment and randomization: Recruited Mar–Aug 2022 via email/newsletters. After online screening and a baseline clinic visit (week −1), eligible participants were randomized using minimization (MinimPy) to PDP or control, balanced on sex, waist circumference (above/below ethnicity-specific median) and fruit/vegetable intake (above/below US median). Trained coordinators informed participants by email. Interventions: Control received standard USDA Dietary Guidelines for Americans (2020–2025) via digital leaflet, short video, online resources and weekly email check-ins from a nutrition coach for weeks 2–18. PDP participants completed the ZOE test kit (CGM, standardized muffins for mixed-meal tests, and a dried blood spot for postprandial TG/HDL-C/total cholesterol) and logged diet via the ZOE app. They received 4 weeks of generalized nutrition/lifestyle lessons (weeks 2–6) followed at week 6 by personalized ‘Insights’ including blood sugar and blood fat scores, gut diversity and microbiome scores, and personalized food scores generated by the ZOE 2022 algorithm. From week 6, PDP participants engaged in a 12-week action plan with app lessons and optional phone/video walkthrough of results; app-based coach support was available. The PDP did not prescribe calorie restriction. Personalization algorithm: Food quality scores (0–100) combined macronutrient content and metadata (glycemic load, fat quality, processing level, food group) and were personalized to each participant’s glucose control, postprandial TG response, ASCVD risk, health history and microbiome composition. Personalized meal/day scores guided adherence targets. Assessments and timeline: Baseline 1 (week −1): fasting venous blood, blood pressure, anthropometrics (height, weight, waist/hip), questionnaires, stool sample. Baseline 2 (week 0): fasting bloods, questionnaires, stool; FFQ (PREDICT FFQ with 264 items). Interim (week 12): fasting bloods, questionnaires, stool, FFQ. Endpoint (week 18/19th week): fasting bloods, questionnaires, stool, FFQ; PDP repeated ZOE test kit (CGM, test meals, DBS). PDP additional follow-ups at 8 and 12 months; controls offered cross-over or commercial product after endpoint. Dietary assessment: PREDICT FFQ at weeks 0, 12, and 18; exclusions for incomplete items or implausible energy intake relative to BMR. Energy density calculated as kcal/g of foods (excluding beverages). Microbiome: Stool collected at home (DNA/RNA Shield tubes), shipped to lab. Shotgun metagenomics (NovaSeq 6000), preprocessing with quality control and human read removal, average 35±13 million reads/sample. Taxonomic profiling with MetaPhlAn 3.0 and 4.0. Alpha diversity (richness, Shannon) and beta-diversity (Bray–Curtis). Machine learning (random forest, 100-fold CV) linked species-level changes to health outcomes. CGM: Freestyle Libre 14-day; data used from 12 h post-activation; participants unblinded. Outcomes: Primary—18-week changes from baseline in fasting serum triglycerides (TG) and direct LDL-C. Secondary—changes in weight, waist and hip circumference, systolic/diastolic blood pressure, HbA1c, fasting insulin, glucose, C-peptide, apolipoprotein A1 and B, gut microbiome (richness, Shannon, Bray–Curtis), postprandial TG (DBS), diet quality (HEI), self-reported energy level; other labs (e.g., liver enzymes, CRP, TNF-α, FBC) and self-reported mood and hunger. Adherence: Both groups self-reported adherence (0–10 scale) every ~6 weeks. PDP adherence additionally evaluated via logging metrics and personalized day scores; minimum logging of 4 days/month including a weekend day and caloric thresholds. Sample size: Powered for n=150/group (total n=300) at 90% power, alpha<0.05 to detect between-group differences of 0.21 mmol/L in TG and 0.30 mmol/L in LDL-C (with assumed SDs 0.55 and 0.8 mmol/L, respectively). With two primary outcomes, significance threshold P<0.025. Statistical analysis: Intention-to-treat (ITT; n=347) and per-protocol (PP; n=225; endpoint 18±2 weeks). Baseline defined as average of two clinical baseline measures. Repeated-measures models estimated diet group×time interaction adjusted for age and sex, with participant random effect; non-normal outcomes log-transformed. Between-group differences reported with 95% CIs; primary outcomes tested at P<0.025. Microbiome longitudinal within-individual changes tested by paired, one-sided Wilcoxon; across-group differences by Kolmogorov–Smirnov parameter. Favorable/unfavorable species tested by Mann–Whitney–Wilcoxon; fold changes reported. Subgroup analyses contrasted high vs low adherence within and across arms.

Participants: n=347 randomized (PDP n=177; control n=170); 86% female; mean age 52±7.5 years; BMI 34±5.8 kg/m²; fasting glucose 5.32 mmol/L; total cholesterol 5.41 mmol/L; TG 1.35 mmol/L; LDL-C 3.38 mmol/L. ITT included all 347; PP n=225 (PDP 108; control 117). Primary outcomes (18 weeks, ITT): - Triglycerides: Greater reduction with PDP vs control. Between-group mean difference in change = −0.13 mmol/L (log-transformed 95% CI = −0.07 to −0.01), P=0.016 (adjusted); unadjusted P=0.018. Mean change from baseline: PDP −0.21 mmol/L (95% CI −0.33 to −0.10); control −0.07 mmol/L (95% CI −0.15 to 0.02). - LDL-C: No significant between-group difference. Between-group difference = −0.04 mmol/L (95% CI −0.16 to 0.08), P=0.521 (unadjusted P=0.504). Mean change: PDP −0.01 mmol/L (95% CI −0.08 to 0.09); control 0.04 mmol/L (95% CI −0.05 to 0.13). Secondary outcomes (18 weeks, ITT): - Body weight: Between-group difference −2.46 kg (95% CI −3.67 to −1.25). Within-group: PDP −2.17 kg (95% CI −3.03 to −1.31); control 0.30 kg (95% CI −0.56 to 1.15). - Waist circumference: Between-group difference −2.35 cm (95% CI −4.07 to −0.63). Within-group: PDP −2.94 cm (95% CI −4.17 to −1.71); control −0.59 cm (95% CI −1.81 to 0.63). - HbA1c: Between-group difference −0.05% (95% CI −0.10 to −0.001). Within-group: PDP −0.02% (95% CI −0.05 to 0.01); control 0.03% (95% CI −0.01 to 0.07). - Diet quality (HEI): Between-group difference +7.08 points (95% CI 5.02 to 9.15). Within-group: PDP +7.01 (95% CI 5.51 to 8.51); control −0.08 (95% CI −1.35 to 1.50). - No significant between-group differences for hip circumference, blood pressure, fasting insulin, glucose, C-peptide, apolipoproteins A1 and B, or postprandial TG. Other labs largely unchanged except mean platelet volume and absolute lymphocytes. Dietary intake: - PDP had lower dietary energy density at endpoint vs control (1.67±0.38 vs 1.87±0.38 kcal/g, P<0.001). Significant between-group differences at week 18 for % energy from carbohydrates, fat, PUFA, fiber, and energy density. PDP reduced energy intake more than control (between-group difference 162 kcal/day, 95% CI 22 to 302; P<0.001 for group×time adjusted for age/sex). High interindividual variability in nutrient and food intake changes in both groups. Adherence and perceptions: - 30% more PDP participants reported high/very high adherence (scores ≥8/10). Top 30% by diet quality improvement increased HEI by 12.9% (PDP) vs 6.15% (control). Subjective improvements reported in PDP vs control: energy (43% vs 11%), sleep quality (35% vs 9%), mood (33% vs 15%), reduced hunger (22% vs 14%) (all P<0.01). Microbiome: - Within-individual beta-diversity (Bray–Curtis) increased at week 12 in both groups vs baseline (paired Wilcoxon P<0.01). PDP showed greater divergence over time than control; across-group difference at week 18 significant (KSp=0.04). Favorable species: 8/15 increased in PDP vs 0/15 in control; summed abundance change 0.48±9.05 vs −0.73±8.63 (MWWp=0.015). Unfavorable species showed no between-group difference. ML models discriminated participants by changes in weight and hip circumference in PDP (AUC 0.65 and 0.59) but not control (AUC 0.49 for both). Safety: - Four adverse events, none severe; no withdrawals due to injury. One probable mild allergic reaction to test muffin (undisclosed nut allergy) led to withdrawal; others included bruising, light-headedness at blood draw, and mild CGM-site bleeding; all resolved. Per-protocol (PP): - TG reduction greater with PDP vs control: between-group difference −0.17 mmol/L (log-transformed 95% CI −0.07 to −0.01; P=0.032). Mean change: PDP −0.23 mmol/L (95% CI −0.33 to −0.12); control −0.06 mmol/L (95% CI −0.16 to 0.05). LDL-C differences remained non-significant (0.05 mmol/L; 95% CI −0.08 to 0.19; P=0.430). Secondary outcomes differences were generally larger in PP and in highly adherent PDP participants.

The trial demonstrates that a multi-input personalized dietary program can elicit modest but significant improvements in fasting triglycerides, body weight, waist circumference, HbA1c, and diet quality relative to standard USDA guidance, in generally healthy middle-aged/older adults. LDL-C did not differ between groups, and many cardiometabolic markers were unchanged, indicating the benefits are selective. The PDP’s integration of individual postprandial glucose and lipid responses, microbiome composition, and health history likely contributed to tailored food choices, lower dietary energy density, and improved satiety, supporting modest weight loss without explicit calorie restriction. Microbiome analyses show that dietary changes in both groups alter gut composition, but PDP induced greater and more sustained divergence and increased several taxa previously associated with favorable cardiometabolic health; microbiome shifts in PDP were modestly predictive of anthropometric improvements. Compared with prior personalized nutrition trials that often improved diet quality without superior weight effects, this multilevel PDP produced small but statistically significant anthropometric benefits, especially among highly adherent participants. These findings support the concept that addressing metabolic heterogeneity with personalized guidance can enhance select cardiometabolic outcomes beyond generalized advice, while acknowledging that effects on LDL-C and several biomarkers were not observed over 18 weeks.

An app-delivered personalized nutrition program leveraging individual postprandial responses, microbiome profiles, and health history improved fasting triglycerides, body weight, waist circumference, HbA1c, diet quality, and gut microbiome beta-diversity versus standard USDA dietary advice over 18 weeks in generally healthy adults. LDL-C and several other biomarkers were unchanged. The approach supports modest cardiometabolic benefits, particularly with higher adherence, and underscores the potential of personalization to complement population guidelines. Future research should: (1) evaluate longer-term sustainability and clinical endpoints; (2) refine algorithms using larger, more diverse cohorts and integrate lifestyle factors (sleep, physical activity, meal timing); (3) compare the isolated effect of personalized food scores versus broader program components; and (4) improve representativeness across sex, age, and ethnicity.

- Physical activity changes were not accurately captured, potentially confounding outcomes. - Interventions were not matched for contact intensity: the PDP involved more interaction and app engagement than the control, possibly influencing adherence and behavior. - Some behavioral change occurred in controls (e.g., increased fiber, reduced fat), and awareness of trial participation may have influenced behavior in both groups. - Limited diversity: cohort predominantly female and White; larger studies needed for broader ethnic and gender representation; findings not applicable to children or older adults beyond the studied age range. - Free-living, remote design enhances generalizability but limits control over compliance and measurement fidelity. - Short duration (18 weeks) may be insufficient to detect changes in LDL-C and other cardiometabolic markers. - Algorithm tested was the first version (2022); further refinement may yield different outcomes.