A multicenter analysis from the REFINE-PET registry examined whether automated ascending aortic calcification (AAC) stratifies risk in patients who otherwise appear low risk on PET. The cohort included 21,923 patients; investigators identified 3,580 (16.3%) as “triple-normal” according to the report’s criteria (stress total perfusion deficit (TPD) 2 and CAC 0). Among these triple-normal patients, 1,322 (36.9%) had measurable AAC, with a reported median Agatston score of 65 (IQR 22–212). The study team framed AAC as a potential residual-risk marker detectable on the non-contrast attenuation correction CT routinely acquired with PET.
Short declarative facts coexist with nuanced observations. The triple-normal subgroup was small relative to the registry but substantial enough to examine event rates over prolonged follow-up. The authors emphasized that routine PET metrics alone do not capture all risk, and that an imaging feature outside the coronary arteries—namely ascending aortic calcification—may carry independent prognostic information.
Investigators quantified AAC using deep-learning aortic segmentation on non-contrast CT images and calculated Agatston scores. The source reported good-to-excellent agreement between the automated method and three blinded human readers, with intraclass correlation coefficients ranging from 0.87 to 0.90. The registry description specified that AAC extraction was fully automated, but the article did not report detailed CT acquisition parameters or the specific deep-learning architecture used. Those methodological details were not provided in the source.
The automated approach was positioned as one that could be applied to attenuation-correction CT without additional patient imaging. The authors noted that such automation enables scalable AAC measurement across sites that already perform PET with attenuation-correction CT.
Median follow-up for the triple-normal group was 52 months (IQR 26–69). During that interval, the study recorded 194 occurrences of the primary composite endpoint, which the authors defined as cardiovascular death, myocardial infarction or heart failure admission. The analysis used Cox proportional hazards models adjusted for clinical confounders.
Per reported results, each standard deviation increase in AAC was independently associated with the primary endpoint (adjusted hazard ratio [HR] 1.24; 95% confidence interval [CI] 1.09–1.40; p<0.001). Association signals extended to individual outcomes: heart failure admission showed an adjusted HR of 1.30 (95% CI 1.10–1.55; p=0.003), and all-cause mortality an adjusted HR of 1.17 (95% CI 1.04–1.32; p=0.009). The investigators therefore observed that AAC portended elevated risk even in patients with normal perfusion, preserved flow metrics and absent coronary artery calcium on PET.
The study team tested whether AAC added prognostic information beyond established clinical and PET-based markers. Data indicated that AAC improved model fit when added to models containing clinical covariates, myocardial flow reserve and total perfusion deficit (reported p ≤ 0.009). However, the incremental discrimination was modest: the change in C-statistic ranged from 0.007 to 0.009.
Those numbers suggest measurable but small gains in overall model performance. The report did not present alternative risk thresholds or decision-analytic metrics that would translate the modest ΔC into clinical action. The source did not supply recommended cutpoints for AAC to trigger downstream testing or therapy in the triple-normal population.
Among the component outcomes, the strongest signal was for heart failure admission. Investigators observed that AAC’s association with heart failure admission remained statistically significant after multivariable adjustment, with an adjusted HR of 1.30 per standard deviation increase in AAC. All-cause mortality also showed a smaller but statistically significant association (adjusted HR 1.17).
These results indicate that aortic wall calcium may relate to processes that predispose to heart failure or to competing systemic risks not captured by coronary-focused PET metrics. The source, however, did not provide mechanistic data, specific heart-failure subtypes, or temporal patterns of events relative to AAC measurement.
The registry analysis carried several constraints that the authors reported. The source did not provide detailed per-site CT acquisition parameters or the specific deep-learning model architecture; those methodological items were not reported. The study used retrospective registry data, and sites either obtained written informed consent or had consent waived; institutional review boards at participating centers approved the protocol. The authors declared multiple competing interests across the investigator team, as noted in the source.
Data supporting the findings are available from the corresponding author under REFINE-PET data-use agreements and subject to institutional approvals, the authors reported. The article did not define AAC thresholds for clinical management or provide prospective validation steps; those remain unanswered questions documented in the source.