Patients with suspected or known CAD referred for 82Rb-PET were screened and invited to participate in this prospective study between 2018 and 2022. Patients with known CAD or scans without available CACS were excluded from this analysis (figure 1).

Figure 1.

Flowchart of patient selection. For this project, only patients without known CAD and with available CACS were included. CACS, coronary artery calcium score; CAD, cardiovascular arterial disease; PET, positron emission tomography.

(0.11MB).

Clinical likelihood (CL) was calculated for each patient with the RF-CL described in figure 4 of the 2024 ESC guidelines for the management of CCS.7 The lowest and highest age categories were used in patients younger than 30 years (n=6) and those older than 80 years (n=133), respectively.

CL was refined with the CACS-CL using the formula published by Winther et al.9 Based on the initial RF-CL, patients were assigned to either lower or higher clinical likelihood according to their CACS value. Details and the formula used are described in the supplementary data.

The definition of CL categories was similar to that of the guidelines: very low 0% to 5%, low> 5% to 15%, moderate> 15% to 50%, high> 50% to 85%, very high> 85%.

Our current diagnostic algorithm (considered the reference standard) consists of a CACS and PET scan for all patients referred with suspected CAD. Accordingly, each patient was scanned with CACS and PET. Three different gatekeeper algorithms were retrospectively applied, as depicted in figure 2.

Figure 2.

Three different diagnostic algorithms for optimizing pretest selection. The upper part depicts 3 different diagnostic algorithms. Depending on the RF-CL, a PET scan would be performed or withheld. The pie chart diagrams indicate the number of patients and the proportion of abnormal findings in the corresponding group. The gray boxes indicate scan results which would not be available if the algorithm had been used. Test characteristics refer to the detection a small ischemia (SDS ≥ 2). CACS, coronary artery calcium score; CACS-CL, coronary artery calcium score-weighted clinical likelihood; NPV, negative predictive value; PET, positron emission tomography; RF-CL, risk factor-weighted clinical likelihood.

(0.75MB).

In algorithm 1, patients with an RF-CL PTP ≤ 5% were deferred from testing as recommended by the guidelines. All patients with an RF-CL PTP> 5% received the standard protocol with CACS and PET.

Algorithm 2 was an extension of algorithm 1: CACS was used to calculate CACS-CL PTP in patients with a low RF-CL PTP (> 5%-15%). If reclassified to 0% to 5%, patients were deferred from testing with PET, whereas patients with RF-CL ≥ 15% or CACS-CL> 5% were tested with PET.

In algorithm 3, CACS-CL was used in all patients for risk stratification. No PET scan was performed if CACS-CL was very low (0%-5%).

Imaging protocols were used as previously described.4,5 In brief, patients were scanned with a 3-dimensional PET/CT scanner (Biograph mCT, Siemens Healthineers, Germany). A low-dose CT scan was obtained for attenuation correction. Subsequently, a second, electrocardiogram-triggered nonenhanced low-dose CT during breath hold was acquired for CACS.

Thereafter, 82Rb-Chloride was intravenously injected in a weight-adjusted manner (30-40 mCi) both at rest and during stress. Rest imaging was always performed first. Stress was pharmacologically induced with adenosine or regadenoson.

Reconstructed images were visually inspected with QGS-QPS software (SyngoVia, Siemens Healthineers). CACS was calculated according to the Agatston method, using 130 HU as the threshold, as previously described.12

Images were jointly analyzed and interpreted by an experienced board-certified nuclear medicine physician and cardiologist, who reached consensus. A visual semiquantitative 17-segment model with a 5-point scale (0: normal tracer uptake, 4: no tracer uptake) was used to calculate summed stress (SSS), rest (SRS), and difference score (SDS=SSS-SRS). Two endpoints were defined: small ischemia (SDS ≥ 2) and relevant ischemia (SDS ≥ 7, consistent with ≥ 10% ischemia of the left ventricular myocardium, as suggested in the guidelines on the criteria to consider invasive evaluation with subsequent revascularization).8

Normally distributed continuous variables are reported as the mean±standard deviation. Nonnormally distributed continuous variables are reported as the median [interquartile range]. Categorical variables are expressed as frequencies and percentages.

The number of missed ischemic findings for each of the 3 algorithms was compared with the reference standard. The difference in absolute risk of missed ischemia was used to calculate the number needed to scan (NNS) to detect 1 patient with a pathological finding for both endpoints (NNS=1/absolute risk difference).

Sensitivity, specificity, negative predictive value (NPV) and negative likelihood ratio (NLR) were calculated for each algorithm. The McNemar test was used to compare the proportion of normal scans with the reference standard. Reclassification among probability cutoffs by algorithms 2 and 3 was assessed using the net reclassification improvement. A P value<.05 was considered statistically significant.

Potential radiation and health care cost savings were estimated using the exact number of CACS and PET scans hypothetically performed in the corresponding algorithm (as outlined in figure 2, table 1 and table 2). The following in-house estimates were applied: CACS, 0.4 mSv and €416; PET, 2.0 mSv and €2,600. An exchange rate of €1.04=CHF 1 was used. Costs to detect an ischemic finding that would otherwise be missed under a given algorithm were calculated as follows: for algorithms 1 and 2 (cost of CACS+PET)×NNS; for algorithm 3, PET cost×NNS, as CACS was already incorporated for all patients.

Statistical analyses were performed using SPSS (V.28.0.1.0, IBM, United States) and RStudio (R V.4.2.2 and R packages TableOne, ggalluvial, PredictABEL) (R Core Team).

A total of 1792 patients were included, with a mean age of 65±11 years, and 43% were female. Typical and atypical angina were reported in 21% and 25% of patients, respectively. Median CACS was 74 [1-413]. Small ischemia was observed in 19.8%, and relevant ischemia in 9.3%. Baseline characteristics are shown in table 3.

The median RF-CL and CACS-CL PTP were 11% [6-19] and 12% [3-28], respectively. The distribution among risk categories is shown in table 3 and figure 3. By RF-CL, most patients (43.5%) were classified as having low likelihood (> 5%-15% PTP), and 22.0% were classified as having very low PTP (≤ 5%). None of the patients was classified as having high PTP (> 50%).

Figure 3.

Refinement of clinical likelihood using the CACS-CL model. Alluvial plot showing classification from RF-CL (left) to CACS-CL (right). CACS-CL, coronary artery calcium score-weighted clinical likelihood; RF-CL, risk factor-weighted clinical likelihood.

(0.24MB).

The CACS-CL model classified more patients (34.5%) as having very low likelihood (≤ 5%). The proportion of low PTP was less than half compared with the RF-CL model (19.7% vs 43.5%, P <.001). A total of 154 (8.6%) were classified as having high PTP (> 50%-85%).

Algorithm 1 did not involve patient reclassification. Algorithm 2 reclassified 612 patients (34.2% of the cohort) from low CL to very low (258, 33.1% from this PTP category), moderate (345, 44.2%) and high CL (9, 1.2%). One hundred sixty-eight (21.5%) patients remained in the low CL category. The overall proportion of low CL was reduced from 43.5% to 9.4%. Net reclassification improvement was in favor of algorithm 2: 0.2571 (95%CI, 0.1969-0.3173; P <.001) for small and 0.2331 (95%CI, 0.1582-0.308; P <.001) for relevant ischemia. Reclassification tables are depicted in tables S1,S2.

Algorithm 3 reclassified more patients (1004, 56.0%) to different likelihood categories: 571 (31.9%) were moved to higher and 433 (24.1%) to lower categories (figure 3). The proportion of patients in the very low likelihood category increased from 22.0% to 34.5%. Net reclassification improvement was in favor of algorithm 3 and higher compared with algorithm 2: 0.4954 (95%CI, 0.4173-0.5735; P <.001) for small and 0.5724 (95%CI, 0.4771-0.6677; P <.001) for relevant ischemia. Details are shown in tables S3,S4.

The overall performance, detailed test characteristics, potential radiation exposure, and health care cost reduction of the 3 algorithms are summarized in figure 2, figure 4, table 4 and table 2.

Figure 4.

Central illustration. Three different diagnostic algorithms based on the current recommendations of the 2024 ESC guidelines on chronic coronary syndromes were retrospectively applied. Sensitivity and NPV refer to the ability to detect or exclude small ischemia.82Rb, rubidium-82; CACS-CL, coronary artery calcium score-weighted clinical likelihood; ESC, European Society of Cardiology; NPV, negative predictive value; PET, positron emission tomography; RF-CL, risk factor-weighted clinical likelihood.

(0.16MB).

Algorithm 1 deferred the smallest proportions of patients (22.0%) from testing with CACS and PET. Overall, 1.5% of ischemic findings were missed. In patients deferred from testing, 4% and 3% had undetected small and relevant ischemia, respectively. The NNS for small and relevant ischemia were 69 and 163, respectively. Compared with the gold standard, sensitivity and NPV were good and excellent, respectively. Per patient, 0.3 mSv and €665 could have been saved.

Compared with algorithm 1, algorithm 2 additionally deferred 258 (14.4%) patients from testing, resulting in a 34.0% radiation and 34.5% cost reduction. Overall, there was a trend to missing more small ischemic findings (1.3% vs 0.8%, P=.18). The rate of undetected relevant ischemia was similar. In patients deferred from PET, 4% small and 2% relevant ischemia were not diagnosed. Sensitivity and NPV were lower than with algorithm 1. However, the NPV to exclude relevant ischemia was still high (99.3%). Potential radiation and cost reduction per patient were 0.8 mSv and €1039, respectively.

Algorithm 3 deferred 618 (34.5%) from PET. The proportion of missed diagnoses was lowest, in particular for relevant ischemia (1.1% small ischemia, 0.3% relevant ischemia). Patients deferred from PET had missed small and relevant ischemia in 3% and 1%, respectively. NNS was highest compared with the other algorithms (75 for small, 358 for relevant ischemia). With algorithm 3, 28.7% radiation and 29.7% costs could have been saved, which translates to a per patient reduction of 0.7 mSv and €897. Sensitivity and NPV were highest compared with the other algorithms.

The proportion of normal scan results was significantly reduced from 80.2% to 76.5%, 72.1% and 71.9% for algorithms 1, 2 and 3, respectively (P <.001 each).

Estimated costs to prevent missed diagnoses of small ischemia were €208 104, €150 000, and €195 000 for algorithms 1, 2 and 3, respectively. The costs to detect all patients with relevant ischemia were €491 608, €449 384, and €930 800 using algorithms 1, 2 and 3, respectively.

First, CAD was diagnosed using functional ischemia testing rather than anatomical modalities such as invasive coronary angiography or CTCA, which introduces a residual risk of endpoint misclassification. However, only a minority of angiographically significant luminal stenoses result in myocardial ischemia.14 Moreover, noninvasive functional tests are widely used in clinical practice to guide management and to exclude hemodynamically significant CAD.

Second, the median CACS in this cohort was 74 [1-413], suggesting that CTCA would have been feasible in most patients. However, referring physicians likely judged the CL to be higher, leading to direct referral for PET. Due to logistical constraints, it was not feasible to perform both CACS and CTCA in all patients, highlighting the importance of optimizing preselection with gatekeeper algorithms.

Third, despite prospective, consecutive patient enrolment, the generalizability of the proposed diagnostic algorithms is somewhat limited, as the results are derived from a single academic center and may reflect local referral patterns and preferences. Nevertheless, most patients were referred by cardiologists in private practice who generally adhere to guideline recommendations. In addition, PET is not uniformly available in all health care systems, which may further limit generalizability. However, as PET is among the most sensitive functional imaging modalities,24 the safety of the proposed algorithms is unlikely to be compromised if alternative functional tests, such as stress echocardiography or stress magnetic resonance imaging, are used instead.

Fourth, a substantial proportion of patients were asymptomatic yet were referred for advanced CAD testing. As demonstrated in the supplementary safety analysis restricted to symptomatic patients, these individuals had a higher burden of cardiovascular risk factors, comorbidities, and electrocardiographic abnormalities (table S5), which likely prompted referring cardiologists to choose functional ischemia testing rather than CTCA. Aside from significantly higher CACS values, no meaningful differences were observed in scan results, radiation or cost reduction, missed diagnoses, or algorithm performance when analyses were limited to symptomatic patients (tables S6-S8). Nevertheless, the inclusion of asymptomatic patients may have biased the findings toward lower pretest probability and higher rates of normal scans, potentially influencing the observed gatekeeper performance.

Fifth, the pretest probability in this cohort was low, which limits the effectiveness of the described algorithms. Moreover, according to the 2024 ESC guidelines, many of these patients would now be referred for CTCA as the initial test; however, referrals in this study were made prior to the publication of the current recommendations. Finally, as a center performing advanced cardiac imaging, we sought to evaluate and compare different gatekeeper strategies to optimize patient selection and resource allocation.

Sixth, only PET scans were analyzed in this cohort, although other functional ischemia tests that do not involve ionizing radiation, such as stress echocardiography and stress magnetic resonance imaging, are available. In theory, applying the proposed algorithms to these modalities could further enhance reductions in radiation exposure and costs; however, at present, only CT and PET are available as hybrid imaging techniques.

Seventh, the safety of the proposed gatekeeper algorithms was inferred from imaging results rather than from clinical outcome data. Definitive confirmation of safety would require prospective prognostic studies with hard clinical endpoints.