Minimal measurement strategies for cardiometabolic risk classification within the Positive Health framework: A cross-sectional analysis of NHANES 1999-2004 data

Technical Summary: Minimal Measurement Strategies for Cardiometabolic Risk Classification within the Positive Health Framework

Problem Statement
Noncommunicable diseases (NCDs) remain a primary global mortality cause, driven by risk factors such as adiposity, metabolic dysfunction, and reduced physical fitness. While the "Positive Health" (PH) framework has expanded the definition of health to include physiological, social, and emotional dimensions, current PH instruments lack objective, scalable physiological indicators for metabolic and cardiovascular health. Although various anthropometric indicators are known to correlate with cardiometabolic risk, it remains unclear which specific combination of simple, self-measurable indicators provides meaningful improvements in risk classification over simpler approaches. There is a need to determine if a parsimonious set of measurements can achieve classification performance comparable to a full candidate set, thereby reducing the measurement burden in population health monitoring.

Methodology
This cross-sectional study utilized data from 8,275 adults in the National Health and Nutrition Examination Survey (NHANES) cohorts from 1999–2004. The researchers evaluated a predefined minimal set of four self-measurable indicators selected based on health standards and expert consultation:

1. Body Mass Index (BMI): Proxy for lipid metabolism.
2. Waist-to-Height Ratio (WHtR): Proxy for body fat percentage.
3. Mid-Upper Arm Circumference (MUAC): Proxy for glucose metabolism.
4. VO2max: Used as a proxy for the 6-minute walk test (6MWT) to represent cardiorespiratory fitness (measured in a subsample of n=2,919).

These indicators were linked to clinical risk outcomes categorized into low and high-risk groups based on epidemiological cut-offs: body fat percentage, HbA1c, glucose, insulin, cholesterol, and triglycerides.

The study employed two complementary analytical approaches:

• Nested Logistic Regression: Models were constructed in steps: (i) base model (age, sex); (ii) +BMI; (iii) +WHtR; (iv) +BMI+WHtR; and (v) full model (including MUAC and VO2max where available). Performance was assessed using Likelihood Ratio Tests (LRTs), ROC-AUC, and Precision-Recall AUC (PR-AUC) to handle class imbalances.
• Decision Tree Analysis: Used to identify non-linear relationships and determine the primary variables driving classification decisions (splitting variables) within the minimal set.

Key Results

• Dominance of WHtR: Across almost all models and outcomes, WHtR consistently demonstrated the strongest discriminative performance. In nested models, the difference in PR-AUC between BMI and WHtR ranged from -0.002 to -0.037, with WHtR emerging as the primary splitting variable in decision trees.
• Incremental Value of BMI: Adding BMI to a model containing WHtR resulted in only marginal gains in PR-AUC for most outcomes (0.000 to 0.008). A notable exception was triglycerides, where the addition of BMI yielded a larger gain (ΔPR-AUC = 0.039).
• Limited Value of MUAC and VO2max: The further inclusion of MUAC and VO2max provided limited additional classification value overall.
  • VO2max: In the subsample where VO2max was measured, adding it to the model did not improve classification beyond the anthropometric measures. However, in sex-stratified decision tree analyses, VO2max was an influential variable for men, while MUAC appeared more important for women.
  • Glucose/Insulin Outcomes: No meaningful decision trees could be generated for HbA1c, insulin, or glucose, indicating that the available predictors did not effectively split these specific outcomes.
• Model Performance: Decision trees combining the minimal set (including VO2max) showed higher sensitivity and balanced accuracy compared to BMI-only models, particularly for detecting individuals with elevated body fat, triglycerides, and total cholesterol.

Significance and Claims
The paper concludes that a parsimonious measurement strategy utilizing a limited number of simple, self-measurable indicators is sufficient for the classification of cardiometabolic risk. Specifically, the study claims that:

• WHtR is the primary driver: It consistently outperforms BMI and serves as the most effective single indicator for identifying individuals at risk for elevated body fat, triglycerides, and cholesterol.
• Diminishing Returns: Adding further measurements (MUAC, VO2max) provides little additional benefit to classification performance in the general population, suggesting that complex measurement protocols may not be necessary for initial risk stratification.
• Feasibility for Positive Health: These findings support the integration of simple, resource-efficient anthropometric strategies into the Positive Health framework. This approach can alleviate measurement burdens in public health and prevention settings while maintaining the ability to classify risk effectively.

The authors note limitations, including the cross-sectional nature of the data (precluding long-term risk prediction), the use of researcher-measured rather than self-measured data in NHANES, and potential selection bias in the VO2max subsample (which was younger and healthier). They suggest that future research should validate these findings in longitudinal cohorts and explore subgroup-specific thresholds.