Optimisation of children z-score calculation based on new statistical techniques

Area of Science:

• Pediatric endocrinology and nutrition
• Biostatistics and computational modeling
• Growth assessment and anthropometry

Background:

• Accurate anthropometric parameter (height, weight, BMI) z-score calculation is crucial for pediatric clinical assessment.
• Current methods like the CDC-LMS are widely used but can lack precision in certain percentiles.
• Existing growth charts and z-score calculators may not fully capture growth nuances.

Purpose of the Study:

• To enhance the accuracy of z-score calculation for pediatric anthropometric parameters.
• To revise the statistical methodology using original data from established growth chart development.
• To provide a more precise tool for assessing growth and nutritional status in children and adolescents.

Main Methods:

• Development and internal validation of a Gaussian Process Regression (GPR) model.
• Comparison of GPR-derived z-scores against WHO and CDC-LMS methods.
• Evaluation using standard z-score cut-off points, a simulated cohort (n=3000), and real patient data (n=212, ages 2-18).

Main Results:

• GPR demonstrated significantly higher accuracy in z-score calculation compared to CDC-LMS and WHO methods (p<<0.001).
• Weight-for-age, height-for-age, and BMI-for-age z-score calculations showed considerable variation across methods in simulated and real data.
• Observed variations were larger for z-scores near 0 ± 1 compared to those outside ± 2.

Conclusions:

• The revised GPR method offers superior accuracy for standard z-score cut-off points over existing CDC-LMS and WHO approaches.
• GPR-based calculations provide more accurate z-score determinations, improving patient classification relative to growth cut-offs.
• Clinicians and statisticians should consider adopting updated z-score calculation methods for enhanced accuracy in pediatric growth assessment.