Gender Differences in Fractional Flow Reserve Derived from Machine Learning Computed Tomography

Study population

Our study population consisted of 471 consecutive patients who underwent clinically indicated CCTA and positron emission tomography (SPECT) for suspected CAD between January 1, 2016 and June 22, 2020 at a large referral center.

This patient cohort was originally established to assess the comparative prognostic role of functional assessment with SPECT vs ML-FFRCT when added to the CCTA anatomical assessment, which we have published before14. The dual SPECT-CCTA test was performed at the discretion of the treating physician. In the current sub-analysis, we present sex-specific differences. A flowchart is provided (Supplementary Fig. 1).

Approval from the Institutional Review Board of the Houston Methodist Academic Institute was obtained prior to the start of the study. Informed consent was withdrawn by the Houston Methodist Academic Institute Institutional Review Board due to the retrospective nature of the study. All methods were performed in accordance with current guidelines and regulations.

Evaluation of covariates

Information on sociodemographic variables, medical history, comorbidities, and medication use was obtained through electronic health record review within 30 days of imaging.

Follow-up and result

The primary endpoint was death or non-fatal myocardial infarction (ID/MI). The secondary outcome included major adverse cardiovascular events (MACE), a composite of death, MI, and unplanned revascularization (PCI or CAP occurring more than 90 days after index imaging). Myocardial infarction has been defined as the 4th universal definition of myocardial infarction15. All results were obtained from a chart review and assessed by blinded expert physicians. All patients were followed from the date of CCTA imaging until the results appeared or the last known date of contact noted in the patient records.


CCTA scans were performed using the 3rd generation SOMATOM FORCE scanner (Siemens, Forchheim, Germany). Image acquisition was performed according to the guidelines of the Society of Cardiovascular Computed Tomography (SCCT) and has been described previously.14.16. Briefly, intravenous metoprolol was given to patients with a heart rate ≥ 65 beats/min and sublingual nitroglycerin 0.4 mg was given immediately before image acquisition. During image acquisition, 60 to 100 cc of contrast medium was injected, followed by saline solution. Axial scans were obtained with prospective electrocardiographic gating. Image acquisition was performed to include the coronary arteries, left ventricle, and proximal ascending aorta.

Images were analyzed with a three-dimensional workstation using one of several post-processing methods, including axial and multiplanar reformatting, maximum intensity projection, and cross-sectional analysis. Lesion type and location were assessed visually using an 18-segment model according to SCCT guidelines16. In each segment, atherosclerosis was defined as tissue structures > 1 mm2 within the lumen of the coronary artery or adjacent to the lumen which could be distinguished from pericardial tissue, epicardial fat, or the lumen of the vessel itself. Coronary stenosis was classified as none (0%), mild (1–49%), moderate (50–69%), or severe (≥70%) based on the degree of luminal diameter narrowing. Anatomically obstructive CAD by CCTA was defined as 70% stenosis severity in non-left main vessels and 50% in the left main artery. Results were reported using the SCCT Coronary Artery Disease Reporting & Data System (CAD-RADS). All interpretations were made by cardiologists experienced in imaging (at least 10 years of experience).


ML-FFRCT was determined using a fractional flow reserve machine learning based calculation (ML-FFRCT, cFFR 3.2, Siemens Healthcare GmbH, Forchheim, Germany, not available for commercial use). Methods for ML-FFRCT determination have already been described14. The coronary tree was isolated semi-automatically to generate a three-dimensional coronary model. All vessels and branches with a diameter of at least 2 mm were included. ML-FFRCT was determined in the middle of a segment for normal segments and 1 cm downstream of stenosis for segments with lesions based on previous work showing a higher prognostic role of measurements downstream of stenosis17. ML-FFR Derivation DetailsCT values, diagnostic accuracy and concordance of results with invasive ML-FFRCT have been previously reported12.18. A three-dimensional quantitative model of the coronary tree was reconstructed and a 17-segment model was used. The image processing was carried out by two investigators unaware of the results of other tests. Both investigators underwent extensive training prior to image processing. The reproducibility of the results was assessed on a random subset of patients and found to be high (ICC 0.981 per patient and 0.970 per segment, absolute mean difference 0.019 per and 0.027 per segment) and on the quality of the image and the degrees of stenosis by CAD-RADS. The minimum ML-FFRCT values ​​were recorded in each segment per patient and summarized using the median (interquartile range) and categorized. Hemodynamically significant ischemia was defined as ML-FFRCT 19,20,21.

statistical analyzes

Continuous variables were presented as mean (standard deviation)/median (interquartile range) and categorical variables were presented as number (percentage) stratified by sex. Results were compared using Student’s t-test for normally distributed continuous variables and median test for non-normally distributed continuous variables or chi-square test for categorical variables.

Median (IQR) and minimum ML-FFR per patientCT were calculated for each CAD-RAD category (0, 1–2, 3, 4A, and 4B) per patient and for each coronary vessel and stratified by gender to unmask potential differences by severity of stenosis. The prevalence of ML-FFRCT

Multivariate adjusted Cox proportional hazards models were used to investigate the association between ML-FFRCT 2). In sensitivity analyses, we used the median ML-FFR per patientCT and lower thresholds (ML-FFR significantCTdefined as

To determine if there should be separate thresholds for ML-FFRCTfor males versus females, mean ML-FFRCTfor each coronary vessel, the proximal, midline, and distal segments were summarized for the overall study cohort and stratified by sex. The difference between the segments was assessed using the Kruskal-Wallis equality of proportions rank test.

All analyzes were performed using Stata 17.0 (StataCorp, College Station, TX) and a p-value

Comments are closed.