Diagnostic accuracy of machine-learning models for detecting pulmonary embolism in patients with pneumonia: a systematic review and meta-analysis.

What was studied

This review looked at computer models that learn from data. The models tried to spot a blood clot in the lung, called pulmonary embolism. The people in the studies had pneumonia. Both illnesses can look alike, so the clot can be hard to find. The team searched several research databases up to May 2026. Then they pooled the results of studies that gave enough data.

What they found

Overall, the models were good at telling clot from no clot. The main pooled result gave a sensitivity of 0.91. Sensitivity shows how well the models caught real clots. The specificity was 0.92. Specificity shows how well the models ruled clots out. Results varied a lot from study to study. Deep-learning models using CTPA scans were generally more accurate. They did better than models using clinical or mixed data. The certainty of the evidence was rated very low.

What it means for patients

The authors say these models may help doctors as an extra tool. They would be used alongside usual tests for clots in pneumonia. But the authors say the evidence is not yet strong enough for daily care. They say new studies at many centers are needed first. Ways of testing the models also need to be the same across studies.

Why it is not certain yet

Study results differed widely, so the pooled numbers may not hold. The range of likely results was very wide. The authors say the pooled result does not show how a real tool would work. The evidence was mixed, indirect and rated very low in certainty. Some comparisons were only described and still need testing. A review can only be as good as the studies in it.

Source

Diagnostic accuracy of machine-learning models for detecting pulmonary embolism in patients with pneumonia: a systematic review and meta-analysis.

We restate facts from the source record in our own words. The original wording belongs to its publisher.