Background
The distinction of hypertrophic cardiomyopathy (HCM) or cardiac amyloidosis (CA) from hypertensive heart disease may be difficult. The aim of this study was to determine the impact of parametric (polar) maps of regional longitudinal strain on identification of the etiology of mild to moderate left ventricular hypertrophy (LVH).
Methods
Twenty-four consecutive echocardiographic studies with mild to moderate LVH (eight with CA, eight with HCM, and eight with hypertensive heart disease) were selected on the basis of the availability of adequate images to assess longitudinal strain and absence of electrocardiographic criteria for low voltage or LVH or a pseudoinfarct pattern. Twenty level 3–trained readers provided the most likely of three diagnoses (CA, HCM, or hypertensive heart disease) and scored their confidence in making the diagnosis from two-dimensional images and diastolic parameters. A teaching exercise was provided on the interpretation of longitudinal strain in these cohorts, and interpretation was repeated with the addition of the strain polar map.
Results
Baseline concordance among the readers was poor (κ = 0.28) and improved with the addition of strain data (κ = 0.57). Accuracy was improved with the addition of polar maps for the entire study cohort ( P < .001), with 22% of cases reclassified correctly. The largest improvements in sensitivity (from 40% to 86%, P < .001), specificity (from 84% to 95%, P < .001), and accuracy (from 70% to 92%, P < .001) were seen for CA. The strain polar map significantly improved reader confidence in making the correct diagnosis overall ( P < .001).
Conclusions
Regional variations in strain are easily recognizable, accurate, and reproducible means of differentiating causes of LVH. The detection of LVH etiology may be a useful clinical application for strain.
Despite the availability of systolic longitudinal strain (LS) analysis using two-dimensional (2D) speckle-tracking over the past decade, its clinical uptake has been slow. In addition to the identification of subclinical ventricular dysfunction, recent studies have suggested a role for strain in the differentiation of causes of increased left ventricular (LV) wall thickness. Echocardiography plays an important role in differentiating the causes of increased LV wall thickness, but standard diagnostic techniques have limited ability to identify the etiology of mild to moderate thickening. Although this is most commonly associated with LV hypertrophy (LVH) secondary to hypertensive heart disease (HHD), there is increasing interest in reliable early detection of hypertrophic cardiomyopathy (HCM) and cardiac amyloidosis (CA), because the prognosis and management of these entities are very different. Unfortunately, the characteristic features of these entities—electrocardiographic abnormalities and echocardiographic morphologic patterns such as eccentric LVH, dynamic outflow tract obstruction, advanced diastolic dysfunction, pericardial effusion or diffuse cardiac chamber, and valvular thickening—are associated with late-stage disease.
We and others have described regional patterns of LS in CA and HCM. However, these were generally proof-of-concept studies in which the LS patterns were described in patients with more advanced disease in whom other echocardiographic and electrocardiographic features may have provided the diagnosis. In addition, these studies did not evaluate the impact of simple pattern recognition using strain parametric (polar) maps on the accuracy of diagnosis. The aim of this study was to assess the impact of adding a strain polar map to traditional echocardiographic parameters on concordance, reviewer confidence, and diagnostic accuracy in patients with mild to moderate LVH without characteristic electrocardiographic patterns, because this cohort often proves to be a diagnostic dilemma.
Methods
Patient Selection
Consecutive cases were identified for inclusion retrospectively from the echocardiography database at the Cleveland Clinic between April 2011 and March 2012 using the search terms “mild hypertrophy,” “moderate hypertrophy,” “cardiac amyloid,” and “hypertrophic cardiomyopathy” and their derivatives, but strain measurements were obtained prospectively and blinded to other data. Of the cases identified, patients were included if they had mild or moderate LVH (on the basis of 2D LV wall thickness defined according to American Society of Echocardiography guidelines ), normal-voltage QRS complexes on electrocardiography, and confirmed diagnoses of CA, HCM, or HHD. The diagnosis of CA was based on myocardial biopsy or, in the absence of biopsy, characteristic findings on cardiac magnetic resonance imaging, which had to have been independently read as consistent with CA, in the absence of hypertension. The diagnosis of HCM was based on myocardial biopsy or LVH in the absence of hypertension with either genetically proven HCM or a family history of genetically proven HCM. The diagnosis of HHD was based on a history of hypertension and the absence of alternative diagnoses as potential causes of LVH. All patients were reviewed and their diagnoses confirmed by an independent cardiologist at the same institution, and the study was approved by the institutional review board.
Exclusion criteria included greater than moderate LVH, a history of obstructive coronary artery disease, a low-voltage or pseudoinfarct pattern on electrocardiography, greater than trivial pericardial effusion, and dynamic outflow tract obstruction at rest. Because CA is the least common diagnosis, this cohort was identified first. A decision was made a priori to match the number of suitable patients with CA with the same number of patients with HCM and HHD. A quantitative approach to the pattern of LS has been published previously in the CA group ; the remaining groups were identified from a de novo search of the echocardiographic database.
LS Analysis
To avoid concerns regarding variability of strain among manufacturers, only studies performed on Vivid 7 or Vivid 9 ultrasound systems (GE Medical Systems, Milwaukee, WI) were included. Speckle-tracking analysis was performed offline using dedicated software (EchoPAC; GE Medical Systems). For the three apical views, two sample points were manually placed at the endocardial border of the LV base, and a single sample point was placed at the apex at end-diastole. A region of interest was automatically generated to encompass the myocardial thickness along the length of the LV wall. The thickness and position of the region of interest were adjusted manually to ensure that the inner margin conformed to the endocardial border and included the entire thickness of the myocardium. Aortic ejection time was obtained from the LV outflow tract velocity-time integral pulsed-wave Doppler signal. The software then automatically identified and tracked the speckles, frame by frame, throughout systole. Adequacy of tracking and timing was verified by visualization of the speckle-tracking and review of the strain curves generated. Only those with adequate quality 2D images were included; as described previously, inadequate 2D quality was defined as suboptimal visualization of more than two segments or if more than two segments failed to track adequately after 2 manual adjustments of the endocardial border.
Review Process
This assessment involved 20 echocardiographic readers (level 3 physicians), who reviewed 24 cases in two 1-hour sessions performed 1 week apart. The readers were aware that this work was being performed for research purposes and assented to participation. For each case, the group was shown selected cine images from clinical 2D echocardiograms that included parasternal long-axis, apical long-axis, and apical two- and four-chamber views. In addition, images of Doppler evaluation of mitral inflow and tissue Doppler evaluation of medial and lateral annuli were provided. The following values (as measured according to American Society of Echocardiography guidelines), were provided: ejection fraction using the biplane Simpson’s method from apical two- and four-chamber windows, mean LV wall thickness (calculated as [interventricular septum + posterior wall thickness]/2), LV mass index, left atrial volume index using a biplane area-length formula, peak early (E) velocity, ratio of peak early and late (E/A) velocities and E-wave deceleration time from the mitral inflow, average early (e′) diastolic annular velocities from the medial and lateral wall LV walls, and E/e′ ratio. To avoid bias in the assessment of diagnosis, no clinical data were provided for any case.
At each assessment, readers were given approximately 2 to 3 minutes to review each case and provide one of the possible diagnoses: HHD, HCM, or CA. No guidelines or criteria were provided to aid the readers in making the diagnosis. Each diagnosis was graded in terms of reader confidence in the correctness of the diagnosis from 0 to 2 (0 = low likelihood [<30%], 1 = intermediate likelihood [30%–70%], 2 = high likelihood [>70%]) on the basis of the imaging parameters provided.
The follow-up involved a 5-min teaching exercise on the interpretation of LS in the context of LVH followed by re-presentation of the 20 initial cases with the addition of strain polar maps. An example of the data presented to the reviewers from a representative patient from each of the three cohorts (CA, HCM, and HHD) is provided in Figure 1 . At the initial read, only the 2D cine and Doppler measurements were provided; the polar map was added at the follow-up read. The cases were interpreted in a manner identical to the baseline exercise. The teaching exercise involved a brief review of previously published work describing typical patterns of LS seen in CA (relative apical sparing of LS with reduction in LS at the basal and middle levels of the left ventricle) and HCM (reduced strain at the site of greatest hypertrophy). No cases used in the study were shown or discussed during the teaching exercise.
Statistical Analysis
Summary statistics are presented as mean ± SD for continuous data. Comparisons across groups were performed using one-way analysis of variance. Concordance among readers was measured using the chance-adjusted multirater free marginal κ coefficient. The NcNemar test was used to assess for change in diagnoses with and without strain polar maps. Sensitivity, specificity, positive predictive value, negative predictive value and accuracy were calculated for each diagnosis using 2 × 2 tables. Paired Wilcoxon signed-rank tests were used to assess for differences in these parameters. On the basis of the interpretation provided by the readers, the data were recoded for analysis as follows: low-likelihood disease was scored 0, intermediate likelihood for the correct diagnosis was scored +1 and for the incorrect diagnosis −1, and high likelihood for the correct diagnosis was scored +2 and for the incorrect diagnosis −2. Paired Wilcoxon signed-rank tests were then used to assess differences between the two reads, and corresponding Z scores for each reader were obtained separately. Z scores thus obtained represent the effect size of the intervention within individual readers and follow a normal distribution. To assess whether the intervention had an overall effect, Z scores for all 20 readers were averaged and the standard deviation obtained, and the significance of a difference from zero was assessed using a one-sample t test. The net reclassification index was calculated for the improvement in diagnostic accuracy using the polar maps. Statistical analysis was performed using SPSS version 18 (IBM, Armonk, NY).
Results
Study Selection
Of the 55 patients with confirmed CA and suitable echocardiographic studies with measurable LS identified between April 2011 and March 2012, only eight had less than severe hypertrophy, had normal-voltage QRS complexes, and fulfilled all of the remaining inclusion and exclusion criteria. These patients were then matched to eight consecutive suitable patients with HCM and HHD. All patients with CA who met the inclusion criteria had adequate image quality for strain assessment. A total of nine patients with HCM and 10 patients with HHD were screened; one patient with HCM and two patients with HHD were excluded because of inadequate image quality for strain analysis. The mean follow-up period was 72 ± 57 months without a change in diagnosis (calculated from the date of diagnosis or, if unavailable, the date of first recording of the diagnosis, to the most recent review at this institution).
A summary of the echocardiographic data of the 24 cases, as presented to the reviewers, is shown in Table 1 .
| Variable | Overall ( n = 24) | CA ( n = 8) | HCM ( n = 8) | HHD ( n = 8) | P |
|---|---|---|---|---|---|
| EF (%) | 61 ± 5 | 58 ± 6 | 62 ± 5 | 61 ± 4 | .25 |
| MWT (mm) | 14 ± 1 | 14 ± 1 | 13 ± 1 | 14 ± 1 | .28 |
| LVMI (g/m 2 ) ∗ | 117 ± 17 | 113 ± 24 | 119 ± 16 | 117 ± 10 | .79 |
| LAVI (mL/m 2 ) | 33 ± 9 | 34 ± 7 | 35 ± 11 | 30 ± 8 | .44 |
| E (cm/sec) | 83 ± 32 | 87 ± 22 | 85 ± 46 | 76 ± 26 | .77 |
| E/A ratio (cm/sec) | 1.2 ± 0.7 | 1.6 ± 0.9 | 1.1 ± 0.4 | 0.9 ± 0.3 | .11 |
| DT (m/sec) | 226 ± 51 | 211 ± 63 | 213 ± 44 | 253 ± 38 | .19 |
| e′ (m/sec) | 6 ± 2 | 6 ± 2 | 7 ± 2 | 6 ± 3 | .67 |
| E/e′ ratio | 15 ± 6 | 16 ± 7 | 16 ± 11 | 13 ± 8 | .52 |
∗ LVMI was calculated using the truncated ellipsoid formula per American Society of Echocardiography guidelines.
The same participants took part in both baseline and follow-up exercises. There were a total of 960 diagnoses (24 cases, 20 readers at two time points) and a total of 2,785 likelihood data points analyzed (24 cases, three potential diagnoses, 20 readers at two time points; 95 data points were left blank, presumably because the reader was undecided within the allocated time frame).
Concordance
Concordance among readers significantly improved with the addition of the strain polar maps (baseline κ = 0.28, increased to 0.57). Improvement was most notable in the CA group (baseline κ = 0.16, increased to 0.73). However, it was also seen in the HCM (baseline κ = 0.33, increased to 0.47) and HHD (baseline κ = 0.33, increased to 0.51) groups.
Accuracy
The accuracy of diagnosis for the entire study cohort was improved with the addition of the polar maps (65% vs 82%, P < .001). This was driven primarily by an improvement in the accurate diagnosis of CA (70% vs 92%, P < .001) but also by improvement in the diagnosis of HHD (60% vs 73%, P = .001) and HCM (65% to 73%, P = .001).
Overall, 22% of cases were reclassified correctly with the addition of the strain polar maps. The largest correct reclassification was seen in the CA group (46%), with 9% of HCM cases and 10% of HHD cases also reclassified correctly. Table 2 summarizes the numbers reclassified by individual readers.
| Reader | Overall | CA | HCM | HHD |
|---|---|---|---|---|
| 1 | +25% | +50% | 0% | +25% |
| 2 | +29% | +38% | +13% | +38% |
| 3 | +29% | +63% | +25% | 0% |
| 4 | +4% | +25% | −25% | +13% |
| 5 | +21% | +50% | 0% | +13% |
| 6 | +4% | +13% | 0% | 0% |
| 7 | +21% | +38% | +25% | 0% |
| 8 | +33% | +63% | +13% | +25% |
| 9 | +21% | +75% | +25% | −38% |
| 10 | +21% | +50% | +13% | 0% |
| 11 | +25% | +25% | +25% | +25% |
| 12 | +13% | +38% | 0% | 0% |
| 13 | +38% | +38% | +38% | +38% |
| 14 | +13% | +50% | 0% | −13% |
| 15 | +38% | +38% | +25% | +50% |
| 16 | +13% | +25% | +13% | 0% |
| 17 | +25% | +25% | +13% | +13% |
| 18 | +13% | +75% | −13% | −25% |
| 19 | +25% | +63% | −13% | +25% |
| 20 | +29% | +75% | 0% | +13% |
The addition of polar maps provided statistically significant improvements in the sensitivity, specificity, accuracy, and positive and negative predictive values for the echocardiographic diagnosis of CA ( Table 3 ). Although accuracy and specificity parameters for the diagnosis of HCM or HHD were improved by the addition of strain polar maps ( Table 3 ), there was only a trend toward improvement in sensitivity ( P = .054 and P = .061, respectively).
| Diagnosis | Sensitivity (%) | Specificity (%) | Accuracy (%) | PPV (%) | NPV (%) |
|---|---|---|---|---|---|
| CA | |||||
| Baseline read | 40 | 84 | 70 | 55 | 75 |
| Strain read | 86 | 95 | 92 | 92 | 94 |
| P | <.001 | .002 | <.001 | <.001 | <.001 |
| HCM | |||||
| Baseline read | 44 | 75 | 65 | 45 | 73 |
| Strain read | 52 | 84 | 73 | 63 | 78 |
| P | .054 | .01 | .001 | <.001 | .005 |
| HHD | |||||
| Baseline read | 60 | 59 | 60 | 42 | 72 |
| Strain read | 70 | 74 | 73 | 59 | 84 |
| P | .061 | .002 | .001 | .001 | .004 |
Qualitative versus Quantitative Evaluation of Regional Strain
Figure 1 demonstrates the typical patterns of regional strain seen in the different forms of LVH included in this study: the patient with CA has relative sparing of apical LS with reduction in middle and basal segmental LS, the patient with HCM has a reduction in strain of the basal inferior and inferoseptal segments at the site of greatest hypertrophy seen on the 2D images, and the patient with HHD has essentially normal strain, although this can vary as described below.
Strain polar maps of all eight subjects with CA are presented in Figure 2 . These cases show a typical apical-sparing pattern of LS with reduced strain (less negative) at the base and higher strain at the apex (more negative); however, cases A and H are less obvious using the visual color gradations. In the reading session with polar maps, only 22 of 160 reads were misdiagnosed (∼14%), and of these, >85% (19 of 22) were cases A and H as seen in Figure 2 . Using the previously described relative LS formula (average apical LS/[average basal LS + average middle LS]) a cutoff value of 0.93 was 100% sensitive and specific for the diagnosis of CA.
