Consecutive treatment-naive patients diagnosed with CTEPH who had CMR and right heart catheterization (RHC) up to February 2017 were prospectively recorded in hospital databases as part of the ASPIRE (Assessing the Spectrum of Pulmonary Hypertension Identified at a Referral Center) registry as previously described (23, 24) (ClinicalTrials.gov identifier NCT02565030). Patients' demographics, imaging, and clinical metrics with follow-up data were prospectively collected using a census date of April 29th 2019. CMR and RHC metrics, exercise test, pulmonary function, and treatment were included.

The diagnosis of CTEPH required patients to have undergone RHC with a measured mean pulmonary artery pressure (mPAP) ≥ 25 mmHg at rest and at least one segmental perfusion defect on perfusion lung Q scan or pulmonary artery (PA) obstruction seen by multidetector computed tomography pulmonary angiography (CTPA) or conventional pulmonary angiography with other causes of PH excluded (9).

Following multi-modality imaging patients were discussed at a multi-professional meeting. Patients with CTEPH were subsequently discussed at the national pulmonary endarterectomy MDT at Papworth Hospital, Cambridge where surgical accessibility and suitability for surgery were assessed.

All patients were followed up until the date of death or census date. Ethical approval for this study was granted by our institutional review board (ref c06/Q2308/8). This study was funded by grants from the Wellcome Trust (A.J.S.). The funding body was not involved in the study design or data interpretation.

CMR was performed using an eight channel cardiac coil on a GE HDx (GE Healthcare, Milwaukee, WI) whole body scanner at 1.5 T, as previously described (15). Short-axis cine images were acquired using a cardiac gated multislice balanced SSFP sequence (20 frames per cardiac cycle; slice thickness, 8 mm; field of view, 48 cm; matrix, 2563256; BW, 125 kHz/pixel; TR/TE, 3.7/1.6 ms). A stack of images in the short-axis plane with slice thickness of 8 mm (2-mm inter-slice gap) were acquired fully covering both ventricles from base to apex. End-systole was considered to be the smallest cavity area. End-diastole was defined as the first cine phase of the R-wave triggered acquisition or largest volume. Through plane phase contrast imaging was performed orthogonal to the main pulmonary trunk. Phase contrast imaging parameters were as follows: repetition time, TR 5.6 ms; echo time, TE 2.7 ms; slice thickness, 10 mm; field of view, 48 cm, bandwidth, 62.5 kHz; matrix, 256 3128; 20 reconstructed cardiac phases; and velocity encoding of flow, 150 cm/s. Patients were in the supine position with a surface coil and with retrospective ECG gating.

Image analysis was performed on a GE Advantage Workstation 4.1 with the observer blinded to the patient clinical information, and cardiac catheter parameters. Right and left endocardial and epicardial surfaces were manually traced from the stack of short-axis cine images, using proprietary MR workstation software to obtain RV end-diastolic volume (RVEDV) and RV end-systolic volume (RVESV), and left ventricular (LV) end-diastolic volume (LVEDV) and LV end-systolic volume (LVESV). From end-diastolic and end-systolic volumes, RVEF and LV ejection fraction (LVEF) and RV and LV stroke volumes (SV) were calculated. With the exception of RVEF and LVEF, these measurements were all corrected for age and sex (%pred) based on previously published referenced data (25) and they were indexed for body surface area based on Mostellar formula (26). Based on previous work, SV was considered to be the most accurate from LV volumetry (27) and was used for MRI estimation of RV–PA coupling. For calculation of ventricular mass, the interventricular septum was considered as part of the LV. RV end-diastolic mass (RVEDM) and LV end-diastolic mass (LVEDM) were derived. Ventricular mass index (VMI) was defined as RV mass divided by LV mass, as previously described (28). Maximal and minimal PA areas were measured, and relative area change was defined by the following equation: relative area change = (maximum area – minimum area)/minimum area (29). Reproducibility for CMR measurements in our center has been described previously (15).

RHC was performed using a balloon-tipped 7.5F thermodilution catheter (Becton-Dickinson, Franklin Lakes, NJ). RHC was usually performed via the internal jugular vein using a Swan-Ganz catheter. Measurements of right atrial pressure, pulmonary arterial wedge pressure (PAWP), cardiac output (CO) using thermodilution and mixed venous oxygen saturation were also made. Pulmonary vascular resistance (PVR) was calculated as [(mPAP-PAWP) / CO] / 80 and expressed as dyne.s.cm⁻⁵.

As previously described (15, 30–32), RV elastance (Ees) was estimated as mPAP divided by RVESV. PA elastance (Ea) was estimated using mPAP-PAWP divided by LVSV. Ees/Ea by a combined RHC and CMR approach was defined as follows: (mPAP/RVESV)/[(mPAP-PAWP)/LVSV]. CMR estimated Ees/Ea was defined by LVSV/RVESV. Distensibility, a measurement of PA stiffness, was defined as PA relative area change divided by pulse pressure.

Continuous variables were expressed as mean (SD) for parametric variables and median (interquartile range) for non-parametric variables. Categorical data was presented as the number of subjects and percentage. Continuous variables were compared using independent sample t-test. Categorical variables were compared using Pearson's Chi-square test. Survival analysis was conducted using Kaplan-Meier plots and survival was compared using the log-rank test. Survival was calculated from date of diagnosis to date of death or census date to compare survival between the PEA and non-PEA groups. To assess the prognostic value of CMR metrics, survival was calculated from date of CMR to date of death or census date. Univariate Cox proportional hazard regression analysis was used to assess the prognostic value of CMR metrics in terms of biventricular volume, function, mass and PA stiffness metrics in the whole cohort, PEA and non-PEA groups. Hazard ratios (HR) generated from univariate analysis were scaled by dividing the actual individual value of the variable by SD. This is to allow direct comparison of HR. Multivariate Cox proportional hazard regression analysis was performed for predicted CMR variables / MRI coupling measurement, clinical variables and combined CMR and clinical variables. Selected variables were entered into multivariate models if they were reported in literature as predictors of mortality (33, 34), ≤ 10% missing values and a p < 0.200 at univariate analysis. Multivariate models for the whole cohort, PEA and non-PEA groups were generated. To overcome multicollinearity, highly correlated variables (r > 0.80) were entered separately in the models. Receiver operating characteristics (ROC) analysis was used to assess the prognostic significance of the CMR, clinical and combined CMR and clinical models generated on the whole, PEA and non-PEA cohorts and presented with area under the curve (AUC). In order to evaluate the stability (internal validation) of the CMR prognostic model of the whole cohort, PEA group and non-PEA groups, the bootstrap approach with 1,000 bootstrap samples (default settings) was performed (35). In the bootstrap, simple sampling method with bias-corrected and accelerated 95% confidence interval (95% C.I.) type were selected. Locally estimated scatterplot smoothing regression analysis (LOESS) was performed for 1 year mortality, where significance was demonstrated at multivariate analysis, for patients undergoing and not undergoing PEA. All statistical tests were two-sided and a p-value of < 0.050 was considered statistically significant. A Statistical Package for the Social Sciences Program (SPSS) version 26 for Windows (SPSS Inc. Chicago, IL) was used for statistical analysis and for presentation of data GraphPad Prism 8.3.0 (GraphPad Software, San Diego, CA) was used.

A total of 375 patients (mean+/-SD age 64+/- 14 years, 185 (49%) females) with CTEPH were included in the analysis. One hundred and eighty-one (48%) patients had PEA. Table 1 shows demographic, RHC, and lung function results and Table 2 CMR, RV-PA coupling metrics and pulmonary stiffness data, for the whole cohort and for the PEA and non-PEA groups. The mean+/- SD time delay between RHC and CMR was 45+/-15 days.

Compared with the non-PEA group, patients in the PEA group were younger (p < 0.001), had a higher incremental shuttle walking distance (ISWD) (p = 0.001), lower RVESVI%pred (p=0.003), lower RVEDMI%pred (p = 0.046), higher forced expiratory volume at 1 min (FEV1) (p < 0.001) and higher forced vital capacity (FVC) (p = 0.003).

During the follow up period, 104 (28%) patients died. The median overall survival for the whole cohort was 143 months. Median survival of patients undergoing PEA was higher than patients not undergoing surgery (146 vs. 97 months, 95% C.I. (121–162); p < 0.001). Patients' demographics, CMR, RHC, lung function tests and pulmonary arterial stiffness data between survivors and non-survivors is summarized in Supplementary Table 1. Survivors were younger (p < 0.001) with a higher percentage of females (p < 0.001), had better lung function (p < 0.001), exercise capacity and less likely to have a history of malignancy, coronary artery disease, chronic kidney disease or COPD.

Univariate Cox proportional hazard regression analysis for CMR metrics and clinical variables for the whole cohort is presented in Supplementary Table 2. Only variables with p < 0.200 are shown. CMR measures of RV size and function [RVEDVI%pred (p = 0.002), RVESVI%pred (p < 0.001), and RVEF%pred (p < 0.001)], LV function [LVSVI%pred (p = 0.001) and LVEF%pred (p = 0.011)], LA volume [LAVI (p = 0.010)] and invasive (p = 0.027) and non-invasive CMR-derived Ea/Ees ratio (p = 0.001), were significant predictors of mortality at univariate Cox regression analysis for the whole cohort. Separate univariate Cox regression analysis was performed for PEA and non-PEA groups and is shown in Tables 3, 4, respectively.

Multivariate Cox proportional hazard regression analysis models of only CMR metrics, clinical variables and CMR metrics and clinical variables (p < 0.200) at univariate analysis are presented in Table 5. In the whole cohort, higher RVESVI%pred (p = 0.007), lower LVSVI%pred (p = 0.001) and higher LAVI (p = 0.001) were independent predictors of increased mortality in the CMR metrics model. The regression equation for the multivariate Cox regression model in the whole cohort is as follows: Expected hazard = (RVESV%pred^*0.002)-(LVSVI%pred^*0.019) + (LAVI^*0.013). In the PEA group, lower LVSVI%pred (p = 0.035) and higher LAVI (p < 0.001) (Expected hazard = (LAVI^*0.030) - (LVSVI%pred^*0.018) and in the non-PEA group, lower RVEF%pred (p = 0.015) and lower LVSVI%pred (p = 0.024) were independent predictors of increased mortality [Expected hazard = (RVEF%pred^*-0.167) - (LVSVI%pred^*0.014)]. In the combined model, lower LVSVI%pred (p = 0.040) and higher LAVI (p = 0.033) in the whole cohort, higher LAVI (p = 0.009) in the PEA group and lower RVEF%pred (p = 0.040) in the non-PEA group were predictors of increased mortality.

Figure 1 shows Kaplan-Meier survival analysis of the PEA group (LVSVI%pred and LAVI) and non-PEA group (LVSVI%pred and RVEF%pred). LAVI is plotted above and below a value of 41 ml/m² based on previous work and for other values based on median (36). Figure 2 shows LOESS regression analysis for risk of 1 year mortality for LVSVI%pred and LAVI for patients undergoing PEA and for LVSVI%pred and RVEF%pred for the Non-PEA group.

The accuracy of the multivariate Cox models for the whole, PEA and non-PEA cohorts was tested using ROC analysis and presented as AUC. The prognostic accuracy of the CMR model in the whole cohort (AUC, 0.69, p = 0.001), clinical model (AUC, 0.62, p < 0.001) and in the combined CMR and clinical model (AUC, 0.65, p < 0.001). Sensitivity and specificity of the CMR whole cohort model was 71% (95% CI, 65–76) and 62% (95% CI, 52–72) respectively, clinical model 78% (95% CI, 69–86) and 43% (95% CI, 37–50) and for the combined clinical and CMR model 70% (95% CI, 63–76) and 55% (95% CI, 45–65), respectively (Figure 3A). In the PEA cohort, the prognostic accuracy of the CMR model (AUC, 0.73, p < 0.001), clinical model (AUC, 0.69, p = 0.002) and in the combined clinical and CMR model (AUC, 0.76, p < 0.001). Sensitivity and specificity of the CMR PEA model 80% (95% CI, 61–91) and 53% (95% CI, 46–62), respectively, clinical model 72% (95% CI, 65–79) and 65% (95% CI, 46–79) and for the combined model 80% (95% CI, 61–91) and 53% (95% CI, 46–62), respectively (Figure 3B). The prognostic accuracy of the CMR model in the non-PEA group (AUC, 0.63, p = 0.003), clinical model (AUC, 0.63, p = 0.003) and for the combined model (AUC, 0.56, p = 0.030). The sensitivity and specificity were 65% (95% CI, 56–73) and 60% (95% CI, 50–71), respectively for the CMR model, 60% (95% CI, 48–70) and 58% (95% CI, 50–67), respectively for the clinical model and 60% (95% CI, 48–70) and 58% (95% CI, 50–67), respectively for the combined model (Figure 3C).

Internal validation was performed using the bootstrap method as previously described. Supplementary Table 3 summarizes the results of the multivariate Cox regression model using the bootstrap method for the whole cohort CMR model. Variables remained significant with no bias after performing 1,000 bootstraps using bias corrected and accelerated confidence intervals. Similarly, variables for CMR prognostic model for the PEA and non-PEA groups remained significant after the bootstrap method (All p < 0.050).

This was a single center study with retrospective analysis of prospectively collected data which might have introduced selection or misclassification bias. CMR metrics were corrected for age and sex but not for race, which impacts on CMR metrics (53). Despite identifying LAVI as an independent predictor of outcome in the whole cohort and patients undergoing PEA further work is required to identify thresholds to aid clinical decision making. This study also used CMR as a gold standard technique to assess cardiac volumetric and functional data rather than echocardiography and echocardiography may be able to provide comparative data although this is not answered by this study. Balloon pulmonary angioplasty has only recently been available in the UK and we were unable therefore to provide any data regarding the potential for CMR to assess for outcomes in patients undergoing this intervention.