Reply to the Letter to the Editor: “CT compared to MRI for functional evaluation of the right ventricle: a systematic review and meta-analysis”
by Hang Fu, Xuedong Wang, Kaiyue Diao, Shan Huang, Hui Liu, Yue Gao, Qin Zhao, Zhi-gang Yang & Ying-kun Guo (gykpanda@163.com, fuhang66666@163.com)
CT compared to MRI for functional evaluation of the right ventricle: a systematic review and meta-analysisWe thank Prof Jawdat Abdulla very much for his interest in our research [1] and for his kind opinions. We hope we have resolved the issue he was concerned about.
We extracted the mean difference (MD) and limit of agreement (LoA) of each right ventricular function indexes form Bland-Altman analysis in each individual study that we enrolled. We pooled MD and LoA based on fixed model (inverse variance method), which we have written in method. The Meta module (Effect variables) in Stata software was used to calculate the overall results. However, the 95% CI was the default label on the picture generated by Stata software even though LoA was the real parameter analysed. We forgot to replace ‘95% CI’ with ‘95% LoA’, which should not have happened. Additionally, we found we misused ‘95% CI’ at another two places in our original paper: the second sentence in the Results of the Abstract and the fourth sentence of paragraph 4 in the Quantitative results. We have written a letter to editor to correct it.
We did not refer to the results as having “good” or “poor” agreement directly. However, we wrote that CT achieves acceptable accuracy for assessing cardiac function, which means that we think good agreement was achieved by CT. Bland-Altman is a good method to assess the agreement between two methods. A normal distribution of the difference between two methods is the criterion to use Bland-Altman analysis. In our research, we mainly focused on the limitations of the enrolled studies but neglected the probable bias caused by the statistical methods. This limitation should be reported in any meta-analysis that pools variables for Bland-Altman analysis. For example, this meta-analysis assessing quantification of mitral valve regurgitation among 2D echocardiography, 3D echocardiography, and cardiac magnetic resonance not only reported the intrinsic limitations of individual studies but also reported the probable bias of Bland-Altman analysis [2].
In addition, in the part on the qualitative results, we have reported the mean age and the percentage of males in the overall population. The interval between the two examinations in each enrolled study is presented in Table 1 of our original paper. We did not report the overall interval because a specific mean value of that interval, such as 1 ± 0.9 days, was not provided in some studies. We did not report the mean RVEDV, RVESV, RVEF or SV of the two comparison groups because we pooled the MD directly. However, it is better to report these variables to help the reader understand our study. Thus, the mean value of each right ventricle function parameter is presented in Table 1.
Table 1. Mean value of each right ventricle function indexes in the two measurement
| Study | EDV(ml) | ESV(ml) | RVEF(%) | SV(ml) | ||||
| CT | CMR | CT | CMR | CT | CMR | CT | CMR | |
| Elgeti Thomas | 168.6 | 153.7 | 104.7 | 95.1 | 40.2 | 40.2 | 63.2 | 58.7 |
| Alexander Lembcke | 96.2 | 95.4 | 58.4 | 55.9 | 41.9 | 43.3 | 37.8 | 39.8 |
| K. Koch | 155.4 | 151.9 | 79.1 | 75.0 | 50.8 | 51.9 | 76.2 | 76.9 |
| Subha V. Raman 1 | 101.0 | 102.3 | 43.1 | 44.0 | 57.1 | 57.2 | – | – |
| Cedric Plumhans | 78.3 | 77.9 | 38.3 | 37.3 | 51.0 | 51.4 | 40.0 | 40.0 |
| Janina Schroeder | 171.4 | 148.9 | 100.9 | 65.1 | 43.8 | 58.3 | – | – |
| Mira Müller | 90.0 | 86.1 | 37.9 | 35.5 | 58.6 | 58.7 | 52.1 | 50.4 |
| Ying-Kun Guo | 138.1 | 137.7 | 61.7 | 62.5 | 55.3 | 54.4 | 76.3 | 75.1 |
| Lissa Sugeng | – | 205 | – | 131 | – | 40 | – | – |
| J. Jensen | 123.7 | 124.3 | 64.0 | 62.6 | 49.0 | 50.2 | – | – |
| Erica Maffei | 84 | 80 | 46 | 43 | 47 | 47 | 38 | 37 |
| Richard A.P. Takx | 176.5 | 172.7 | 93.2 | 73.4 | 48.0 | 57.9 | 83.3 | 99.2 |
| Xiaoyong Huang | 108.5 | 113.5 | 69.8 | 73.2 | 38.8 | 39.1 | 39.0 | 40.2 |
| Zhang Xiao-chun | 136.5 | 136.7 | 80.1 | 81.1 | 40.6 | 40.1 | 56.4 | 55.6 |
| Andreas Fuchs | 172 | 145 | 77 | 58 | 56 | 60 | 95 | 87 |
| Lei Wang | 115 | 111 | 86 | 85 | 26 | 24 | 26 | 28 |
| Ying-kun Guo 1a | 117.9 | 116.0 | 50.4 | 49.9 | 57.4 | 56.6 | 66.9 | 66.1 |
| Ying-kun Guo 1b | 117.3 | 115.9 | 63.4 | 60.8 | 46.2 | 47.4 | 53.6 | 54.8 |
| Yuzo Yamasaki | 164.4 | 147.3 | 93.0 | 80.0 | 44.2 | 46.8 | 71.5 | 67.3 |
Abbreviation: End Diastolic Volume, EDV; End Systolic Volume, ESD; Right Ventricular Ejection Fraction, RVEF; Stroke Volume, SV; Computed Tomography, CT; Cardiac Magnetic Resonance, CMR
After checking our paper further, consulting statistics experts, and doing the meta-analysis again, we found that we misused the weighted method in Stata software. The weighted method based on sample size was the right method to do the meta-analysis, as we wrote in the Methods. To correct our mistake, we performed meta-analysis with the weighted method based on sample size again. Compared with the results in our original published paper, the specific values of the quantitative pooled results of each right ventricle function parameter changed slightly. However, the final conclusion was not affected. The contents that need to be modified are the Results section of the Abstract, the Quantitative analysis section of the Results, Table 2, Figure 3 and Figure 5. The correct values that belong in the Results, pictures and table are shown in the following corrigendum. The contents that need to be corrected are crossed out, followed by the correct contents written in red. In addition, the corrected Figure 3 and Figure 5 were listed as follows. These corrections have no impact on the final conclusion of the study. The authors sincerely apologize for these errors and the inconvenience caused by them.
Corrigendum
Results in abstract:
Results: Sixteen studies that used disk summation (637 subjects) and three studies that used three-dimensional reconstruction were included. For the 16 studies, the pooled standard mean differences (95% confidence interval) (95% limit of agreement) were 1.04 (− 2.59, 4.67) 2.97(− 4.96,10.91) for EDV, 1.22 (1.50, 3.95) 4.05(− 4.79,12.88) for ESV, − 0.65 (− 2.60, 1.29) − 1.42(− 6.93,4.10) for RVEF, and − 0.37 (− 3.64, 2.90) − 0.05(− 9.06,8.95) for SV. The overall correlation coefficient (r) values were 0.98 0.88 for EDV, 0.95 0.89 for ESV, 0.98 0.83 for RVEF, and 0.97 0.84 for SV. The mean difference between the two methods was not statistically significant (overall effect Z test, p > 0.1).
Quantitative results
The included studies reported a wide range of RVF index values, with RVEF ranging from 24 to 58.7%, EDV 80–172.7 ml, ESV 15.5–131 ml, and SV 37–99.2 ml. Five papers reported EDV, ESV, and RVEF but not SV [20, 26, 32, 33, 39].One study reported a comparison between two subgroups: normal subjects versus patients with mitral valve regurgitation[37]. This amounted to two specific cohorts.
The pooled results for all 19 studies were summarized below. The overall MD (and LOA) were 3.30 (0.92, 5.67) 6.07 (− 1.96, 14.10) for EDV, 4.87 (2.60, 7.14) 6.01 (− 2.25, 14.27) for ESV, − 1.65 (− 3.02, − 0.28) – 1.73 (− 6.59, 3.12) for RVEF, and 1.45 (− 1.06, 3.95) 0.90 (− 7.38, 9.19) for SV, with mild heterogeneity (40.0% 15.6% for EDV and 24.7% 3.5% for ESV) .
Subgroup analysis was conducted according to the CT scanning algorithm used. Among the included studies, 16 applied disk summation and 3 applied 3D reconstruction. Heterogeneity was negligible for all RVF indexes among the 16 studies that employed disk summation. The pooled results (95% CI) (95% LoA) were as follows: 1.04 (− 2.59, 4.67) 2.97 (− 4.96, 10.91) for EDV, 1.22 (1.50, 3.95) 4.05 (− 4.79, 12.88) for ESV, − 0.65 (− 2.60, 1.29) – 1.42 (− 6.93, 4.10) for RVEF, and − 0.37 (− 3.64, 2.90) − 0.05( − 9.06, 8.95) for SV. The pooled results of the 3 studies that applied 3D reconstruction were 16.05 (10.59, 21.51) 23.54 (− 5.42, 52.50) for EDV, 13.02 (8.94, 17.10) 17.16 ( −5.82, 40.13) for ESV, − 2.63 (− 4.56, − 0.70) −3.54 (−12.01, 4.93) for RVEF, and 4.01 (0.12, 7.90) 6.51(− 14.38, 27.40) for SV. The homogeneity of the pooled results was negligible except for EDV (5.4%) excellent.
Among the 16 studies that used disk summation, we conducted subgroup analysis to determine the effect of CT scanner slice/detector number < 64 (subgroup A) or ≥ 64 (subgroup B). The overall results and heterogeneity are presented in Table 2. The overall MD (and LOA) of subgroup B were 0.60 (− 3.27, 4.47) 1.27 (−8.50, 11.05) for EDV, 0.67 (− 2.49, 3.83) 1.68 (−10.31, 13.7) for ESV, − 0.17 (− 2.52, 2.18) −0.56 (− 8.44, 7.33) for RVEF, and − 0.47 (− 3.98, 3.05) 0.36 (−11.1, 11.8) for SV. For subgroup A, the pooled results were 4.30 (− 6.19, 14.78) 6.66 (− 9.83, 23.15) for EDV, 3.28 (− 4.53, 11.09) 8.64 (− 6.50, 23.78) for ESV, − 1.60 (− 6.39, 3.18) −3.17 (−10.87, 4.52) for RVEF, and 0.02 (− 9.97, 10.00) 0.26 (−12.66, 13.18) for SV.
The overall heterogeneity of all 19 studies was greatly affected by the 3 studies that applied a 3D reconstruction algorithm. When the 3 studies were excluded, the heterogeneity decreased, and the pooled results of all RVF indexes improved (Fig. 3). In addition, a funnel plot (Fig. 4) suggested that one of the 3 studies may have publication bias. Although there was a marked variation in the correlation coefficient for each RVF index across studies, the overall results indicated a strong relationship between CT and MRI measures of EDV (0.98 [0.98, 0.98]) (0.88 [0.86,0.90]), ESV (0.96 [0.96, 0.97]) (0.89 [0.87,0.91]), RVEF (0.98 [0.97, 0.99]) (0.83 [0.80,0.85]), and SV (0.97 [0.96, 0.97]) (0.84 [0.81,0.86]) (forest plots shown in Fig. 5).
Table 3 Pooled results and heterogeneity of the 4 indexes in our meta-analysis
| EDV | ESV | RVEF | SV | |||||
| Overall Results | Heterogeneity | Overall Results | Heterogeneity | Overall Results | Heterogeneity | Overall Results | Heterogeneity | |
| 19 studies | ||||||||
| 16 studies | ||||||||
| 3 studies | ||||||||
| Subgroup A | ||||||||
| Subgroup B | ||||||||
| EDV | ESV | RVEF | SV | |||||
| Overall Results | Heterogeneity | Overall Results | Heterogeneity | Overall Results | Heterogeneity | Overall Results | Heterogeneity | |
| 19 studies | 6.07(-1.96,14.10) | 15.6% | 6.01(-2.25,14.27) | 3.5% | -1.73(-6.59,3.12) | 0.0% | 0.90(-7.38,9.19) | 0.0% |
| 16 studies | 2.97(-4.96,10.91) | 0.0% | 4.05(-4.79,12.88) | 0.0% | -1.42(-6.93,4.10) | 0.0% | -0.05(-9.06,8.95) | 0.0% |
| 3 studies | 23.54(-5.42,52.50) | 0.0% | 17.16(-5.82,40.1) | 0.0% | -3.54(-12.01,4.93) | 0.0% | 6.51(-14.4,27.4) | 0.0% |
| Subgroup A | 6.66(-9.83,23.15) | 0.0% | 8.64(-6.50,23.78) | 0.0% | -3.17(-10.87,4.52) | 0.0% | 0.26(-12.7,13.2) | 0.0% |
| Subgroup B | 1.27(-8.50,11.05) | 0.0% | 1.68(-10.31,13.7) | 0.0% | -0.56(-8.44,7.33) | 0.0% | 0.36(-11.1,11.9) | 0.0% |
Fig. 3 a Forest plots of RVEF in all 19 studies; b forest plots of RVEF in 16 studies using disk summation; c forest plots of RVEF in 3 studies using 3D reconstruction; d forest plots of RVEF in subgroups based on slices of CT scanner of the 16 studies as defined before
Fig. 5 Forest plots of correlation coefficient of EDV, ESV, RVEF and SV measured by CT and MRI in all included studies
