📄 medicine

Development and validation of CMI5: a novel composite metabolic index for predicting cardiometabolic multimorbidity using routine health check-up data

This study developed and validated CMI5, a novel composite metabolic index based on five routine clinical indicators (HbA1c, age, fasting glucose, blood urea nitrogen, and BMI), which demonstrated strong predictive performance for cardiometabolic multimorbidity in US data while showing attenuated but still discriminative utility in external validation.

Original authors: Yanyan Hu¹, Gongquan Wang², Yuhua Chi, Shuhan Fang, Manman Niu, Bing Du¹

Published 2026-08-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Yanyan Hu¹, Gongquan Wang², Yuhua Chi, Shuhan Fang, Manman Niu, Bing Du¹

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your body is like a bustling city. Usually, the city runs smoothly, but sometimes, the power grid (your metabolism) starts to flicker, the traffic jams (blood flow) get worse, and the water pipes (your kidneys) start to clog. When these problems happen one by one, doctors call them individual diseases like diabetes or heart trouble. But when they all crash at the same time, creating a perfect storm of multiple health issues, it's called "cardiometabolic multimorbidity." It's a heavy burden that can shorten lives and make daily living much harder.

For years, doctors have had a toolbox full of routine health check-up items—like checking your blood sugar, weighing you, or looking at your age—to spot these problems early. But here's the tricky part: there are so many different tests, and it's often unclear which ones are the real "superheroes" for predicting when the whole city is about to collapse. Scientists have been trying to build a single, simple "weather forecast" that combines the best clues to tell us who is at risk of this multi-disease storm, without needing expensive or complicated new tests. The big question is: Can we find a handful of common, everyday numbers that, when mixed together, act like a crystal ball for our future health?

That is exactly what a team of researchers set out to do. They created a new "metabolic weather forecast" called CMI5. Think of it as a special recipe where they took the five most powerful ingredients from a massive pantry of 48 different health checks and mixed them into one score. To test if this recipe worked, they cooked up a model using a huge database of health records from the United States (involving over 32,000 people) and then tried to serve it to a different group of people in China (over 13,000 people) to see if the taste held up.

The researchers found that their five secret ingredients were: Hemoglobin A1c (a measure of long-term blood sugar), Age, Fasting Blood Glucose (sugar levels after not eating), Blood Urea Nitrogen (a sign of how well kidneys are working), and Body Mass Index (a measure of body size). When they combined these into the CMI5 score, the results were impressive. In the US group, the new index was a fantastic predictor. It could tell the difference between people with and without the multi-disease storm with an accuracy score (called an AUC) of 0.904. To put that in perspective, a perfect score is 1.0, and a random guess is 0.5. This means the CMI5 was very sharp at spotting the danger.

The study also discovered a "tipping point" in the data. They found that as long as your CMI5 score stayed below 9.5, the risk of having multiple diseases rose slowly, like a gentle hill. But once the score crossed 9.5, the risk didn't just climb; it skyrocketed up a steep cliff. It was as if the city's defenses suddenly gave way, and the danger of a health crisis jumped dramatically. For every single point the score went up, the odds of having these combined diseases increased by about 6.5 times.

However, the story has a twist when the recipe traveled to China. When the researchers tested the same CMI5 score on the Chinese group, it still worked, but it wasn't quite as sharp. The accuracy score dropped to 0.742. The authors suggest this happened because the "ingredients" might behave slightly differently in different populations. For instance, the relationship between body size and health risks can vary between people of different ethnic backgrounds, and dietary habits (which affect kidney markers) differ between the US and China. It's like a recipe that tastes perfect in one kitchen but needs a little adjustment of the spices to taste right in another.

So, what does this mean? The study suggests that CMI5 is a powerful, low-cost tool that could help doctors in the US quickly screen for people at high risk of developing multiple heart and metabolic diseases using just five numbers they already check. It highlights that when these five factors combine, they tell a much stronger story than any single test could alone. However, the researchers are careful to say that because the score performed differently in another country, it shouldn't be used as a final diagnosis everywhere just yet. It's a promising new compass for navigating health risks, but it might need to be recalibrated before it can guide everyone on the planet with the same precision.

Technical Summary: Development and Validation of CMI5

Problem Statement
Cardiometabolic multimorbidity (CMM), defined as the coexistence of two or more cardiometabolic diseases (diabetes, heart disease, stroke, and hypertension), represents a severe public health burden associated with reduced life expectancy and quality of life. While routine health check-ups generate extensive clinical data, the predictive utility of common parameters for CMM specifically remains insufficiently characterized. Existing composite indices (e.g., TyG, TyG-BMI, AIP) primarily target single cardiometabolic diseases rather than the multimorbid state. Furthermore, current screening panels often include tests with unclear clinical value, leading to inefficiency and unnecessary costs. There is a critical need to identify the most diagnostically valuable routine indicators from a broad panel to construct an efficient, low-cost prediction model for CMM risk stratification.

Methodology
This study employed a cross-sectional design utilizing data from two major cohorts:

Discovery Set: The National Health and Nutrition Examination Survey (NHANES) 2013–2023 ( $n = 32,843$ ).
External Validation Set: The China Health and Retirement Longitudinal Study (CHARLS) 2015–2016 ( $n = 13,725$ ).

Data Processing and Feature Selection:

Indicator Screening: Forty-eight routine clinical indicators (including demographics, blood pressure, routine blood tests, liver/kidney function, and lipid/glucose profiles) were evaluated.
Univariate Analysis: Receiver Operating Characteristic (ROC) analysis was performed to determine the Area Under the Curve (AUC) for each indicator individually.
Index Construction: The top five predictors based on AUC were selected to construct the Composite Metabolic Index (CMI5).
Modeling: A regularized random forest algorithm (using the rfsrc package in R) was trained on the NHANES data. The model utilized the five CMI5 components as input features to predict the binary outcome of CMM. Hyperparameters were tuned via ten-fold cross-validation.
Statistical Analysis: Multivariable logistic regression was used to assess independent associations, and Restricted Cubic Spline (RCS) analysis was employed to explore non-linear dose-response relationships and threshold effects.

Key Contributions

Systematic Screening: The study systematically evaluated 48 routine indicators, identifying a specific subset with the highest predictive performance for CMM, moving beyond single-disease indices.
Novel Index (CMI5): The authors constructed CMI5, integrating five readily accessible indicators: Glycated Hemoglobin (HbA1c), Age, Fasting Blood Glucose (FBG), Blood Urea Nitrogen (BUN), and Body Mass Index (BMI).
Threshold Identification: The study identified a specific non-linear threshold effect, revealing a sharp escalation in CMM risk when CMI5 exceeds 9.5.
Cross-Population Validation: The model was rigorously tested in both US (NHANES) and Chinese (CHARLS) populations to assess generalizability, providing a realistic assessment of cross-ethnic applicability.

Results

Predictor Ranking: In the NHANES discovery cohort, the top five predictors were HbA1c (AUC = 0.831), Age (AUC = 0.777), FBG (AUC = 0.762), BUN (AUC = 0.658), and BMI (AUC = 0.621).
Model Performance (Internal): In the NHANES internal test set, the CMI5-based regularized random forest model achieved an AUC of 0.904 (95% CI: 0.896–0.912). The training set AUC was 0.969.
Risk Association: Multivariable logistic regression indicated that each one-unit increase in CMI5 corresponded to a 7.53-fold higher odds of CMM (OR = 7.53, 95% CI: 7.30–7.88, $p < 0.001$ ), a magnitude significantly higher than any single component indicator.
Dose-Response: RCS analysis confirmed a non-linear relationship where CMM risk increased gradually below a CMI5 of 9.5 but rose exponentially above this threshold.
External Validation: In the CHARLS cohort, the model achieved an AUC of 0.742. While discriminative, this represents a notable performance attenuation compared to the US cohort. The optimal cutoff for the CHARLS set was 0.144 (sensitivity 0.90, specificity 0.85).

Significance and Claims
The paper claims that CMI5 serves as a low-cost, easy-to-implement screening tool for primary CMM risk stratification, particularly within US populations where the model was derived. The integration of five routine indicators allows for the early identification of high-risk individuals without additional testing costs.

However, the authors maintain a modest and cautious stance regarding generalizability. They explicitly state that the model's cross-population generalizability is limited by ethnic heterogeneity and the low-dimensional feature space (only five indicators). The performance drop in the Chinese cohort (CHARLS) is attributed to differences in metabolic implications of BMI and BUN across ethnicities, as well as potential overfitting. Consequently, the authors conclude that the current version of CMI5 should not replace comprehensive clinical evaluation in non-US settings and requires calibration or retraining before application in Chinese or other non-US populations. The study emphasizes that while CMI5 is a promising auxiliary tool for primary screening, future research must focus on multi-center validation, prospective studies for incident CMM prediction, and model calibration to improve cross-population stability.

Technical Summary: Development and Validation of CMI5

More like this