Radiology Feature: Model predicts CVD from ancillary CT findings

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
After reviewing nearly 7,000 chest CT scans, researchers in the Netherlands developed a model that reliably predicts the onset of cardiovascular disease (CVD) based exclusively on ancillary CT findings, according to an article published in the October issue of Radiology.

The new model “could potentially change the way radiologists contribute to the efficiency of daily patient care,” according to lead author Martijn Gondrie, MD, of the University Medical Center of Utrecht, the Netherlands, and colleagues.

Ancillary CT findings occur frequently given the widespread use of clinical CT. The radiology community has not yet quantified the prevalence and medical importance of these unexpected findings.

The study is part of the PROVIDI (PROgnostic Value of Ancillary Information in Diagnostic Imaging) study, which is the first longitudinal study to investigate the prevalence and clinical importance of ancillary findings in diagnostic chest CTs.

The authors began with a sample of 23,443 patients who underwent chest CTs between 2002 and 2005 in the Netherlands. Researchers excluded studies for a number of variables, including diagnoses of terminal cancer (which would prevent follow-up of cardiovascular events) and CT scans that were directly indicated for CVD (since these scans would make this study’s CVD findings primary rather than ancillary).

Three radiologists then reviewed 6,975 chest CTs, looking for four types of aortic abnormalities in the routinely ordered CTs: aortic calcification, aortic wall irregularity, aortic plaque and elongation of the descending thoracic aorta. Researchers scored each CT according to the predominance of any of these characteristics; for example, aortic wall calcifications with five or fewer foci earned a score of 1, 2 or fewer calcifications earned a 2 and any additional aortic wall observations, or nine or more foci, generated a score of 3.

The researchers used these non-indicated characteristics, in addition to patients’ age, sex and indication for the original CT, to construct hazard ratios or scores used to predict patients’ risks for CVD. Higher hazard ratios, which increased in persons with older age, men and increased ancillary findings, indicated higher risks for CVD.

CVD events included fatal or non-fatal heart disease, heart failure, peripheral arterial disease, aortic aneurysm, cerebrovascular disease and nonrheumatic valvular disease.

The study found that of the patients with scores in the top quartile, 7 percent encountered CVD events within the first year of the CT and 12 percent experienced CVD events within the first two years of the CT. This was compared with only 1 percent of low-risk patients experiencing CVD in the two years following the CT.

For all four of the ancillary aortic abnormalities, increases in abnormalities were significantly correlated with increased CVD events. When adding all of the abnormal and demographic characteristics together, however, the researchers found that a comprehensive model did not increase the predictability of CVD any more than each predictor did individually.

Instead, the authors developed a model for predicting CVD based solely on the sum score for aortic calcifications, between 1 and 3, which increased along with increases in the number of foci or the number of aortic sections across which the calcifications crossed.

The authors chose this model for its strong statistical correlation with CVD and for the convenience and applicability with which the model can be used by radiologists to predict CVD. According to the authors, their model “allows radiologists who are working from routinely recorded patient and image characteristics only to make meaningful and relatively accurate predictions that are on par with established models and thus contribute meaningfully to the early identification of patients at risk of CVD.”

The mean aortic calcification score for patients that experienced CVD events (817 patients) was 2.88, while the risk-score for CVD event-free patients (totaling 347 patients) was 1.75, a statistically significant difference.

“The results of this study show that radiologists can predict cardiovascular disease fairly well using unrequested calcifications of the aortic wall on CT, as well as the sparse patient information that radiologists typically have immediate access to (age, gender, CT indication),” explained Gondrie in an interview.

“Ultimately, this has the potential to become part of an easily executed risk stratification by radiologists which may reduce future cardiovascular events,Gondrie continued.

The authors pointed to several limitations of their study, including the study’s short average follow-up, 17 months, which might have excluded some late-onset CVD cases. The study also relied on medical records for excluding patients with histories of CVD, which the authors acknowledged may not be entirely accurate. The authors argued, though, that the study’s exclusion rate did not differ substantially with national CVD rates.

Gondrie offered, “Currently, radiological practice has struggled to keep pace with the accelerating improvements in imaging technology,” but emphasized that the ancillary predictions made possible in models like this one “are obtained for free (without extra radiation exposure or costs) and may hold valuable clues as to the patient’s overall health and their risk for future disease.”