Summary: In a secondary analysis of a 6-month randomized trial in 155 adults with obesity and abnormal glucose metabolism, higher baseline weight-loss self-efficacy predicted achieving at least 5% weight loss in both the low-fat and machine-learning personalized nutrition arms. Higher BMI predicted success in the low-fat arm (p < 0.0001) and older age predicted success in the personalized arm (p < 0.0001). The analysis identified predictors within each arm and did not test whether one diet outperformed the other.
PICO Summary
| Element | Detail |
|---|---|
| Population | 155 adults with obesity and abnormal glucose metabolism (mean age 59 [SD 10] years; 66.5% female; mean BMI 33.4 [SD 4.6] kg/m2), drawn from a 6-month behavioural counselling randomized controlled trial. Secondary analysis conducted in the United States. |
| Intervention | Personalized calorie-restricted nutrition diet guided by a machine-learning algorithm (gradient boosting machine used to identify baseline predictors of success), arm n = 71. |
| Comparison | Standardized low-fat calorie-restricted diet without algorithmic personalization, arm n = 84. Predictors were modelled in each arm separately, not compared head-to-head. |
| Outcome | Predictors of weight-loss success (at least 5% loss at 6 months), modelled with repeated five-fold cross-validation (100 repetitions). Higher baseline weight-loss self-efficacy predicted success in both arms. Higher BMI predicted success in the low-fat arm (p < 0.0001); older age predicted success in the personalized arm (p < 0.0001). No between-arm effect size, 95% CI, ARR, or NNT was reported; this analysis did not test relative diet efficacy. |
Expert Commentary
This is a hypothesis-generating secondary analysis, and it should be read as such rather than as evidence that algorithm-guided nutrition produces superior weight loss. The parent trial was randomized, but the question asked here is different: which baseline traits travel with reaching at least 5% loss within each diet arm. The consistent signal is psychological. Higher baseline weight-loss self-efficacy predicted success regardless of diet assignment, which aligns with a sizeable behavioural literature and is the most actionable takeaway. The arm-specific findings, higher BMI favouring the low-fat arm and older age favouring the personalized arm, are more fragile. With only 71 to 84 participants per arm and a machine-learning approach that screens many candidate variables, the risk of overfitting and spurious associations is real, and the reported p-values reflect predictor importance within models rather than confirmatory tests. The principal limitation is therefore sample size relative to model complexity, leaving these predictors in need of external validation before they guide triage. Can I use this with my patients? Not yet as a selection rule, but the self-efficacy finding reinforces assessing a patient’s confidence and readiness before prescribing any restrictive diet. I would like to see these predictors prospectively validated in a larger, pre-registered cohort. Notably, one co-author developed the personalized algorithm, so independent replication matters.
References
Popp CJ, Wang C, Berube L, Curran M, Hu L, Pompeii ML, et al. Baseline Characteristics of Weight-Loss Success in a Personalized Nutrition Intervention: A Secondary Analysis. Nutrients. 2025;17(13):2178. doi:10.3390/nu17132178