Comparison of cNORM and LMS methods for estimating reference percentile curves from biometric data

Area of Science:

• Biostatistics
• Public Health
• Growth Curve Analysis

Background:

• Accurate biometric reference curves are vital for population health monitoring and screening.
• The Lambda Mu Sigma (LMS) method is a standard for generating age-specific reference percentiles.
• Alternative methods are needed to enhance accuracy, especially in extreme ranges.

Purpose of the Study:

• To compare the performance of the Lambda Mu Sigma (LMS) method with cNORM, a distribution-free approach, for modelling biometric reference curves.
• To evaluate accuracy and bias of both methods across different percentile ranges using NHANES data.
• To determine the suitability of cNORM as an alternative to LMS in public health applications.

Main Methods:

• Utilized National Health and Nutrition Examination Survey (NHANES) datasets.
• Compared Lambda Mu Sigma (LMS) and cNORM methods for body mass index (BMI) and maximum oxygen consumption (VO₂max) reference curves.
• Employed random sampling, cross-validation, and performance metrics (R², RMSE, Bias).

Main Results:

• Both cNORM and LMS demonstrated high accuracy across full BMI and VO₂max distributions.
• cNORM exhibited superior precision in extreme percentiles (±2 SD), vital for identifying at-risk individuals.
• Accuracy improved with larger sample sizes, with a more pronounced effect for LMS, though interactions were inconsistent.

Conclusions:

• The distribution-free cNORM approach is a viable alternative to LMS for public health reference curves.
• cNORM is particularly advantageous when precise classification in extreme ranges is critical for screening.
• This study highlights the importance of method selection for accurate health screening based on biometric data.