Estimating regional centile curves from mixed data sources and countries.

Area of Science:

• Biostatistics
• Global Health
• Anthropometry

Background:

• Accurate population growth data is vital for health assessment, especially in resource-poor nations.
• Existing anthropometric data is often unavailable or fragmented across different countries.
• Traditional growth surveys are expensive and not always feasible.

Purpose of the Study:

• To develop a practical, cost-effective method for estimating regional age-conditional weight distributions.
• To create country-level and multi-country regional growth references from existing survey data.
• To demonstrate the method's utility for applications like region-specific drug dosing.

Main Methods:

• A two-step modeling approach using generalized additive models for location, scale, and shape (GAMLSS).
• Step 1: Modeling country-specific centile curves using Box-Cox t and Box-Cox power exponential distributions.
• Step 2: Defining regional references via a finite mixture of country distributions, weighted by population size.

Main Results:

• Successfully fitted weight-for-age distributions for 12 countries in South East Asia and the Western Pacific using 273,270 observations.
• Achieved a good fit for both raw body weight and Z-scores using the proposed modeling technique.
• Generated reliable country-level and regional growth references.

Conclusions:

• The proposed method provides an affordable and efficient strategy for estimating regional growth distributions.
• This approach is particularly valuable for resource-poor countries lacking comprehensive anthropometric data.
• The generated regional references have practical applications, including optimizing pediatric drug dosages in tropical regions.