Study population

A cross-sectional study was conducted to include women with normal pregnancies (19-42 weeks of gestation) by searching our database of patients enrolled into protocols that included examination of the fetal heart by 3- and 4-dimensional ultrasound. Women were considered to have normal pregnancies if they had no medical or obstetrical complications, carried a singleton fetus without chromosomal or congenital anomalies, were dated by either a first- or second-trimester ultrasound scan, and delivered an appropriate-for-gestational-age neonate at term (≥37 weeks of gestation) with an uncomplicated neonatal course.

All women provided written informed consent prior to undergoing sonographic examination. Participation was approved by the Institutional Review Board of the National Institute of Child Health and Human Development and the Human Investigation Committee of Wayne State University.

Examination technique

Ultrasound examinations were performed by 6 experienced sonographers using systems with STIC capability (Voluson 730 Expert, Voluson E8 Expert; GE Medical Systems, Kretztechnik GmbH, Zipf, Austria) and utilizing a motorized curved array transabdominal transducer (2-5 or 4-8 MHz). A transverse view of the fetal chest at the level of the 4-chamber view was obtained, from which STIC datasets were acquired. The transducer was oriented so that the fetal spine was positioned posteriorly for each acquisition. Acquisition time was 10 seconds with a sweep angle that was sufficient to encompass the fetal cardiac structures (25-35°). Color Doppler sonography was not utilized during the acquisition process. Adequate cardiac datasets were accepted for postprocessing if acoustic shadowing (signal loss in the sound path secondary to echogenic structures), dropout (signal loss in the sound path without intervening structures), and motion artifact were absent.

Cardiac datasets were acquired to investigate the following fetal cardiovascular parameters: (1) ventricular volume; (2) stroke volume (end-diastolic volume – end-systolic volume); (3) cardiac output (stroke volume × fetal heart rate); and (4) ejection fraction (stroke volume/end diastolic volume × 100%). Fetal biometric measurements of the biparietal diameter (BPD), head circumference (HC), abdominal circumference (AC), and femoral diaphysis length (FL) were obtained utilizing 2D sonography at the time of cardiac dataset collection. Cardiac output was expressed both as a function of estimated fetal weight and as a function of biometric parameters (HC, AC, FL).

Analysis was performed offline (4D View versions 5.0-7.0; GE Healthcare, Milwaukee, WI) in a standardized manner. In the A plane of the multiplanar display, the fetal heart was reoriented such that the left ventricle was located on the left side of the screen with the apex of the heart directed upward. The interventricular septum was then rotated to 90° in both the A plane and the C plane. The AV valves were located by scrolling from front to back in the A plane. The image was then optimized by selecting Chroma Color 1 (Sepia) and SRI 5. After image brightness and contrast settings were optimized, end-systole and end-diastole were identified by scrolling through each frame and identifying the image preceding the AV valve opening (systole) and following AV valve closure (diastole) ( Supplemental Video Clip 1 ).

Cardiac ventricular volumes were calculated in a semiautomated fashion utilizing virtual organ computer-aided analysis (VOCAL). VOCAL II was selected and the Contour Finder: Trace option was utilized with 15° of rotation and a sensitivity of 1 (default = 5). The image was enlarged and the reference dot repositioned into the ventricle of interest. Because of the complex geometry of the ventricles, the location of the reference dot within the ventricle was selected to meet the software requirement that the contour only cross the rotation line twice. With these selections, 12 rotational steps were made and a volume was computed. Datasets were accepted for analysis if the ventricular septum, ventricular walls, and AV valves were visible throughout each rotational step ( Supplemental Video Clip 2 ).

We previously reported the repeatability and reproducibility of cardiac ventricular volume measurements utilizing this technique. Volume measurements were repeatable with good agreement (coefficient of variation [CV], <10%) and excellent reliability (intraclass correlation [ICC], >0.95) for both intraobserver and interobserver measurements. Additionally, ventricular volumes were reproducible with negligible difference in agreement (CV <1%), good reliability (ICC >0.9), and minimal bias (mean percent difference, −0.4%; 95% limits of agreement, −5.4% to 5.9%) when different STIC datasets for the same patient were compared.

Statistical analysis

Data were first assessed using numerical and graphical techniques, including scatter plots of each response vs gestational age, to determine whether they met the assumptions of the statistical tests being used to analyze them. All but 2 scatter plots revealed the presence of curvilinear relationships and heteroscedasticity; hence, natural logarithmic transformations (from the Box-Cox family of transformations) of each response and gestational age were performed to linearize the data and correct for heteroscedasticity.

Weighted regression analysis was performed using these transformed data for each of the models to address heteroscedasticity. The weights were calculated as 1/(predicted value of response × 1.253) because they are the best linear unbiased estimates (BLUE) of each of the true responses. The weights were multiplied by √(π/2) = 1.253, using a half-standard normal distribution. Because all of the responses are a function of gestational age, the residuals should have a normal distribution at each value of gestational age, and the absolute values of the residuals should have a half-normal distribution. It follows that the mean of the absolute residuals multiplied by √(π/2) is an estimate of the SD of the residuals. If the SD is not fairly constant for each parameter, their predicted values from regressing absolute residuals against gestational age multiplied by √(π/2) will provide age-specific estimates of the SD of the signed residuals, and hence of the response.

For each model, studentized residuals from the weighted regression analysis were assessed for normality using a normal probability plot. Main-effects polynomial weighted regression models were fitted to the data to capture the curvature in the data. Because the 2 scatter plots of ejection fraction for the left and right ventricles did not indicate curvilinear relationships or the presence of heteroscedasticity, simple linear regression analysis was used to evaluate the mean change in ejection fraction for each ventricle due to gestational age.

Linear and/or quadratic terms for gestational age in the model typically provided a good fit to the data, as evidenced by random patterns in the residual analyses. The quadratic term was not dropped from the model whenever it was found to be nonsignificant because it provided a better fit to the data than its corresponding simplified model with only the linear term in it. Model P values and corresponding adjusted R-squares were reported. The 5th, 50th, and 95th percentiles were computed on the back-transformed data and weighted regression equations.

For bivariate analysis, the Shapiro-Wilk and Kolmogorov-Smirnov tests were used to test for normal distribution. Because the data were not normally distributed, nonparametric statistics were employed. The Wilcoxon signed-rank test was used to determine the difference between paired variables, and Spearman’s rank correlation coefficient (r _s ) was utilized to assess correlations. A P < .05 was considered statistically significant for all comparisons. Statistical analyses were performed with SPSS package version 14 (SPSS Inc, Chicago IL) as well as The SAS System for Windows version 9.2 (SAS Institute Inc, Cary, NC).