Between April 2015 and September 2016, 227 patients were considered for inclusion in this prospective study. In this study, we excluded patients with a left ventricular ejection fraction of 0.30 or less on transthoracic echocardiography before CCTA. Their serum creatinine level was obtained within 3 months prior to contrast-enhanced studies, and their estimated glomerular filtration rate (eGFR) was calculated using the modification of diet in renal disease formula of the Japanese Society of Nephrology (1314). Our inclusion criteria were suspected or confirmed coronary artery disease and referral for a CCTA study for clinical reasons, based on guidelines promulgated by the American College of Cardiology (15). We recorded the total body weight (TBW) to tailor the amount of CM used. We also recorded the patients' height for the calculation of other body parameters and other demographic data (Table 1).

As we excluded 5 patients with renal failure (eGFR less than 30 mL/min/1.73 m²), a history of allergic reactions to iodinated CM, or proven or suspected pregnancy, our final study population consisted of 222 patients. This included 102 males and 120 females, ranging in age from 40 years to 95 years (mean, 71.6 years); their TBW ranged from 30.0 kg to 83.0 kg (mean, 58.0 kg).

All patients were scanned on a 64-detector row CT scanner (Lightspeed VCT; GE Healthcare, Milwaukee, WI, USA); retrospective electrocardiography-triggered helical scans were performed. The CT scanning parameters were 0.35-second and rotation, 0.625-mm detector row width, 0.2 helical pitch (beam pitch), 8.0-mm table movement, 50-cm scan field-of-view (FOV), 100 kVp, and 400–770 mA. All scans were from the top of the left atrial appendage to the level of the inferior margin of the cardiac apex, in the craniocaudal direction. All patients were able to hold their breaths during the examination. Image reconstruction was performed in a 15- to 20-cm display FOV, depending on the patient's body size. Each patient was given nitroglycerin sublingually (0.3 mg) 5 minutes before scanning. Patients whose heart rates exceeded 65 beats per minutes after its administration additionally received landiolol hydrochloride (Corebeta; Ono pharmacological Co., Ltd., Osaka, Japan). The injection protocols are summarized in Table 2. We injected CM (iomeprol [Iomeron]; Eisai Co., Ltd., Osaka, Japan) through a 20-gauge catheter into the antecubital vein using a power injector (Dual Shot; Nemoto-Kyorindo Co., Ltd., Osaka, Japan). For the test bolus scanning, the CM was diluted (30% contrast material, 70% saline); the injection volume and rate were TBW × 0.6 mL and TBW × 0.05 mL/s administered for 12 seconds, respectively. For the CCTA scanning, the injection volume and rate were TBW × 0.6 and TBW × 0.05 mL/s administered for 12 seconds, respectively. CM delivery was followed by flushing with 20 mL of physiological saline at the same injection rate. To monitor the ascending aorta, we obtained dynamic low-dose (100 kVp, 50 mAs) scans; the interscan interval was 1.0 seconds. Acquisition of the dynamic monitoring scans began 10 seconds after the start of contrast injection. A region of interest (ROI) was placed inside the ascending aorta to obtain a time-attenuation curve for aortic peak-time measurements. We recorded aortic peak enhancement by constructing time-enhancement curves by connecting all time points. The arrival time in the ascending aorta was defined as the duration from the scan delay of the test bolus injection to the time of peak aortic enhancement. Using the arrival time data, the scan delay for CCTA was set at the arrival time plus 2.0 seconds post-injection (16).

The mean CT number (in HU) for the ascending aorta was recorded for all patients on a CT console monitor by placement of a circular ROI cursor; the ROI diameter ranged from 10 mm to 30 mm. CT numbers in the ascending aorta were measured on an unenhanced image of the test bolus with acquisition for the dynamic monitoring scans and subsequent CCTA with a standard kernel. Areas of calcification and artifacts were carefully excluded from the ROI. The degree of contrast enhancement was expressed as the change in the CT number (ΔHU) and was calculated by subtracting the CT number on unenhanced images from that on contrast-enhanced images of the ascending aorta.

As in earlier studies, factors with an effect on contrast enhancement, i.e., the patient's age, sex, TBW, and height, were recorded (1718). We acquired their age and sex from their electronic health records. Their TBW and height were obtained immediately prior to CCTA scanning. We measured the patients' cardiac output (CO) with a non-invasive CO monitor (Aesculon mini; Ospyka Medical, Berlin, Germany) that continuously displayed the CO; the average CO during 30 valid cardiac cycles was recorded.

We developed the GLM using a combination of the independent variables that had a significant effect on enhancement per gram of iodine on CCTA (ΔHUCCTA) (enhancement per gram of iodine on test bolus [ΔHUTEST] and TBW) in multivariate analysis, and also developed two conventional predicting models, using ΔHUTEST and TBW, as controls.

Previous reports have suggested that the vessel enhancement at the test bolus is linearly correlated to the vessel enhancement by the full bolus (910). Therefore, the predicted ΔHUCCTA in the following equation of a given ΔHUTEST (model 1) was calculated as follows: Pr=ΔHU/gIave×ΔHU/gItestΔHU/gItest-ave where ΔHU/gI_ave (ΔHUCCTA_ave) is the average of ΔHUCCTA (HU/grams of iodine [gI]), ΔHU/gI_test-ave (ΔHUTESTave) is the average of ΔHUTEST, and ΔHU/gI_test (ΔHUTEST) is a given ΔHUTEST (HU/gI). Previous reports have suggested that body size parameters, such as TBW, are inversely correlated to the enhancement by the fixed amount of contrast material (1219). Therefore, the predicted ΔHUCCTA of a given TBW (model 2) was determined as follows: Pr=ΔHU/gIave×TBWaveTBW where ΔHU/gI_ave is the average of ΔHUCCTA (HU/gI), TBW_ave (TBW_ave) is the average of TBW (kg), and TBW is a given TBW (kg).

We also developed a GLM to predict ΔHUCCTA using all independent variables (patients' age, sex, TBW, CO, ΔHUTEST, and peak time of test). With the aid of Akaike Information Criterion (AIC) analysis, we selected two independent variables (ΔHUCCTA and TBW) for the predictive model. Therefore, the GLM-predicted ΔHUCCTA, using a given ΔHUTEST and TBW (model 3), was determined as follows: Pr = e^{a × ΔHU/gI_test + b × TBW + c} where ΔHU/gI_test is a given ΔHUTEST (HU/gI), TBW is a given TBW (kg), a and b are the estimated coefficients, and c is a constant term.

Statistical analyses were performed with the free statistical software “R” (version 3.2.2; The R Project for Statistical Computing; http://www.r-project.org/). The relationship between ΔHUCCTA and the patient's age, sex, TBW, CO, ΔHUTEST, and the peak time with the test bolus was assessed by univariate linear regression analysis. We calculated the Pearson product-moment correlation coefficient (r) to determine the strength of associations. Welch's t test was used to compare ΔHUCCTA of males and females. We also performed multivariate regression analysis to determine independent and significant covariates that affected the ΔHUTEST values and calculated the standardized regression coefficient (β) to assess the strength of associations.

We developed predictive models using independent factors that had significant effects on ΔHUCCTA, and constructed GLMs using a combination of all the independent variables in multivariate analysis. The decision to include or exclude parameters in the final model was based on the AIC, a measure that is a function of both training error and complexity, because additional factors may result in a better mathematical fit that yields no additional biological information by overfitting to the training data.

To assess the validity of the models across various samples, we performed a 6-fold cross-validation; in this process, we trained the GLMs using 185 (37 × 5) patients and validated these models on another 37 patients. The correlation among the models with variables independently associated with ΔHUCCTA and GLM was assessed by calculating Pearson's correlation coefficient. Bland–Altman analysis was used to predict the contrast enhancement errors among all models. We calculated the residual values between the predicted values and the true values for all models. We compared residual values by using analysis of variance (ANOVA). When the residual value was significantly different according to ANOVA, we compared each model by using the t test with Holm post-hoc correction.

A p value of less than 0.05 was considered to indicate a statistically significant difference, and all interval estimations shown are 95% confidence intervals (CIs).

As shown in Figure 1A and F, univariate linear regression analysis revealed a correlation between the ΔHUCCTA and patients' age (r = 0.34), and ΔHUTEST (r = 0.75). The radiation dose for the dose-length product (mGy-cm) and scan duration of the test bolus were 3.8 ± 1.5 mGy-cm and 10.2 ± 4.2 seconds. There was an inverse correlation between the ΔHUCCTA and the height (r = 0.43), TBW (r = 0.67), and CO (r = 0.34) of patients (Fig. 1B–D) and their effect on the ΔHUCCTA (r = 0.69, p < 0.001 for all). We saw no significant correlation between the peak time of the test bolus and the ΔHUCCTA (r = 0.14) (Fig. 1E). The average ΔHUCCTA was significantly higher in females than in males (34.6 ± 7.2 vs. 29.3 ± 7.0 HU/gI, p < 0.001).

Multivariate linear regression analysis showed that only the TBW and ΔHUTEST retained their independent predictive value (p < 0.001) (Table 3). Calculation of the standardized regression coefficient revealed that the highest correlation between the ΔHUCCTA and independent variables was observed for the TBW (β = −0.303). The strength of association between the ΔHUTEST and the ΔHUCCTA value was β = 0.503.

Figures 2 and 3 show the results of the 6-fold cross-validation analysis. The highest correlation coefficient between ΔHUCCTA and the prediction values was seen in GLMs (r = 0.75), followed by TDC (r = 0.69) and TBW (r = 0.62). The lowest Bland–Altman limit of agreement was observed with GLMs (mean difference −0.0 ± 5.0 HU/gI, 95% CI: −10.1, 10.1 HU/gI), ΔHUCCTA (−0.0 ± 5.9 HU/gI, 95% CI: −11.9, 11.9 HU/gI), and TBW (1.1 ± 6.1 HU/gI, 95% CI: −11.1, 13.3 HU/gI) (Fig. 3). The residual values were 3.67 ± 3.46, 4.29 ± 4.10, and 4.78 ± 4.02 for the GLM, ΔHUTEST, and TBW. There was a significant difference in the residual value with the ANOVA test. In the post-hoc analysis, the residual values of the GLM were significantly lower than that of the TBW (p < 0.001) and ΔHUTEST (p < 0.001). Additionally, there was no significant difference between the residual value of the ΔHUTEST and TBW (p = 0.129).

Finally, we calculated the final parameters of three models. In this study, the ΔHUCCTA_ave was 32.2 ± 7.6 HU/gI, and the ΔHUTESTave was 44.3 ± 10.2 HU/gI; therefore, Pr was predicted by the following formula, Pr = 0.726 × ΔHUTEST with model 1. In this study, the ΔHUCCTA_ave was 32.2 ± 7.6 HU/gI, and the TBW_ave was 58.1 ± 10.4 kg; therefore, Pr was predicted by following formula, Pr = 1870.7 ÷ TBW with model 2. If they were estimated for all patients in this study, a was 0.012 (95% CI, 0.0011, 0.0013), b was −0.0076 (95% CI, 0.0065, 0.087), and c was 3.36 (95% CI, 3.35, 3.37); therefore, Pr was predicted by following formula, Pr = e^{0.012 × ΔHU/gI_test − 0.0076 × TBW + 3.36} with mode 3.

Figure 4 shows a representative case.