Development and Internal Validation of a Nomogram for Concomitant Pulmonary Embolism in Patients With Lower-Extremity Deep Vein Thrombosis: A Comparison With Machine Learning Models
This study developed and internally validated a logistic regression-based nomogram using clinical predictors to estimate the probability of concomitant pulmonary embolism in patients with lower-extremity deep vein thrombosis, demonstrating performance comparable to advanced machine learning models while offering superior interpretability, though it requires external validation before clinical implementation.
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
When a blood clot forms in the deep veins of a leg, it is a serious medical event known as deep vein thrombosis. The greatest fear for doctors and patients is that a piece of this clot might break loose, travel through the bloodstream, and lodge in the lungs. This secondary event is called a pulmonary embolism, and it can be life-threatening. In many cases, the clot in the leg and the clot in the lungs appear at the same time, even if the patient shows no specific signs of trouble in their breathing. Currently, the standard way to check for a lung clot is a specialized CT scan of the chest. However, performing this scan on every single patient with a leg clot is not practical. The scan involves radiation and a dye that can strain the kidneys, so doctors usually reserve it for patients who show clear warning signs, such as sudden shortness of breath or a drop in oxygen levels. This creates a difficult gap: many patients with leg clots might also have silent lung clots that go undetected because they do not look sick enough to trigger a scan.
Researchers at the China-Japan Friendship Hospital in Beijing set out to bridge this gap by creating a tool to estimate the likelihood of a hidden lung clot in patients who have just been diagnosed with a leg clot. They gathered data from over 1,100 patients admitted to their hospital between 2019 and 2024. For each patient, they looked at a specific set of seven factors available at the time of admission: their age, their body mass index, a blood test result called D-dimer, whether they had any existing lung disease, if they had a history of blood clots, where exactly the leg clot was located, and whether clots were found in veins elsewhere in the body, such as the arms or neck. They then compared how well a traditional statistical method, known as a nomogram, performed against three more complex computer algorithms often called machine learning models. The goal was to see if a simple, easy-to-read chart could predict the presence of a lung clot as accurately as these sophisticated digital systems.
The study found that among the 1,151 patients, 173 were confirmed to have a lung clot. The researchers discovered that certain factors made a lung clot much more likely. Patients who had existing lung conditions, such as chronic obstructive pulmonary disease or pneumonia, were nearly eight times more likely to have a lung clot than those without such conditions. The presence of clots in veins outside the legs, like in the arms or neck, was an even stronger indicator, associated with a risk increase of more than nine times. A history of previous blood clots and clots extending into the back of the knee or lower leg also significantly raised the odds. Interestingly, the study noted that older age was associated with a lower likelihood of a lung clot in their data, but the authors caution that this is likely due to how patients were selected for testing rather than a biological protection, as older patients might have been less likely to show the sudden breathing symptoms that trigger a scan.
When the team tested their tools on a separate group of patients to see how well they worked, the results were revealing. The complex machine learning models, including a system called gradient boosting, performed slightly better than the simple statistical model, but the difference was so small that it was not statistically significant. The traditional nomogram, which is essentially a chart where a doctor adds up points for each risk factor to get a total score, performed just as well as the advanced computer models in distinguishing between patients with and without lung clots. The simple model correctly identified about 77 percent of the patients who actually had a lung clot and correctly ruled out about 74 percent of those who did not. This suggests that for this specific medical question, adding layers of complex computer processing did not provide a meaningful advantage over a straightforward calculation based on seven clear clinical facts.
Despite these promising results, the researchers are careful to state that this tool is not a diagnostic test that can replace a CT scan. Because the hospital only performed scans on patients who looked sick, the data contains a bias; many patients without symptoms were never scanned, meaning some silent lung clots might have been missed entirely. When the researchers looked only at the patients who actually received a scan, the strength of the connections between the risk factors and the lung clots became weaker. This indicates that the model is partly predicting which patients are likely to be scanned, rather than just predicting the biological presence of a clot. Therefore, the chart should be viewed as an exploratory guide to help doctors decide who needs closer attention, rather than a definitive answer. Before this tool can be used in clinics to guide patient care, it must be tested in other hospitals with different patient populations and with a more consistent approach to scanning everyone, ensuring that the predictions hold true in the real world.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.