Predicting rapid weight gain in six-month-old infants: an exploratory modeling study
Ana Daniela Ortega-Ramírez1, Carmen Alicia Sánchez-Ramírez1, Benjamín Trujillo-Hernández1
1Facultad de Medicina, Universidad de Colima, Colima, México.
Insights
Predicting rapid weight gain (RWG) in infants is crucial for preventing childhood obesity. While statistical and machine-learning models showed modest results, identifying early predictors remains key for intervention strategies.
Area of Science:
- Pediatric Health
- Biostatistics
- Machine Learning in Healthcare
Background:
- Rapid weight gain (RWG) in early life is a significant risk factor for childhood obesity.
- The multifactorial etiology of RWG necessitates advanced analytical approaches for early prediction.
Purpose of the Study:
- To compare the predictive performance of machine-learning (SVM, Naïve Bayes) and statistical models (LASSO, GLM) for RWG from birth to 6 months.
- To identify key early-life predictors of RWG.
Main Methods:
- Prospective infant cohort data analyzed.
- Four predictive models evaluated: SVM (linear kernel), Naïve Bayes, LASSO regression, and Generalized Linear Model (GLM).
- Performance metrics included AUC, accuracy, precision, sensitivity, specificity, and F1-score with 95% CIs.
Main Results:
- All models demonstrated comparable precision (0.70-0.75).
- The GLM showed the highest point estimates for AUC (0.66), specificity (0.45), accuracy (0.72), and F1-score (0.800).
- LASSO achieved the highest sensitivity (0.91), but confidence intervals overlapped across all metrics, indicating no statistically significant differences between models.
Conclusions:
- All evaluated models exhibited similar, modest discriminative ability for predicting RWG.
- Consistent early-life predictors were identified, underscoring the multifactorial nature of RWG.
- Larger cohorts are required to enhance predictive accuracy and fully evaluate machine-learning model potential for RWG prediction.
Background:
Rapid weight gain (RWG) in early life is a significant risk factor for childhood obesity. Its multifactorial etiology warrants exploratory statistical and machine-learning analysis to aid early prediction.
Methods:
Data from a prospective infant study were used to compare four models for predicting RWG from birth to 6 months: two machine‑learning methods (SVM with a linear kernel and Naïve Bayes), one regularized regression (LASSO), and one traditional statistical model (Generalized Linear Model, GLM). Performance was evaluated using AUC, accuracy, precision, sensitivity, specificity, and F1‑score, each with 95% confidence intervals (CI).
Results:
Precision was comparable across models (0.70-0.75). The GLM showed the highest point estimates for AUC (0.66, 95% CI 0.48-0.83), specificity (0.45, 95% CI 0.37-0.53), accuracy (0.72, 95% CI 0.53-0.86), and F1‑score (0.800), while LASSO achieved the highest sensitivity (0.91, 95% CI 0.84-0.95). However, all CIs overlapped, indicating no statistically significant differences.
Conclusion:
Although the GLM had the highest point estimates, all models showed similar and modest discriminative ability. Consistent early‑life predictors emerged across approaches, highlighting the multifactorial nature of RWG. Larger cohorts are needed to improve predictive accuracy and fully assess machine‑learning methods.
Impact:
Rapid weight gain (RWG) results from the dynamic interplay of biological, dietary, behavioral, and environmental factors. Developing robust models to identify key determinants is therefore essential. In this study, the GLM yielded the highest point estimates across key metrics, while the machine‑learning models nonetheless demonstrated promising potential. Predictive modeling in this context not only enables risk stratification but also provides insight into underlying mechanisms, thereby guiding future longitudinal research and informing preventive strategies to support healthy growth trajectories in early life.
