Defining left ventricular remodeling following acute ST-segment elevation myocardial infarction using cardiovascular magnetic resonance

Analysis was performed using CVI42 software (Version 5.2.2, Calgary, Canada). Forty STEMI patients reperfused by PPCI, with paired acute and follow-up scans from one of the cohorts reported recently [19‐21] were used for inter and intra-observer variability. Semi-automated contours were drawn on the short-axis cine images using the threshold segmentation option for the epicardial border and the automatic detection option for the endocardial border, with manual adjustment when required. The LVEDV, LVESV, LV mass (LVM) and LVEF were quantified using 2 methods as shown in Fig. 1. In method 1, we used rounded endocardial contours and excluded the trabeculae and papillary muscles (T&P) as part of the LVM and they were included as part of the LV volume. In method 2, the T&P were included as part of the LVM and they were excluded from the LV volume. The basal cine slice was included if at least 50% of the cavity circumference was surrounded by ventricular myocardium and this principle was used for both end-systole and end-diastole. %ΔLVEDV, %ΔLVESV, %ΔLVM and %ΔLVEF were calculated as the difference between the follow-up parameters and the corresponding baseline parameters and expressed as a percentage of the baseline parameters. All 40 acute and matching follow-up scans were analysed by 2 experienced CMR operators (twice by HB, 3 years’ experience in CMR, at least 2 months apart and blinded to previous results, and once by YYG, 1 and a half years’ experience in CMR).

Patient level data were obtained from our 2 previously reported cohorts [18, 19], and from 2 previous randomized controlled trials [16, 17] as listed in the Additional file 1: Online appendix Table 1. Only patients with paired acute and follow-up CMR scans were included. The LV parameters reported by the original studies were used for analysis. All cines were acquired using steady-state free precession-based cines as previously described in their respective publications [16‐19]. The CMR details for the acute MI size and microvascular obstruction (MVO) assessment by the 4 different cohorts of patients included are summarized in the Additional file 1: Online appendix Table 1.

Statistical analysis was performed using SPSS version 22 (IBM Corporation, Illinois, US). Normality was assessed using Shapiro-Wilk Test. Continuous data was expressed as mean ± standard deviation (SD) or median (interquartile range) and categorical data was reported as frequencies and percentages. Groups were compared using paired Student t test/Wilcoxon signed rank test or unpaired Student t test/Mann Whitney U test where appropriate. One-way analysis of variance was used to obtain the mean squared error for each LV parameter for inter and intra-observer measurements and their corresponding square root provided their standard error of the measurement (SEM). The 95% confidence interval (CI) for each SEM was calculated as previously described [22]. Coefficient of variation (CoV) was expressed as the standard deviation of the difference divided by the mean and expressed as a percentage and Levene’s test for homogeneity of variance was used to compare CoV between the two methods used for LV parameters quantification (T&P being part of the LV mass or LV volume). Bland-Altman analysis was performed for inter and intra-observer measurements of the LV parameters for comparison. The MDCs with 95% confidence (MDC95) for intra and inter-observer measurements for %ΔLVEDV, %ΔLVESV, % LVM and %ΔLVEF was calculated as 1.96 x SEM x square root of 2. ROC curve analysis was performed to predict an LVEF of <50% at follow-up to identify clinically significant cut-off values for %ΔLVEDV and %ΔLVESV. All statistical tests were two-tailed, and P < 0.05 was considered statistically significant.

Table 2 summarizes the SEM (95%CI), CoV and Bland-Altman analysis of the LV parameters divided into acute and follow-up scans and quantification method. Comparison of CoV did not show any statistical difference for inter-observer or intra-observer measurements (LVEDV, LVESV, LVM, LVEF) on both the acute or follow-up scans between both LV quantification methods (T&P included as part of the LV volume or LV mass) (P values between 0.15 and 0.97).

The LVEDV and LVESV were significantly higher and the LVM and LVEF were significantly lower both on the acute and follow-up scans when the T&P were included as part of the LV volume as shown in Fig. 2. When they were included as part of the LVM, they contributed the same extent to the LV mass on the acute and the follow-up scans (12.9 ± 5.1 and 11.4 ± 6.3% respectively, P = 0.17).

Details on the intra-observer and inter-observer measurements for %ΔLVEDV, %ΔLVESV, %ΔLVM and %ΔLVEF, both for when T&P was included as part of either LV volume or LV mass, are provided in Table 3. The MDC95 values for these LV parameters were similar for inter- and intra-observer measurements, and whether the T&P were included as part of the LV volume or mass.

Irrespective of how the T&P were dealt with, the highest MDC95 was 11% for %ΔLVEDV and 10% for %ΔLVESV for intra-observer measurements. The corresponding values for inter-observer measurements were 12% for both %ΔLVEDV and %ΔLVESV. Further details for %ΔLVM and %ΔLVEF are provided in Table 4.

A total of 146 STEMI patients had matching acute (mean of 4 ± 2 days) and follow-up CMR scans (median of 4 (4–5) months). 12/146 (8%) patients had their scans on a 3 T scanner and the rest were acquired on 1.5 T scanners. Table 5 summarizes the clinical and CMR details of these 146 patients.

ROC curve analysis showed that %ΔLVESV was a better predictor of LVEF of <50% at follow-up, with an area under the curve (AUC) of 0.83 (95% CI 0.77 to 0.90), when compared to an AUC of 0.75 (95% CI 0.67 to 0.83) for %ΔLVESV, P = 0.03 for ROC curves comparison (Fig. 3). An 11% increase in %ΔLVEDV had a sensitivity of 72% and a specificity of 70%, and a 5% increase in LVESV had both a sensitivity and specificity of 78%.

These cut-off values were lower than the MDC95 for inter-observer measurements. Therefore, using the cut-off values for MDC95 for inter-observer measurements (given that the scans from different studies were by different observers), an increase in LVEDV of 12% had a sensitivity of 73% and a specificity of 69% and an increase in LVESV of 12% had a sensitivity of 89% and a specificity of 62% to detect an LVEF of <50%.

Figure 4 shows the relationship between %ΔLVESV and %ΔLVEDV. The dashed lines represent the cut-off values of +12 and −12% change in LVEDV (vertical dashed lines) and +12 and −12% change in LVESV (horizontal dashed lines). Patients were divided into three groups for %ΔLVEF based on the MDC95 cut-off of 13% for inter-observer measurements, namely: blue circles - no change in LVEF at follow-up; green circles - increase in LVEF at follow-up compared to acute scan; red circles - decrease in LVEF at follow-up compared to follow-up. Those with a reduction in LVEF at follow-up were more likely to have an increase in both LVEDV and LVESV, and tended to be in the right upper quadrant (RUQ) of the graph (adverse LV remodeling group). Those with an improvement in LVEF were more likely to have in improvement in LVESV and LVEDV and tended to be in the middle lower quadrant (MLQ) and left lower quadrant (LLQ) of the graph (reverse LV remodeling group). Some patients had an increase in LVEDV only with or without an improvement in LVEF, and tended to lie in the right middle quadrant (RMQ) of the graph (adverse LV remodeling with compensation). Those in the middle quadrant (MQ) of the graph had no change in LVEDV or LVESV and predominantly no change in LVEF (no remodeling group).

Figure 5 provides a schematic representation for evaluating LV remodeling post-STEMI from %ΔLVEDV and %ΔLVESV, using a 2-step approach: firstly the %ΔLVEDV is evaluated (using a cut-off value of 12%) and secondly, the %ΔLVESV is assessed as shown in Fig. 6 (using a cut-off value of 12%). Using this approach, 4 main patterns of post-STEMI LV remodeling were observed: Group 1: reverse LV remodeling (with LVEF predominantly improved, 29% of patients); Group 2: no LV remodeling (with LVEF predominantly unchanged, 19% of patients); Group 3: adverse LV remodeling with compensation (with LVEF predominantly improved, 14%); and Group 4: adverse LV remodeling (with LVEF unchanged or worsened, 31%).

The acute MI size was divided into quartiles as follows: <15%, 15 to 24%, 25 to 33% and ≥33%. Figure 7 shows the distribution of acute MI size divided by quartiles and between those without MVO (Fig. 7a) and those with MVO (Fig. 7b). The incidence of MVO on the acute CMR scan was 43%, 66, 81 and 78% for those in Group 1 to 4 respectively, P = 0.002. Although those with larger MI size and MVO were more likely to have adverse LV remodeling (Group 4 - red box), there were also patients with small MI sizes and no MVO who went on to develop adverse LV remodeling (green dots within red box in Fig. 7a) and adverse LV remodeling with compensation (Group 3 - within yellow box). Likewise, there were also a notable number of patients with large MI size and MVO who developed reverse LV remodeling (Group 1 - black dots within blue box in Fig. 7b).

Inter-observer and intra-observer measurements were performed in only 40 patients (80 scans) but this is significantly larger than the number of patients used in a previous study (n = 10) providing the minimal detectable change in LVEF by echocardiography in patients undergoing chemotherapy (10 patients with echocardiography at 2 time-points) [22]. We only used one analysis tool and LV parameters were quantified using the semi-automated method. We did not have matching echocardiography data for comparison. We did not have complete data on the presence of multi-vessel disease or clinical outcomes and our sample size was relatively small. Therefore, we used an LVEF of <50% at follow-up as a surrogate. [15] There was heterogeneity in the performance of CMR for acute MI size and MVO (scanner strength, dosage and type of contrast, timing of LGE for MVO and MI, quantification technique used – Additional file 1: Online appendix Table 1) and our findings need to be confirmed by future studies.