National Taiwan University Hospital Research Ethics Committee approved this retrospective study and waived the need for informed consent. This study was conducted at a single tertiary center with analysis of CCT images from July 2003 to January 2017 and clinical follow up until January 2019. The study was divided into three steps. Step I was to explore a simple corrective factor to adjust CCT measurements in children of various sizes. Based on the results from Step I, Step II involved building reference curves and a formula for normal LV quality on the proposed CCT measurements. Finally, Step III involved demonstrating the characteristics of LV quality (represented as a percentage to normal references) and tracing the outcomes using CCT measurements in patients with DRVs who received either BiVR or SVP.

Control subjects with a “normal” heart were included in the Step I and Step II control groups, which were used to establish normal references. The indications for CCT studies in the control group included airway problems, mediastinal lesions, or abnormal shadows on echocardiograms, which were eventually determined as ‘normal’ using CCT, or that were due to other minor confirmed pathologies on CCT images that did not affect the heart itself, such as small valvular vegetations. For the Step III group, patients with CHD with a DRV who underwent cardiac CCT were included in the preoperative and postoperative stages.

Three CCT scanner models were used for this study (LightSpeed 16: GE Medical Systems, Milwaukee, WI, USA, Jul 2003–Jun 2006; LightSpeed 64 VCT: GE Healthcare, Waukesha, WI, USA, Jul 2006–Oct 2008; Sensation 64: Siemens Medical Solutions, Forchheim, Germany, Nov 2008–Jan 2017). Scanning was performed using electrocardiography triggering to inhibit cardiac-related motion artifacts. Slice thickness ranged from 0.625–0.8 mm. The matrix size in the X–Y plane was 512 $\times$ 512 pixels. The X-ray tube current was adjusted according to patient body weight to follow the principle of “as low as reasonably achievable” [13, 21]. The mean effective radiation doses range from 2.0 mSv in newborns to 15.6 mSv in young adults [21]. A nonionic iodinated contrast medium was prescribed according to patient body weight. Chloral hydrate was administered to uncooperative patients (aged 5 years old).

Post-processing data quantification was performed using a commercial software (Syngo®; Siemens Medical Solutions, Forchheim, Germany). Three-dimensional volume-rendering images were used to quantify the LVEDV and LVMM. The MAA was measured in two dimensions. All measurements were performed during end-diastole (Fig. 1). A single reader with 23 years of experience in pediatric CCT interpretation assessed the CCT images and obtained the required measurements [22, 23, 24]. The reader was unaware of the participant outcomes while obtaining the measurements.

Fig. 1.

Left ventricular quality measured at end-diastole cardiac computed tomography images.

CCT-measured values (LVEDV, LVMM, and MAA) of the LV varied widely among children of different ages. Therefore, CCT measurements were adjusted for body size. Factors representing body size included age, height, weight, and body surface area (BSA). The correlation coefficients of each factor for all CCT measurements were checked. The factor with the highest correlation coefficient was used in Step II to adjust the normal LV quality in children of different ages and widely-varied LV sizes.

Using CCT-measured LV characteristics adjusted by the most significantly correlated body size factor obtained from Step I, we established the body-size-adjusted “normal range” references of all three measurements. LV quality was defined as the percentage of an individual’s measurements divided by body-size-adjusted normal values.

For Step III, three major patient groups were established from the study group with DRV: the double outlet right ventricle (DORV), unbalanced atrioventricular septal defect (ubAVSD), and hypoplastic left heart syndrome (HLHS) groups. We only included patients who reached their final BiVR or SVP status. Each patient’s LV quality was calculated during every examination, and the calculated data was marked on the figures of normal reference curves made in Step II to use for comparison (Fig. 2). This served to reveal any differences and tendencies in these two interventions (BiVR or SVP) in the present clinical practice, which could be represented by their own regression estimations. Finally, to explore the lowest limit of LV quality on the first visit that could predict survival from a final BiVR, we analyzed CCT measurements from patients with DRVs who had not undergone any interventions. The ranges of LV quality were compared between patients who survived and those who expired.

Fig. 2.

Mean regression curves of LV measurements in different groups. Black lines are the normal reference curves. Red lines and blue lines are mean regression curves of patients who survived after biventricular repair or single ventricle palliation, respectively. Scatter plots of LVEDV (A), LVMM (B), and MAA (C) relative to height in patients with a dominant right ventricle, who survived after their final surgical correction, against normal references (black regression lines). The patients who underwent BiVR are represented by red circles and those who underwent SVP are represented by blue triangles, with their regression estimations in red and blue lines. Solid markers denote measurements of patients before any intervention. The insets at the left upper corners of each figure show the case distribution in early childhood. LVEDV, left ventricular end-diastolic volume; LVMM, left ventricular myocardial mass; MAA, mitral annulus area.

Descriptive statistics and Student’s t-test were used to compare the mean measurements in the study and control groups. A two-tailed Pearson correlation coefficient (r) of $>$0.8 was considered significant. Reference curves were plotted using general linear regression analysis according to the highest coefficient of determination (R${}^2$) values obtained in the curve with the best fit. Statistical analyses were performed using SPSS (version 16.0, SPSS Inc., Chicago, IL, USA). p 0.05 was considered statistically significant.

We included 76 controls (age range: 27 days–20.7 years, mean 9.6 years; female: male = 25:51) with “normal” hearts (Table 1). Height exhibited the highest correlation with all measurements. The correlation coefficients of height with LVEDV, LVMM, and MAA were 0.93, 0.87, and 0.90, respectively, (p 0.001) (Supplementary Table 1).

The study group (N = 76) remained the same as that used in Step I. In Step I, height exhibited the strongest correlation with all three CCT measurements (LVEDV, LVMM, and MAA). By using “height” as an independent variable, we established the normal reference curves of the CCT-measured LV quality (black regression lines, Fig. 2) and their formulas as follows:

(1)$$\begin{array}{c}\hfill \text{𝐋𝐕𝐄𝐃𝐕}(\mathrm{mL})=0.0022826558\times {(\text{𝐡𝐞𝐢𝐠𝐡𝐭})}^{2.1316673352}\mathrm{\dots }[{\text{R}}^2=0.933]\hfill \\ \hfill \text{𝐋𝐕𝐌𝐌}(\mathrm{g})=0.0021273778\times {(\text{𝐡𝐞𝐢𝐠𝐡𝐭})}^{2.1748578761}\mathrm{\dots }[{\text{R}}^2=0.868]\hfill \\ \hfill \text{𝐌𝐀𝐀}\left({\mathrm{cm}}^2\right)=0.0069124642\times {(\text{𝐡𝐞𝐢𝐠𝐡𝐭})}^{1.3680602272\mathrm{\dots }}\mathrm{\dots }[{\text{R}}^2=0.858]\hfill \end{array}$$

The scatter plots of each patient’s LV quality are shown in Fig. 2. The plots show the regression estimations of LV quality in the two subgroups (BiVR vs. SVP) of patients who survived after their final surgical correction. Curves representing BiVR (in red) or SVP (in blue) differed significantly in all LV measurements.

From the regression curves, we found that the initial LVEDV of patients who received BiVR was similar to that of healthy subjects with shorter heights and of younger ages (Fig. 2A). The initial LVMM of the younger patients who received BiVR was slightly heavier than normal. However, this population had a slower increase in LV mass, which eventually became lighter than normal when the patients became taller or older (Fig. 2B). Patients who underwent BiVR always exhibited an MAA similar to that of healthy subjects, even after reaching adulthood (Fig. 2C). Those who initially received SVP had smaller LVEDV, lighter LVMM, and smaller MAA, did not “catch up” as they grew up, and always lagged below the normal range (Fig. 2). A comparison of the initial LV quality of the patients with DRV between BiVR (solid red circles in Fig. 2) and SVP (solid blue triangles in Fig. 2) survival revealed some overlapping LV measurements at similar body statuses. This representation reflects the current clinical practice. It is possible that some patients who do not have poor LV quality should still be considered for BiVR.

To explore the lowest limit of the LV quality that could survive BiVR, we included patients with a DRV who expired or survived after BiVR for further analysis (Table 2). In the DORV group, the mean of all three LV quality values were higher in the BiVR group than those in the SVP group (LVEDV, 92.9% vs. 71.0%; LVMM, 140.9% vs. 93.7%; and MAA, 86.2% vs. 63.3%); however, only MAA differed significantly between the groups (p 0.05). In the HLHS group, the LV quality of patients who expired after BiVR and of those who survived SVP did not differ significantly (LVEDV, 40.7% vs. 45.6%; LVMM, 55.7% vs. 65.5%; MAA, 44.7% vs. 41.5%). In the ubAVSD group, patients who survived after BiVR exhibited a considerably better LV quality than that of those who underwent SVP (LVEDV, 63.6% vs. 13.8%; LVMM, 87.3% vs. 31.7%; MAA, 78.7% vs. 35.0%). The patients with ubAVSD who expired after BiVR exhibited significantly lower LVEDV (15.9% vs. 63.6%; p 0.05) and LVMM (49.4% vs. 87.3%; p 0.05) than those of patients who survived; however, the MAA did not differ significantly (58.7% vs. 78.7%) from the survivors. The two patients with ubAVSD who expired after BiVR exhibited slightly higher initial mean LV quality than the single patient with ubAVSD who survived after SVP (LVEDV: 15.9% vs. 13.8%; LVMM: 49.4% vs. 31.7%; MAA: 58.7% vs. 35.0%).