Diagnostic accuracy of cardiac magnetic resonance in acute myocarditis: a systematic review and meta-analysis

Diagnostic accuracy of cardiac magnetic resonance in acute myocarditis: a systematic review 1
and meta-analysis 2
Christos P. Kotanidis, MD, MSca, Maria-Anna Bazmpani, MDa, Anna-Bettina Haidich, 4
PhDb, Charalambos Karvounis, MD, PhDa, Charalambos Antoniades, MD, PhDc, Theodoros 5
D. Karamitsos, MD, PhDa 6
7
a) First Department of Cardiology, AHEPA Hospital, Aristotle University of Thessaloniki, 8
Thessaloniki, Greece 9
b) Department of Hygiene and Epidemiology, Medical School, Aristotle University of 10
Thessaloniki, Greece 11
Oxford, Oxford, UK 13
14
15
Total word count: 4,459 16 17 Brief title: CMR in acute myocarditis: a meta-analysis 18 19 Keywords 20 Acute myocarditis, CMR, T1 mapping, T2 mapping, Lake-Louise criteria, meta-analysis 21 22 23 No conflicts of interest – no relationships with industry. 24 25 Christos P. Kotanidis acknowledges support by the Alexandros S. Onassis Public Benefit 26 Foundation, Athens, Greece. 27 28 29 Address for correspondence 30 Prof. Theodoros D. Karamitsos 31 First Department of Cardiology 32 Aristotle University of Thessaloniki 33 AHEPA Hospital 34 55136, St. Kyriakidi 1 street 35 Thessaloniki, Greece 36 37 Tel: +30 2310994830 38 Fax: +30 2310994673 39 Email: [email protected] 40 41 42 43
Objectives 46
The aim of this systematic review was to explore the diagnostic accuracy of various Cardiac 47
Magnetic Resonance (CMR) index tests for the diagnosis of acute myocarditis in adult 48
patients. 49
Background 51
Acute myocarditis remains one of the most challenging diagnoses in cardiology. CMR has 52
emerged as the diagnostic tool of choice to detect acute myocardial injury and necrosis in 53
patients with suspected myocarditis. 54
55
We considered all diagnostic cohort and case-control studies. We searched MEDLINE, 57
EMBASE, the Cochrane Library, SCOPUS and Web of Science, up to April 21, 2017. We 58
used the Quality Assessment of Diagnostic Accuracy Studies-2 tool to assess the quality of 59
included studies. PROSPERO registration number: CRD42017055778. 60
61
Twenty-two studies were included in the systematic review. Because significant 63
heterogeneity exists among the studies, we only present hierarchical Receiver Operator 64
Curves. The Areas Under the Curve (AUC) for each index test were: for T1 mapping 0.95 65
[95% CI: 0.93-0.97], for T2 mapping 0.88 [95% CI: 0.85-0.91], for Extracellular Volume 66
Fraction (ECV) 0.81 [95% CI: 0.78-0.85], for increased T2 ratio/signal 0.80 [95% CI: 0.76-67
0.83], for Late Gadolinium Enhancement (LGE) 0.87 [95% CI: 0.84-0.90], for Early 68
Gadolinium Enhancement (EGE) 0.78 [95% CI: 0.74-0.81] and for the Lake-Louise Criteria 69
(LLC) 0.81 [95% CI: 0.77-0.84]. Native T1 mapping had superior diagnostic accuracy across 70
all index tests. The AUC of T2 mapping was greater than the AUC of increased T2 71
ratio/signal and EGE, whereas ECV showed no superiority compared to other index tests. 72
LGE had better diagnostic accuracy compared to the classic CMR index tests, similar 73
accuracy with T2 mapping and ECV, and only T1 mapping surpassed it. 74
75
Novel CMR mapping techniques provide high diagnostic accuracies for the diagnosis of 77
acute myocarditis and constitute promising successors of the classic elements of the LLC, for 78
routine diagnostic protocols. 79
95
Cardiac Magnetic Resonance (CMR) plays a central role in the diagnosis of acute 96
myocarditis. The aim of this systematic review was to explore the accuracy of standard and 97
novel CMR index tests for diagnosing acute myocarditis in adults. Novel mapping techniques 98
have been shown to have superior diagnostic accuracy compared to the Lake-Louise criteria 99
(LLC). Native T1 mapping had a significantly higher accuracy than all other index tests 100
examined. ECV offers no additional accuracy in diagnosing acute myocarditis than the 101
classic components of the LLC, alone or combined. T2 mapping had higher accuracy than 102
increased T2 ratio/signal and early gadolinium enhancement. 103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
CMR = Cardiac Magnetic Resonance 122
ECV = Extracellular Volume Fraction 123
EGE = Early Gadolinium Enhancement 124
HSROC = Hierarchical Summary Receiver-Operator Curve 125
LGE = Late Gadolinium Enhancement 126
LLC = Lake-Louise Criteria 127
ShMOLLI = Shortened modified Look-Locker inversion recovery sequence 129
QUADAS-2= Quality Assessment of Diagnostic Accuracy Studies 2 130
131
132
133
134
135
136
137
138
139
140
141
142
143
145
Acute myocarditis is defined as an inflammatory disease of the myocardium that can 146
result from a wide variety of infectious agents (viruses, bacteria, etc.), systemic diseases, 147
drugs, and toxins(1). The variable clinical presentation of patients with acute myocarditis(2) 148
makes its diagnosis a challenge. Although endomyocardial biopsy is considered the gold 149
standard for the diagnosis of acute myocarditis (3), several limitations restrict its widespread 150
application. Therefore, it is only recommended in a limited number of clinical scenarios such 151
as patients with evidence for heart failure in combination with acute disease (<2 weeks, class 152
I) or left ventricular dilatation (<3 months, class I) or specific other cases of heart failure 153
(class IIa)(4). 154
Cardiac magnetic resonance (CMR) is currently considered the most comprehensive 155
and accurate diagnostic tool in patients with suspected myocarditis(5). The Lake-Louise 156
Criteria (LLC) recommend to combine different CMR techniques in patients with suspected 157
myocarditis to determine myocardial edema (T2-weighted imaging), hyperemia (T1-weighted 158
imaging), and fibrosis (late gadolinium enhancement [LGE])(5). The LLC use a semi-159
quantitative approach and allow the detection of myocardial inflammation and necrosis. 160
However, several technical limitations exist: T1-weighted spin-echo sequences during free 161
breathing frequently suffer from poor image quality whereas T2-weighted spin-echo images 162
have a low signal-to-noise ratio(5). Furthermore, LGE alone may fail to characterize acute 163
myocarditis, as some patients only present with acute myocardial inflammation/edema. T1 164
and T2 mapping are novel CMR techniques for quantitative tissue characterization with tight 165
normal ranges, which allow a more objective assessment of myocardial tissue properties. 166
Importantly, parametric mapping techniques appear to overcome some of the 167
aforementioned technical limitations of the LLC(5) and enable the assessment of diffuse 168
myocardial injury, since they have been shown to be highly sensitive to increased free water 169
content rendering them ideal for detecting acute myocardial inflammation/edema (6). 170
The aim of this systematic review is to explore the diagnostic accuracy of classic and 171
novel CMR index tests for the diagnosis of acute myocarditis in adult patients. Secondary 172
aims are to investigate potential sources of heterogeneity and provide preliminary 173
comparisons of the diagnostic accuracy across studied index tests. 174
175
177
The methods and results of this review are being presented according to the Preferred 178
Reporting Items for Systematic reviews and Meta-Analyses statement (PRISMA)(7). The 179
review protocol was previously registered on PROSPERO (CRD42017055778). 180
181
Eligibility Criteria 182
We considered all diagnostic cohort and case-control studies that used either 183
endomyocardial biopsy or clinical criteria for the diagnosis of acute myocarditis. We 184
excluded studies in which we were not able to reconstruct a 2×2 diagnostic table. 185
186
Search Strategy 187
We searched MEDLINE (via PubMed), EMBASE (via Ovid), the Cochrane Library, 188
SCOPUS and Web of Science using both free-text terms and Medical Subject Headings 189
(MeSH) terms, without any date or language restrictions. We also hand-searched trial 190
registries and reference lists of the included studies. The detailed search strategy that was 191
used and a list of literature sources searched are listed in the online appendix. The last search 192
was performed on April 21, 2017. 193
194
Study Selection and Data Extraction 195
Two authors (CK and MAB) independently screened the records retrieved from the 196
search after deduplication by title and abstract. Selected records were further screened for 197
eligibility in full text independently by the same investigators (CK and MAB). Data 198
collection was initially piloted using a predesigned extraction form on three manuscripts, as 199
proposed by literature(8). After necessary adjustments, two independent reviewers (CK and 200
MAB) extracted data using a customized extraction form. Discrepancies at each stage of 201
selection were arbitrated by a third reviewer (TDK) and resolved by consensus. We contacted 202
the authors of included studies to obtain additional data. 203
204
Assessment of Methodologic Quality 205
We used the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool 206
to assess the quality of included studies(9). Two review authors adjusted both the signaling 207
questions and the assessment questions to form a review-specific version of the tool. The tool 208
was subsequently piloted and, when good agreement was achieved, it was used to assess risk 209
of bias and applicability of all included studies independently by the same authors. 210
Disagreements were resolved by consensus. 211
212
Data Synthesis 213
Initially, for each index test, the derived estimates of sensitivity and specificity were 214
plotted in receiver operating characteristic (ROC) curves and forest plots for preliminary 215
investigation. We expected that studies would use different thresholds to dichotomize test 216
results measured on a continuous scale and therefore planned to perform meta-analyses using 217
hierarchical models to produce summary ROC (HSROC) curves and 95% prediction regions, 218
in Stata 13 (Stata Corporation, College Station, TX)(10,11). We also estimated summary 219
sensitivities and specificities, in separate analyses, where studies reported common 220
thresholds. After visual inspection of the forest plots and the summary ROC curves, we 221
investigated the presence of heterogeneity between studies, by performing appropriate meta-222
regression models and wherever possible, subgroup analyses according to study type 223
(cohort/case-control) and CMR field strength. We also performed sensitivity analyses, if the 224
number of available studies was enough, by restricting analyses to studies at low risk of bias. 225
We used the Bonferroni correction to compensate for the type I error increase that occurs 226
with multiple comparisons(12). 227
Search Results 231
Our search yielded 2.725 results from the various literature sources. After removing 232
duplicates, 1.924 records were screened by title and abstract. 1.764 records were excluded 233
and we screened the remaining 160 records by full-text for eligibility. Finally, 22 studies 234
were included in our systematic review. A detailed flow diagram with the study selection 235
process and various reasons for exclusion is shown in Figure 1. 236
237
238
Native T1 mapping 239
Seven studies provided data regarding the diagnostic accuracy of native T1 240
mapping(13–19). All studies but one were considered to be at high risk of bias (Online Figure 241
1). Consequently, it was impossible to perform a sensitivity analysis based on methodological 242
quality. Details pertaining to study characteristics (Online Table 1) and methodologic quality 243
assessment are available in the appendix. In the main analysis, sensitivity ranged from 0.64 to 244
0.98 and specificity ranged from 0.67 to 1.00 (Online Figure 2). We observed heterogeneity 245
between included studies, which could be attributed to the use of different CMR field 246
strengths and variable positivity thresholds. Hence, we only present an HSROC curve (Figure 247
2) with an area under the curve (AUC) of 0.95 [95%CI: 0.93 to 0.97]. We performed 248
subgroup analysis (Online Figure 3) excluding the two studies that used field strengths other 249
than 1.5 T(15,17), with the AUC retreating to 0.92 [95%CI: 0.89 to 0.94]. In the appendix, 250
we depict the conditional probability plot (Online Figure 5) and Fagan nomogram (Online 251
Figure 4) to support decision making and interpretation of the clinical utility of the index test 252
for detecting patients with acute myocarditis. For example, in an average-risk population with 253
a pretest probability (prevalence) equal to 25%, native T1 mapping increases the probability 254
of acute myocarditis to 75% when the test result is positive, and decreased the probability to 255
4% when the test result was negative. We found no publication bias (Online Figure 6) in the 256
regression test for funnel plot asymmetry (p=0.19). To explore the potential effect of 257
publication year and sample size on heterogeneity, we conducted a meta-regression analysis. 258
Neither one proved to affect sensitivity/specificity values. 259
260
T2 mapping 261
Six of the included studies(13,14,18–21) provided data regarding the diagnostic 262
accuracy of T2 mapping. Details about study characteristics (Online Table 2) and 263
methodologic quality are available in the appendix. Four of the studies included were 264
considered to be in high risk of bias (Online Figure 9). The forest plot depicting sensitivities 265
and specificities of the six included studies in the main analysis is available in Online Figure 266
7. Sensitivity ranged from 0.57 to 0.94 and specificity from 0.60 to 0.92. Because of variable 267
cut-off values being used across the studies, we avoided presenting summary sensitivity and 268
specificity points and chose to present an HSROC curve (Figure 3). We found no publication 269
bias in the regression test for funnel plot asymmetry (p=0.19). All studies used 1.5 T field 270
strength, so we solely sought to perform subgroup analysis based on study design. Only two 271
studied used the one-gated prospective cohort design and we weren’t able to fit proper meta-272
analytical models, since those require a minimum of four studies(22). Sample size proved to 273
affect sensitivity/specificity values (p=0.04) with small studies presenting higher values of 274
test performance, contrary to publication year. 275
276
Extracellular Volume Fraction (ECV) 277
Data on the diagnostic accuracy of ECV was available in five studies(13–15,18,19). 278
Four of the studies included were considered to be at high risk of bias (Online Figure 10). 279
Only one study used a cohort design(19) and therefore sensitivity analysis was not possible. 280
A table with details of study characteristics (Online Table 3) as well as information about 281
methodologic quality assessment is available in the appendix. Online Figure 8 presents the 282
forest plot depicting sensitivities and specificities of the five studies. Sensitivity ranged from 283
0.67 to 0.94 and specificity from 0.56 to 0.90. Because of variable cut-off values being used 284
across the studies, we only present an HSROC curve (Figure 4) with an AUC of 0.81 (95% 285
CI: 0.78 to 0.85) and not summary sensitivity or specificity points. 286
287
Classic CMR techniques 288
In this section, we present the core findings of our review regarding the diagnostic 289
accuracy of increased T2 signal, late gadolinium enhancement (LGE), early gadolinium 290
enhancement (EGE) and their combination, the Lake-Louise criteria (LLC). A more detailed 291
presentation with information about methodologic quality assessment and data synthesis is 292
available in the appendix. In total, 15 of the included studies provided data on the diagnostic 293
accuracy of increased intensity in T2-weighted images(13–18,23–31). Sensitivity ranged 294
from 0.45 to 1.00 and specificity from 0.43 to 1.00. Fitting the HSROC curve yielded an area 295
under the curve equal to 0.80 (95% CI: 0.76 to 0.83). The diagnostic accuracy of LGE was 296
examined in 17 of the included studies(13–19,23,25–30,32–34) with a summary sensitivity 297
point of 0.68, a summary specificity point of 0.96 and AUC of 0.87 (95% CI: 0.84 to 0.90). 298
Ten studies(14,15,18,23,25–27,30,31,35) provided data on the diagnostic accuracy of early 299
enhancement. Sensitivity ranged from 0.49 to 0.85, specificity from 0.43 to 1.00 and, the 300
AUC was equal to 0.78 (95% CI: 0.74 to 0.81). The diagnostic accuracy of the LLC was 301
investigated in eight studies(14,15,18,19,25–27,30). Sensitivity was 0.78, specificity 0.88 and 302
the AUC 0.83 (95% CI: 0.79 to 0.86). 303
304
Comparison of the diagnostic accuracy of index tests 305
In order to compare the diagnostic accuracy of index tests analyzed above, we 306
compared the AUC values derived from fitting summary ROC curves to the individual study 307
data for each index test(36). Overall results from all meta-analyses conducted with details on 308
accuracy can be found in Table 1. Online Table 8 presents the comparison matrix, with 309
corresponding p values. Native T1 mapping proved to be superior in terms of diagnostic 310
accuracy across all index tests, including the comparison with T2 mapping or LGE. The AUC 311
of T2 mapping was significantly greater than the AUC of increased T2 ratio/signal and EGE, 312
whereas ECV showed no superiority compared to other index tests. LGE proved to have 313
better diagnostic accuracy compared to both T2 weighted imaging and EGE, similar accuracy 314
with T2 mapping and ECV, and only T1 mapping managed to surpass it. 315
316
318
This meta-analysis showed that the accuracy of standard CMR techniques for the 319
diagnosis of acute myocarditis (LLC: edema on T2-weighted imaging, hyperemia on T1-320
weighted imaging and fibrosis on late gadolinium enhancement imaging) is good, but can be 321
further improved with the addition of novel parametric mapping techniques. In an acute 322
clinical setting, imaging of myocardial edema by T2 weighted sequences could be replaced 323
by native T2 mapping and, similarly, imaging of myocardial hyperemia using T1 weighted 324
sequences could be replaced by native T1 mapping. LGE has high diagnostic accuracy and 325
can constitute the third component of the revised CMR criteria for acute myocarditis. It 326
should be noted though that we observed significant heterogeneity between studies, which 327
arise from the different sequences, imaging protocols and field strengths used. 328
The diagnostic value of native T1 relaxation times in subjects with acute myocarditis 329
has been reported in numerous studies, achieving an accuracy of 61 to 99%(13,15–17). 330
Different T1 thresholds but within a tight range of 990-1000ms have been used at 1.5T. A 331
cut-off value of 992ms used by Hinojar et al(17) yielded the highest sensitivity of 98% and 332
specificity of 100%. Of note, 6 out of 7 studies(13–15,17–19) used a modified Look-Locker 333
inversion recovery sequence (MOLLI) whereas Ferreira et al(16) used a shortened version of 334
MOLLI (ShMOLLI) which yielded a sensitivity and specificity of 90% and 88%, 335
respectively. Αpart from MOLLI-based inversion sequences which are the most commonly 336
used and validated, several other T1 mapping sequences exist. The plethora of different T1 337
mapping methods generates some uncertainty whether results of one study can be confirmed 338
on a similar patient population with a different T1 mapping methodology. Additionally, there 339
was a variation in the days that CMR was performed. While in most studies, CMR was 340
performed shortly after presentation, Radunski et al(18) report a median interval between 341
onset of symptoms and CMR of two weeks. This resulted in a significantly lower sensitivity 342
of 64% but a high specificity of 90%. 343
Studies using T2 mapping at 1.5 T reported sensitivity that ranged from 57% to 94% 344
while specificity ranged from 60% to 92%. Bohnen et al(21) in a study with one-gated design 345
found the highest sensitivity of 94% with a cut-off of 60ms while Radunski et al using a 346
similar cut-off of 61ms found a significantly lower sensitivity of 57%. Once again, this 347
discrepancy is probably due to the more subacute clinical presentation (2 weeks after onset of 348
symptoms) in the study by Radunski et al(18). Luetkens et al(14) reported an equivalent 349
diagnostic performance of T2 mapping when compared to T1 mapping and increased 350
diagnostic accuracy when compared to T2 weighted imaging, confirming the improved 351
detection of diffuse inflammation by T2 mapping compared to standard T2 weighted images. 352
Several T2 mapping sequences exist (although fewer than T1 mapping) which may affect 353
diagnostic thresholds for detecting disease. 354
In the five studies that used ECV in acute myocarditis, sensitivity ranged from 67% to 355
94% and specificity ranged from 56% to 90%. Von Knobelsdorff-Brenkenhoff et al(13) 356
reported lower ECV values compared to Luetkens et al(14) and Radunski et al(18). This 357
discrepancy may be explained by the focal nature of disease, the…