Development and Validation of a Machine Learning-Based Prediction Model for Atrial Fibrillation in Hospitalized Patients with Coronary Heart Disease and Diabetes Mellitus: A Retrospective Study

In the human body, the heart is a relentless pump, but like any complex machine, it is vulnerable to wear and tear, especially when other systems are under stress. Two of the most common conditions that strain this system are coronary heart disease, where the arteries feeding the heart muscle become narrowed, and diabetes, a metabolic disorder that disrupts how the body processes sugar. When these two conditions occur together, they create a dangerous synergy that significantly raises the risk of a third, often silent threat: atrial fibrillation. This is a chaotic rhythm where the heart's upper chambers quiver instead of beating steadily, a state that can lead to stroke or heart failure. For doctors, the challenge has long been identifying which patients with both heart disease and diabetes are most likely to develop this irregular rhythm before it strikes, allowing for early intervention rather than emergency reaction.

A team of researchers at the Chongqing Emergency Medical Center set out to solve this puzzle by looking backward through the medical records of over two thousand patients. They gathered data from individuals admitted to the hospital between 2017 and 2025 who were already living with both coronary heart disease and diabetes. The researchers split these patients into two groups: those who developed atrial fibrillation during their stay and those who did not. By comparing the two groups, they searched for subtle differences in blood work, heart measurements, and patient history that could serve as warning signs. Instead of relying on a single test, they used a sophisticated computer method known as machine learning. This approach allows a computer to sift through vast amounts of information to find complex patterns that human eyes might miss, much like how a seasoned mechanic might hear a specific rattle in an engine that indicates a problem long before the car breaks down.

The study analyzed a wide array of data points, from the size of the heart's chambers to the levels of specific proteins and minerals in the blood. After testing several different computer algorithms, the researchers found that one method, called a random forest model, was far superior at predicting who would develop the irregular rhythm. This model achieved a high level of accuracy, correctly identifying the risk in the vast majority of cases. The computer learned that the most important warning signs were not just the presence of the heart disease or diabetes themselves, but a specific combination of factors. These included the patient's age, the physical dimensions of the heart's upper chambers, and a mix of blood markers related to inflammation, nutrition, and electrolyte balance.

Specifically, the model identified eight key indicators that, when viewed together, painted a clear picture of risk. Older age and larger dimensions of the left and right upper chambers of the heart were strong predictors, reflecting the physical strain these conditions place on the organ's structure. The blood levels of prealbumin, a protein that signals nutritional status, were also critical; lower levels suggested a body under stress. Similarly, the balance of white blood cells to albumin, a measure of how much inflammation was present relative to the body's reserves, proved to be a powerful signal. Other factors included the levels of calcium and sodium in the blood, which are essential for the heart's electrical signals, and a specific enzyme called HBDH, which can indicate stress on the heart muscle.

To ensure that this computer model was not just a black box making guesses, the researchers used a technique to explain exactly how it reached its conclusions. They found that the model weighed these eight factors differently for each patient, creating a personalized risk profile. For one person, the size of their heart chambers might be the primary driver of risk, while for another, it might be a combination of low nutrition and high inflammation. This ability to break down the prediction into understandable parts means that doctors can look at a specific patient and see exactly which factors are pushing them toward danger. The study suggests that by using this tool, clinicians could screen high-risk patients more effectively, monitor them more closely, and potentially start treatments earlier to prevent the onset of atrial fibrillation.

The researchers were careful to note that their findings came from a single hospital and that the model needs to be tested on patients in other locations to confirm its reliability. However, the results offer a promising new way to look at a common and dangerous medical problem. By combining routine blood tests and standard heart scans with advanced computer analysis, the study demonstrates that it is possible to spot the warning signs of atrial fibrillation in patients with heart disease and diabetes long before the condition becomes life-threatening. This approach moves medicine away from a one-size-fits-all approach toward a more precise, personalized strategy for keeping the heart beating in rhythm.