Coronary Calcium Scoring Improves Risk Prediction in Patients With Suspected Obstructive Coronary Artery Disease
Key result
Adding CACS to clinical likelihood improves MI and death prediction over standard PTP to 0.69.
- C-statistic 0.69
- 95% CI 0.67-0.70
- n=45,129
Why the study?
While RF-CL and CACS-CL models improve identification of obstructive CAD compared to basic PTP models, their prognostic value for predicting myocardial infarction and death needed to be assessed.
Do the RF-CL and CACS-CL models improve risk prediction for myocardial infarction and death compared to the basic PTP model in patients with suspected obstructive CAD?
Key Points
- To assess the prognostic value of the risk factors-weighted clinical likelihood (RF-CL) and coronary artery calcium score-weighted clinical likelihood (CACS-CL) models compared with standard pre-test probability (PTP) models in patients with suspected obstructive coronary artery disease.
- Analyzed symptomatic patients referred for diagnostic testing due to clinical indications from a Danish register (n=41,177) and a North American randomized study (n=3,952), with a median follow-up of 3.6 [1.9 – 5.6] years.
- Stratified the incidence of myocardial infarction and death across likelihood categories of obstructive CAD (very low [≤5%], low [5–≤15%], moderate [15–≤50%], and high [>50%]) using PTP, RF-CL, and CACS-CL models.
- Both RF-CL (45%, n=21,120) and CACS-CL (60%, n=26,363) substantially down-reclassified patients to a likelihood ≤5% compared to PTP (18%, n=8,419), while annualized event rates of myocardial infarction and death remained low: 0.51 (0.46–0.56) for RF-CL, 0.48 (0.44–0.56) for CACS-CL, and 0.37 (0.31–0.44) for PTP.
- Predictive power using Harrell′s C-statistics showed superiority for RF-CL (0.64 (0.63–0.65)) and CACS-CL (0.69 (0.67–0.70)) over the PTP model (0.61 (0.60–0.62)).
- Unadjusted hazard ratios with a binary cutoff of 5% (>5% vs ≤5%) were 3.08 (CI 2.58–3.69) for PTP, 2.77 (CI 2.48–3.09) for RF-CL, and 3.53 (CI 3.18–3.92) for CACS-CL.
Study Design
Type
Cohort (n=45,129)
Multicenter
Yes
Structured PICO
Do the RF-CL and CACS-CL models improve risk prediction for myocardial infarction and death compared to the basic PTP model in patients with suspected obstructive CAD?
P
Population
45,129 symptomatic patients without previously diagnosed CAD referred for non-invasive testing, followed for a median of 3.6 years.
E
Exposure
Risk factors-weighted clinical likelihood (RF-CL) model and coronary artery calcium score-weighted clinical likelihood (CACS-CL) model for risk stratification.
C
Comparator
Basic pre-test probability (PTP) model based on the 2019 ESC guidelines.
O
Outcome
Incidence of all-cause death or non-fatal myocardial infarction.hard clinical
Main Result
Effect estimate: C-statistic 0.69 (95% CI 0.67-0.70)
Absolute Event Rate: 0.69% vs 0.61%
Limitations
- Retrospective analysis of existing data sets
- Potential selection bias as patients had to be eligible for coronary CTA (e.g., severe kidney disease or obesity may be under-represented)
- Predominantly Caucasian patients from Europe and North America, limiting generalizability to other ethnicities
- Low number of events in the PROMISE study cohort
- Selection bias (patients had to be eligible for coronary CTA, so severe kidney disease or severe obesity may be under-represented)
- Low numbers of events in the PROMISE study
Key expert perspectives
Captured external expert commentary on this paper, strongest first. Original sources are linked where available.
“Despite that we are down-classifying a lot of patients, the patients which are down-classified for not testing have an excellent prognosis. Our conclusion is that this seems safe, but of course we need to test this in a randomized setup. But this is very encouraging for the models and [their] application in the future.”
ODOmar DzayePhysician-scientist, Johns Hopkins University School of Medicine
“These models are great in terms of combining data with risk factors like family history for coronary artery disease, dyslipidemia, hypertension, diabetes, and smoking, and if they now include also coronary artery calcium, they definitely exhibit better predictive accuracy than models based only on age, sex, and symptoms like the pretest probability test.”
Implication
The RF-CL and CACS-CL models improve prognostic risk stratification for myocardial infarction and death compared to the standard PTP model, safely identifying more patients who can be deferred from further diagnostic testing.
Abstract
Background: In patients suspect of obstructive coronary artery disease (CAD), the risk factors-weighted clinical likelihood (RF-CL) model and the coronary artery calcium score-weighted clinical likelihood (CACS-CL) model improves the identification of obstructive CAD as compared to basic pre-test probability (PTP) models. Objectives: The aim of this study was to assess the prognostic value of the new models. Methods: The incidence of myocardial infarction and death were stratified according to categories by the RF-CL and CACS-CL and compared to categories by the PTP model. We used cohorts from a Danish register (n=41,177) and a North American randomized study (n=3,952). All patients were symptomatic and referred for diagnostic testing due to clinical indications. Results: Despite substantial down-reclassification of patients to a likelihood ≤5% of CAD with either the RF-CL (45%) or CACS-CL (60%) models compared to the PTP (18%), the annualized event rates of myocardial infarction and death were low using all three models; RF-CL 0.51 (0.46–0.56), CACS-CL 0.48 (0.44–0.56) and PTP 0.37 (0.31–0.44), respectively. Overall, comparison of the predictive power of the three models using Harrell′s C-statistics demonstrated superiority of the RF-CL (0.64 (0.63–0.65)) and CACS-CL (0.69 (0.67–0.70)) as compared to the PTP model (0.61 (0.60–0.62)). Conclusions: The simple clinical likelihood models which include classical risk factors or risk factors combined with CACS provide improved risk stratification for myocardial infarction and death compared to the standard PTP model. Hence, the optimized RF-CL and CACS-CL models identifies 2.5 and 3.3 time more patients, respectively, who may not benefit from further diagnostic testing.
Discussion
No takes yet. Share an insight, caveat, or question.
Cite This Study
Winther et al. (2022) conducted a cohort in Suspected obstructive coronary artery disease (n=45,129). CACS-weighted clinical likelihood (CACS-CL) model vs. Pre-test probability (PTP) model was evaluated on Predictive power for myocardial infarction and death (Harrell's C-statistic) (C-statistic 0.69, 95% CI 0.67-0.70).