The study population consisted of all consecutive patients presenting with new-onset HFrEF of unknown etiology referred for ICA between January 2005 and 2019 and included in a local interventional cardiology database (Cardiobase™). Criteria for referral for ICA were at the discretion of the attending physician. The following exclusion criteria were applied: left ventricular ejection fraction (LVEF) >40%,1 severe valvular disease presumed to be the cause of HF, documented ischemia on non-invasive testing, previous known obstructive CAD (previous MI or coronary revascularization), previous cardiac surgery, and indication for ICA for other purposes. Patients with chest pain were included only if their primary diagnosis and clinical symptoms were predominantly of heart failure (HF).

This was a cross-sectional study performed at a single academic center. Demographic, clinical and angiographic data and vital status were prospectively collected in the ongoing ACROSS (Angiography and Coronary Revascularization On Santa cruz hoSpital) registry, which was approved by the local ethics committee, as previously published.9 All patients provided written informed consent and the study was approved by the ethics committee.

Obstructive CAD was defined as ≥70% luminal diameter stenosis by visual inspection in at least one coronary artery, or >50% in the left main.10

Two subgroups of patients were defined according to the angiographic findings: with obstructive CAD and without obstructive CAD. Logistic regression was used to identify independent clinical predictors of obstructive CAD. The beta regression coefficients of the selected variables were then used to develop a clinical model that estimates the probability of obstructive CAD for a given patient.

Normally and non-normally distributed variables were expressed as mean±standard deviation and median and interquartile range, respectively.

The following variables were assessed for their association with obstructive CAD: age, gender, hypertension, diabetes, hypercholesterolemia, cigarette smoking, family history of CAD, chronic kidney disease, peripheral arterial disease (PAD), LVEF, anginal chest pain, and the presence of regional wall motion abnormalities on two-dimensional echocardiography. Univariate analysis was performed using univariate logistic regression.

In order to derive a predictive model for the presence of obstructive CAD, variables with p<0.05 in the univariate analysis were entered simultaneously in a multivariate logistic regression model performed with bootstrap sampling with replacement using 1000 samples. Bootstrapping was used for internal validation of the model and the 95% confidence intervals (CI) of the odds ratios were estimated using bias corrected accelerated bootstrapping.

To create a practical risk score, each of the independent predictors was assigned a number of weighted points proportional to its beta regression coefficient value, rounded to the nearest integer, yielding the multiplying coefficient for each variable in the score. The resulting score, defined as the sum of the points assigned to each weighted variable, was calculated for every patient. The population was then categorized according to the score into low- (predicted probability <15%), medium- (predicted probability 15–75%) and high-risk strata (predicted probability >75%) for the presence of obstructive CAD.

The performance of the prediction risk score was assessed using model discrimination and calibration. Model discrimination (i.e. the extent to which the model can differentiate between those who have and those who do not have obstructive CAD) was summarized using the area under the curve (AUC) on receiver operating characteristic analysis. The AUC can range from 0.5 to 1.0, with higher values indicating improved ability to predict or identify the endpoint of interest (in this case, the presence of obstructive CAD). Model calibration (reflecting how close predicted obstructive CAD prevalence rates are to the actual observed rates) was assessed using the Hosmer–Lemeshow goodness-of-fit test. A statistically significant p-value from this test would reject the null hypothesis that the features in the model fit the data well.

The statistical analysis was performed using IBM SPSS version 25.0 (IBM SPSS Inc., Chicago, IL, USA) for Mac OS. Statistical significance was set at p<0.05 (two-sided).

The results of univariate and multivariate logistic regression analysis are presented in Tables 2 and 3, respectively. Eight variables associated with the presence of obstructive CAD were identified in univariate analysis and entered into a multivariate model, which was simplified by removing age and hypertension, as these were not significantly associated with obstructive CAD. The AUC of the simplified model was 0.872 (95% CI 0.834–0.909). This model is summarized in Table 3. The p-value from the goodness-of-fit test was 0.333, indicating that the simplified model fit the data well.

The parameter estimates from the simplified model were used to develop a risk score to predict obstructive CAD. Each of the variables included in the simplified model was assigned a number of points proportional to its beta coefficient, i.e. the natural logarithm of the odds ratio point estimate for each variable in the model. The resulting scores assigned to each feature are presented in Table 4.

Individual calculated scores ranged from 0 to 9. The internally validated AUC for the score was 0.872 (95% CI 0.834–0.909), as depicted in Figure 2. Probabilities of obstructive CAD predicted by the score are summarized in Table 4 and ranged from 0.12 to 1.00.

Figure 2.

Receiver operating characteristic curve for the clinical score. AUC: area under the curve; CI: confidence interval.

(0.11MB).

Three risk categories for the presence of obstructive CAD were defined according to the probability of CAD predicted by the score: low (score ≤3 points; predicted probability 1.21–13.0%), intermediate (4–5 points; predicted probability 34.5–55.4%) and high risk (≥6 points; predicted probability: 75.0–100.0%). The distribution of patients in our population by risk category according to the presence or absence of obstructive CAD, and the probability of obstructive CAD predicted by the risk score, are illustrated in Figures 3 and 4, respectively. These three risk categories (high, intermediate and low) represent 63.3%, 25.0% and 11.7% of the total population, respectively, and may help the clinician in decision-making for the selection of the most appropriate diagnostic test according to the predicted probability.

Figure 3.

Distribution of the population by risk category according to the presence or absence of obstructive coronary artery disease. CAD: coronary artery disease.

(0.18MB).

Figure 4.

Distribution of the predicted probability by risk score.

(0.11MB).

The present investigation has limitations that should be acknowledged. First, this was a retrospective analysis with the limitations and biases inherent to its design. Also, it relies on the quality of registry data, another limitation that is universally inherent to this type of study design. However, we provide an analysis that is based on common and easily reported variables and is thus potentially less prone to uncertainty. Furthermore, it was a single-center study, although from a large academic referral center, reflecting the diagnostic approach of multiple secondary centers. The indication for ICA was left at the referring physician's discretion and the echocardiographic findings were not reviewed by a core lab.

The exclusion of patients with acute symptoms and a history of MI, percutaneous coronary intervention or coronary artery bypass grafting may have increased the prevalence of non-CAD HFrEF. However, this might be regarded as a strength of our study, as it addresses the clinical question of the best strategy to exclude CAD as a cause of HFrEF in patients without an apparent etiology.

Demonstration of obstructive CAD as assessed by ICA does not necessarily imply causation, as it may be a concomitant bystander rather than the main etiology of HFrEF. The fact that the majority of patients with obstructive CAD in our population had single-vessel disease is further evidence that the prevalence of severe coronary disease in patients presenting with new-onset HFrEF without a previous history of ACS, angina or ischemia in non-invasive assessment is low, and that obstructive CAD as the main etiology for HFrEF is generally secondary to sequelae from acute coronary events, since HFrEF is a relatively uncommon form of presentation of obstructive CAD. This being so, the presence of obstructive CAD in HFrEF patients may not have a direct impact on prognosis. In addition, even though our score might be less specific for severe obstructive CAD (which has a higher probability of a causal relationship with left ventricular systolic dysfunction in the individual patient), it showed a high sensitivity for this subgroup, demonstrating the clinical safety of its use. This should be taken into account and highlights the importance of the appropriate selection of patients who should undergo ICA.

Although previously reported by Doukky et al.17 as a predictor of obstructive CAD in patients with HFrEF, electrocardiographic data were not included in the present investigation due to a high percentage of missing data, precluding proper statistical analysis.

Importantly, our model was not validated in a separate dataset, which precludes conclusions about its external validity. As such, further studies are needed to validate this score in other HFrEF cohorts and to confirm its value as a generalizable clinical prediction tool.