The institutional human research committee approved this retrospective study and waived the need for informed consent. The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki. This study was conducted in a single hospital. The first study period was from October 2004 to September 2015. After obtaining the results of patients during the first study period (steps I–III), additional patient data for the second study period (step IV), from October 2015 to September 2018, were collected for verification. Children with proven pulmonary thromboembolism and underlying diffuse lung parenchyma disorders were excluded. In step I, we enrolled 40 patients with mean age of 16.7 years. In step II, we recruited 115 subjects with mean age of 9.1 years. In step III, we enrolled 78 patients with mean age of 10.6 years. Finally, in step IV, we enrolled 27 patients with mean age of 7.4 years. Patient demographics are presented in Table 1. Cardiac catheterization for mPAP measurement and CCT were performed within 3 months of each other, and no other intervention was performed between the two examinations.

Until June 2007, CCT was performed using a 16-slice scanner (LightSpeed 16; GE Healthcare, Milwaukee, WI, USA), followed by a 64-slice scanner (VCT; GE Healthcare) from July 2007 to October 2008, and another 64-slice scanner (SOMATOM Sensation 40; Siemens Healthineers, Erlangen, Germany) from November 2008 onwards. Data were acquired in the caudocranial direction, with a section thickness of 0.625–0.8 mm without gaps. The matrix size in the X–Y plane was 512 × 512 pixels. Nonionic iodinated contrast medium (2 mL/kg iopromide, Ultravist® 370; Bayer AG, Berlin, Germany) was delivered through a power injector to all patients at a flow rate of 0.8–4 mL/sec from newborn to adolescence, respectively. The delay time between the start of contrast medium administration and the start of imaging was determined using a bolus-tracking technique. A region in the ascending aorta was used for bolus-tracking; the threshold level was set at 150 HU. All cooperative patients were asked to hold their breath during scanning. Quiet respiration without crying was deemed acceptable in younger children and infants who could not comply with this request; therefore, patients aged < 5 years were routinely sedated with chloral hydrate (50 mg/kg) during examination. The mAs and kVp values were age-adjusted in order to reduce the radiation dose, following the as low as reasonably achievable principle (142728). The mean effective radiation doses ranged from 2.0 mSv in newborns to 15.6 mSv in teenagers. This range of radiation doses was lower than that reported in our previous study, although the two studies had 43 overlapping cases (14). Image reconstruction was performed through multi-planar reformatting on an independent workstation. A single reader with 23 years of experience in pediatric CCT assessed the CCT images and obtained the required measurements. The reader was blinded to the mPAP data.

An elevation in mPAP leads to progressive dilatation of the pulmonary trunk, bilateral pulmonary arteries, and right ventricle (RV) as well as RV hypertrophy (2930). As the RV pressure approaches that of the left ventricle (LV), the LV becomes compressed and exhibits increased pressure, causing LV hypertrophy (31). Clinically, we can also observe the dilatation of the inferior vena cava (IVC) during passive congestion due to PAH. Studies have reported the pulmonary trunk diameter (PTD) and PTD-to-ascending aorta diameter ratio (rPTAo) on CCT images to be positive predictors of underlying PAH in adults (2232). In step I, we tried to explore more indicators other than PTD and rPTAo in children (Fig. 1). The legend in Figure 1 provides a more comprehensive description of the measurements we performed.

Obtaining both the CCT and cardiac catheterization data of an individual with an absolutely normal heart is difficult, except in volunteers. The indications of this step I period for CCT were used to identify any unusual pathology limited by or uncertain in echocardiography in patients who had suspected PAH on the basis of clinical presentation. Therefore, patients with “normal” cardiac anatomy on CCT images with or without a final diagnosis of PAH were selected. The so-called “normal” cardiac anatomy on CCT images is defined by situs solitus, concordant atrioventricular and ventriculoarterial connection, no intracardiac defect, and no anatomical vascular anomaly at the great vessels connecting to the heart. These subjects received both CCT and right heart catheterization.

Children with complete normal cardiac anatomy rarely undergo cardiac catheterization for mPAP measurement. Thus, selection of individuals with a normal heart and an mPAP < 25 mm Hg determined through right cardiac catheterization may yield limited cases. Consequently, in this step II period, the inclusion criteria were extended to subjects with minor cardiovascular or bronchopulmonary problems who had no clinical presentation of any cardiovascular symptoms and signs. These CCT indications in this step II period ranged from checking for suspected coronary artery abnormalities, small intracardiac vegetations, any incidental shadows on echocardiograms or suspected bronchopulmonary anomalies that were finally cleared as “normal” cardiac anatomy, or anomalies that were not related to the target parameters we measured.

To reduce time, only the representative parameters from step I were measured. In this step II, by using the CCT-image parameters strongly correlated with the mPAP obtained in step I, we constructed “normal” reference curves of CCT images. To apply these reference curves to children, we adjusted the parameters according to their highest correlated body sizes, which varied widely.

In the step III period, by using the representative CCT-image parameters that were strongly correlated with mPAP and the most representative parameters for body size obtained in step II, we created formulas to predict mPAP in children. By using these formulas, we could set a threshold line on the “normal” reference curves constructed in step II to enable the diagnosis of PAH in children.

The number of cases included in our first study period in step I was insufficient. Therefore, we extended our inclusion criteria to also include patients with one of the different types of small ASDs, small ventricular septal defects, and partial anomalous pulmonary venous connections. The term “small” was not used to describe small size in the structure but to suggest that the shunting was limited hemodynamically. Usually, we could estimate the pulmonary (Qp) and aortic (Qs) flow from the velocity time integrals on echocardiography. If Qp/Qs was less than 1.5, the defect was thought to be small. The reason for extending the inclusion criteria in this manner was because the primary pathologies were far away from the measured PTD and rPTAo, as reported in the literature (2232). The indications for CCT of these added subjects in this step III were to check for the presence of any associated anomalies, such as coronary artery or airway anomalies, hidden in echocardiography, which was finally cleared. Right heart catheterizations were indicated in these additional patients for verifying cardiovascular anomalies found with echocardiography and demonstrating their hemodynamic significance for the decision to perform surgery on them.

In step IV, patients who had underlying cardiovascular anomalies directly related to the analyzed results from the preceding steps were excluded. All these patients were different from the patients of the previous steps I, II, and III. Indications for CCT were to check for coronary pathology, cardiovascular anomaly, and myocardial and pericardial abnormalities. By using the eligible patient data from CCT and right heart catheterization, we verified the proposed threshold lines and formulas from the preceding steps.

Descriptive statistics and the Student's t test were used to compare the mean measurements of the study and control groups. Relationships between parameters were checked by using 2-tailed Pearson correlation. Correlation coefficient (r) values > 0.8 were considered significant. Reference curves were plotted through general linear regression analysis according to the highest coefficient of determination (R²) values obtained in curve-fitting. Statistical analyses were performed using the commercially available software EXCEL (version 15, Microsoft, Redmond, WA, USA) and SPSS (version 16, SPSS Inc., Chicago, IL, USA). A formula for predicting mPAP by using the most correlated CCT-image parameters, with adjustments for the most correlated body size parameter of children, was constructed through surface regression (Centurion XVII, StatGraphics, Warrenton, VA, USA). P < 0.05 was considered significant.

CCT was used to check any subtle abnormality in the heart, thromboembolism, and lung parenchymal pathology that finally cleared as normal cardiopulmonary anatomy. The difference in the age distribution between the PAH and non-PAH groups was not significant (p = 0.28), whereas mPAP values differed significantly between these groups (p < 0.01). The parameters on the CCT images that correlated the most with mPAP were the PTD and rPTAo. The remaining 25 CCT parameters showed limited correlation with mPAP (Table 2).

Height was the independent variable that correlated the most with both PTD and rPTAo (r = 0.91 and −0.69 respectively; both p < 0.01). Other variables of children's body size had weaker correlations than height. The correlation coefficients to PTD were 0.82 for age, 0.82 for body surface area (BSA), and 0.75 for weight, respectively. The correlation coefficients to rPTAo were −0.61 for age, −0.56 for BSA, and −0.5 for weight, respectively. By using height as a single key independent variable, we established the normal reference curves of PTD and rPTAo, adjusted for variable body size in children. By using general linear regression analysis on the data with the best-fitting algorithm, we obtained a height versus PTD curve through quadratic curve estimation (PTD = 1.05635 × [height^0.59132], R² = 0.71, p < 0.001) (Fig. 2A) and a height versus rPTAo curve through linear estimation (rPTAo = −0.002134 × height + 1.343391, R² = 0.49, p < 0.001) (Fig. 2B).

However, height data were available for only 72 patients. We performed surface regression analysis to obtain the best-fitting surfaces and formulas to estimate mPAP (in mm Hg) by using PTD (in cm) or rPTAo which both took height (in cm) of children into consideration (Supplementary Fig. 1 in the online-only Data Supplement).

The resulting formulas to estimate mPAP were as follows:

These formulas were used to determine the value of PTD or rPTAo at different heights that had an estimated mPAP of 25 mm Hg. Thereafter, we could arbitrarily set solid threshold lines on the reference curves (Fig. 2). Thus, PAH could be straightforwardly diagnosed in patients in whom CCT parameters lay above the threshold lines, provided that their height was known. The formula could be written into a computer program for calculation in daily practice.

The CCT-image parameters we used in the first study period that predict mPAP were PTD and rPTAo. Therefore, in this step, we excluded patients with native pathologies at the great arteries (e.g., tetralogy of Fallot, transposition of the great arteries and pulmonary atresia, etc.) and patients who had undergone surgery directly on the great arteries.

From these reference curves, we could diagnose PAH in patients with parameters lying above the solid threshold lines. The sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of the threshold line of height versus PTD were 88.9%, 77.8%, 66.7%, 93.3%, and 81.5%, respectively. The sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of the threshold line of height versus rPTAo were 88.9%, 72.2%, 61.5%, 92.9%, and 77.8%, respectively (Supplementary Fig. 2 in the online-only Data Supplement).

By using our formulas, derived from step III, we estimated mPAP using PTD and height or rPTAo and height. Values of mPAP estimated using both formulas showed no significant difference (p < 0.05).