How diet classification shapes conclusions in network meta-analysis: an empirical study of macronutrient trials and blood lipids

This repository contains the dataset and R code supporting a methodological study investigating the impact of different intervention classification strategies (lumping and splitting) on the structure and findings of network meta-analyses (NMAs) of dietary macronutrient interventions.

Data were extracted from randomised controlled trials evaluating the effects of dietary interventions differing in macronutrient composition and quality on four lipid outcomes: low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), and triglycerides (TG).

Three intervention classification systems were applied:

1. Splitting: A detailed classification based on carbohydrate, fat, and protein content, incorporating additional indicators of macronutrient quality, including high monounsaturated fatty acid intake, low glycaemic index, and high dietary fibre content.

2. Lumping 1: A broader classification primarily based on dietary fat and carbohydrate content, allowing variation in protein intake within categories while maintaining distinct classifications for high-MUFA and ketogenic diets.

3. Lumping 2: A simplified classification using three levels of carbohydrate, fat, and protein intake without additional macronutrient quality indicators.

The repository includes:

• Excel dataset: Twelve outcome-specific datasets corresponding to the four lipid outcomes and three classification strategies. Each dataset includes study identifiers, treatment comparisons, effect estimates (mean differences), corresponding standard errors, and study-level variables, including the proportion of female participants, energy restriction status, and intervention duration.

• R code: Scripts for conducting the NMAs using the netmeta package in R, including network visualisation, estimation of relative treatment effects, treatment ranking, assessment of heterogeneity and inconsistency, and exploration of potential small-study effects. Meta-regression analyses are also included for LDL-C.

The materials are provided to facilitate transparency, reproducibility, and further methodological research on the implications of intervention node definition in nutritional network meta-analysis.

The datasets contain study-level aggregate data extracted from published research and do not include individual participant data.