Background
Clinicians rely on age- and size-specific measures of cardiac structures to diagnose cardiac disease. No universally accepted normative data exist for fetal cardiac structures, and most fetal cardiac centers do not use the same standards. The aim of this study was to derive predictive models for Z scores for 13 commonly evaluated fetal cardiac structures using a large heterogeneous population of fetuses without structural cardiac defects.
Methods
The study used archived normal fetal echocardiograms in representative fetuses aged 12 to 39 weeks. Thirteen cardiac dimensions were remeasured by a blinded echocardiographer from digitally stored clips. Studies with inadequate imaging views were excluded. Regression models were developed to relate each dimension to estimated gestational age (EGA) by dates, biparietal diameter, femur length, and estimated fetal weight by the Hadlock formula. Dimension outcomes were transformed (e.g., using the logarithm or square root) as necessary to meet the normality assumption. Higher order terms, quadratic or cubic, were added as needed to improve model fit. Information criteria and adjusted R 2 values were used to guide final model selection.
Results
Each Z -score equation is based on measurements derived from 296 to 414 unique fetuses. EGA yielded the best predictive model for the majority of dimensions; adjusted R 2 values ranged from 0.72 to 0.893. However, each of the other highly correlated ( r > 0.94) biometric parameters was an acceptable surrogate for EGA. In most cases, the best fitting model included squared and cubic terms to introduce curvilinearity.
Conclusions
For each dimension, models based on EGA provided the best fit for determining normal measurements of fetal cardiac structures. Nevertheless, other biometric parameters, including femur length, biparietal diameter, and estimated fetal weight provided results that were nearly as good. Comprehensive Z -score results are available on the basis of highly predictive models derived from gestational age or other biometrics as preferable for clinical reasons. These results supplant current equations and will provide a foundation for future multicenter collaborations.
Highlights
- •
Normative data for fetal cardiac structures are essential to diagnose abnormalities and make predictions about the need for surgical intervention.
- •
Previous studies are limited by the use of older methods such as videotape, small sample sizes, or a focus on obstetric parameters.
- •
This study highlights the best models for “ Z -score equations” for 13 key fetal cardiac structures as well as reasonable alternatives standardized to other biometric parameters.
High-quality normative data for two-dimensional measures of cardiac structures is useful in the accurate diagnosis of congenital heart disease in the fetus. This is particularly true in early gestation, when signs of valve and outflow tract abnormalities may be very subtle. In addition to diagnosing abnormalities, fetal cardiologists must predict and counsel about the most likely postnatal clinical course, including surgical repair and long-term outcomes. As part of clinical care and decision making, pediatric cardiologists and surgeons use “ Z -score” calculations of two-dimensional measurements of structures to make decisions about the type of surgery to perform. Universally accepted predictive models do not yet exist for fetuses, although a small body of literature suggests that they may be useful.
Calculation of Z scores normalizes cardiac structures to body size. This technique has become the standard method of assessing normality in pediatrics and obstetrics. Yet no comprehensive set of normative Z -score equations exist to express the size of fetal cardiac structures in the literature. Existing studies have been based on modest patient numbers, used M-mode measurements, studied only small numbers of parameters, or focused on parameters that would be more useful for obstetric purposes. The largest body of literature available examined absolute values of normal structures across gestation. Using our growing experience with fetal echocardiography, we sought to determine the best predictive models for normal cardiac structures across a broad range of fetal development.
We derived predictive models for Z scores for 13 commonly evaluated fetal cardiac structures using a large heterogeneous population of fetuses without significant structural cardiac disease.
Methods
This study was approved by the institutional review board at our institution.
Patient Selection
The echocardiography database at our hospital was queried to collect normal fetal echocardiographic reports from a representative population over a 1-year period (2009–2010) in subjects ranging from 12 to 39 weeks’ gestation, as determined by early obstetric ultrasound dating. Both singleton and twin fetuses were included. Excluded were fetuses with structural cardiac disease. Selection was performed using a search engine with finding codes for “normal fetal echocardiogram” and “no structural fetal cardiac disease.” These terms would exclude any fetus with a definitive diagnosis of congenital heart defect. If a fetus in the cohort had multiple measurements, only the first study was used.
Echocardiography
Echocardiography was performed using a curvilinear C4-7 MHz probe on either a Vivid 7 (GE Healthcare, Milwaukee, WI) or a Philips iE33 (Philips Medical Systems, Best, The Netherlands) echocardiographic machine with fetal preset. A standard fetal echocardiographic imaging protocol is used at our institution on the basis of the guidelines of the American Society of Echocardiography. Measurements were performed by a cardiologist with specialized training in fetal echocardiography (J.I.P.), who was blinded to demographics and to the prior measurements. Studies with substandard imaging planes were not included (5%–32%). Measurements were performed on a stand-alone digital workstation (Philips Medical Systems, Andover, MA). All fetal echocardiograms coded as normal or as depicting no structural cardiac defects were considered for inclusion in the study. Estimated gestational age (EGA) at our institution is taken from the determination of the referring maternal fetal medicine specialist or in some cases obstetrician and is calculated by early ultrasound dating or last menstrual period. The most appropriate measurement is determined by the referring physician, and in this case an assumption was made that they use national standards for dating. Femur length (FL), biparietal diameter (BPD), and abdominal circumference (incorporated into fetal weight [FW]) were measured using standard methods, and a calculated gestational age if performed at the time of the study was also recorded from these measurements. FW was calculated using the Hadlock formula. Although postnatal follow-up could not be obtained, diagnostic accuracy of fetal diagnosis for major cardiac defects is 97%.
Structures Assessed
A total of 13 cardiac dimensions were evaluated: aortic valve annulus (AoVA), pulmonary valve annulus (PVA), tricuspid valve annulus (TVA), mitral valve annulus (MVA), main pulmonary artery (MPA), ascending aorta (Asc Ao), descending aorta (Dsc Ao), aortic isthmus (AoI), patent ductus arteriosus (PDA), left pulmonary artery (LPA), right pulmonary artery (RPA), left ventricular length (LVL), and right ventricular length (RVL). Examples of echocardiographic images illustrating the measured structures can be found in Figures 1–6 . Measurements of the AoVA, PVA, MVA, TVA, LVL, and RVL were performed in the four-chamber transverse view. Measurements for the Asc Ao diameter, Dsc Ao diameter, and AoI were performed in the sagittal view. Measurements for the MPA, RPA, LPA, and PDA were performed in either the sagittal or short-axis view, and standard imaging planes used are detailed in guidelines from the American Society of Echocardiography. Measurements were made from inner edge to inner edge. Measurements of the MVA, TVA, LVL, and right ventricular length were made in end-diastole and of the AoVA and PVA in end-systole. End-diastole and end-systole were determined by identifying the onset of atrioventricular valve closure and semilunar valve closure, respectively.
Data Analysis
Statistical analysis was performed using the regress procedure in Stata (StataCorp, College Station, TX).
Before regression analysis, an assessment of the correlation of EGA with calculated gestational age by biometric parameters was performed for the study population. Correlation coefficients were calculated.
Linear regression analyses were used to model the relationship between the sizes of typical cardiac structures by indicators of stage of development. We used robust regression to reduce the impact of extreme values and to improve model fit in the face of modest departure from the normality after data transformation. Before implementing analyses, we evaluated the normality assumption for each dimension (dependent variable) and when necessary implemented normalizing transformations (e.g., logarithm or square root) and evaluated the correlation among predictors (independent variables). Modeling of the relationship between dimension measurements and predictors followed, including an assessment of the need to include higher order, quadratic, and cubic terms to introduce curvilinearity. The decision to include a higher order term in a model first driven by whether the term itself made a statistically significant difference. However, the sample size was large, and thus most such additions were statistically significant. That left us in some cases with the question of determining which among statistically significant alternative models provided the best fit, and that was when we used both visual inspection, R 2 , and the Akaike information criterion to assist help determine the best fitting model.
Results
Studies and Demographics
The original search criteria yielded 504 studies (389 were done on fetuses with only one study and 115 on fetuses with multiple measurements). One hundred sixty-nine studies were selected as “no structural fetal cardiac disease,” and 335 were selected via the “normal fetal echocardiogram” code. Indications for the studies performed are highlighted in Table 1 . The gestational age range of the studies is provided in Table 2 . Cardiac dimensions were measured in normal fetal echocardiograms. The mean EGA was 24.04 ± 5.4 weeks (range, 12–39, weeks). Calculated gestational age from biometrics had strong correlation with obstetric determination of gestational age that is typically by early ultrasound dating ( Figure 7 ). The relationship of EGA to calculated gestational age had an adjusted R 2 value of 0.98, with a standard error of 0.81 ( Figure 7 ). The total number of fetuses studied for each parameter is represented in Table 3 . Measured fetal cardiac dimensions were highly correlated with all of the independent variables.
| Indication for fetal echocardiography | Number of echocardiograms ( n = 504) |
|---|---|
| Unclear images of heart | 32 |
| Twin with heart defect | 1 |
| Twin gestation | 7 |
| Trisomy 21 | 1 |
| Connective tissue antibodies in mother | 19 |
| Single umbilical artery | 2 |
| Abnormal nuchal translucency | 54 |
| Family history of heart defect | 127 |
| Echogenic focus | 22 |
| Arrhythmia | 1 |
| Medication exposure | 9 |
| Fetus of diabetic mother | 84 |
| Abnormal ultrasound finding | 62 |
| Abnormal rhythm | 67 |
| Advanced maternal age | 2 |
| Cystic hygroma | 1 |
| Gastroschisis | 1 |
| Mosaic Turner syndrome | 1 |
| Omphalocele | 1 |
| Increased risk for trisomy 21 | 1 |
| Abnormal amniocentesis | 1 |
| Hereditary disease possibly affecting fetus | 1 |
| Prior study with valve regurgitation | 1 |
| Suspected atrial septal defect | 1 |
| Abnormal fetal echocardiogram | 5 |
| Gestational age (wk) | Number of echocardiograms |
|---|---|
| 12 | 2 |
| 13 | 3 |
| 14 | 5 |
| 15 | 9 |
| 16 | 10 |
| 17 | 5 |
| 18 | 17 |
| 19 | 39 |
| 20 | 55 |
| 21 | 48 |
| 22 | 33 |
| 23 | 40 |
| 24 | 25 |
| 25 | 34 |
| 26 | 19 |
| 27 | 29 |
| 28 | 25 |
| 29 | 13 |
| 30 | 21 |
| 31 | 12 |
| 32 | 16 |
| 33 | 12 |
| 34 | 8 |
| 35 | 13 |
| 36 | 4 |
| 37 | 2 |
| 38 | 4 |
| 39 | 1 |
| Parameter | Biometric parameter | Number of subjects |
|---|---|---|
| AoVA | EGA | 402 |
| BPD | 394 | |
| FL | 388 | |
| EFW | 390 | |
| MVA | EGA | 405 |
| BPD | 397 | |
| FL | 390 | |
| EFW | 379 | |
| PVA | EGA | 402 |
| BPD | 394 | |
| FL | 401 | |
| EFW | 377 | |
| TVA | EGA | 407 |
| BPD | 399 | |
| FL | 392 | |
| EFW | 381 | |
| MPA | EGA | 316 |
| BPD | 310 | |
| FL | 305 | |
| EFW | 296 | |
| LPA | EGA | 336 |
| BPD | 329 | |
| FL | 324 | |
| EFW | 319 | |
| RPA | EGA | 338 |
| BPD | 333 | |
| FL | 358 | |
| EFW | 322 | |
| AoI | EGA | 391 |
| BPD | 384 | |
| FL | 378 | |
| EFW | 368 | |
| PDA | EGA | 370 |
| BPD | 363 | |
| FL | 358 | |
| EFW | 346 | |
| Asc Ao | EGA | 407 |
| BPD | 398 | |
| FL | 392 | |
| EFW | 381 | |
| Dsc Ao | EGA | 407 |
| BPD | 399 | |
| FL | 393 | |
| EFW | 381 | |
| LVL | EGA | 414 |
| BPD | 406 | |
| FL | 401 | |
| EFW | 390 | |
| RVL | EGA | 414 |
| BPD | 406 | |
| FL | 401 | |
| EFW | 390 |
For analysis, fetuses with extensive missing data or images that could not be viewed in the archive were excluded, and if more than one study was done, only the first study was included. This yielded a total of 435 unique fetuses/studies.
There was a broad range of study indications, as no indications were excluded. Indications for echocardiography are described in Table 1 , with most common indication being a family history of cardiac defect. A total of 14 sets of twins were included in the study cohort. Chorionicity and amnionicity were not noted for all the twins in the echocardiography database.
Models
Most dimensions required a normalizing transformation, primarily a log transform to reduce skewing, to meet the normality assumption. The correlation between any two predictors was >0.94, thus precluding, because of serious colinearity, the use of multiple predictors in the same model. The form used in modeling predictors of each dimension, the complexity (order) of the model, as well as model characteristics are shown in Tables 4A–4D . The methods and comparative adjusted R 2 values as well as Akaike information criterion values for alternative models are shown in Table 5 . The best model for all parameters except MPA used EGA; for MPA, FL was superior to EGA. In the event that one parameter is unavailable or less reliable for a specific clinical scenario such as growth restriction or limb abnormality, relationships for all independent variables (FL, BPD, FW, and EGA) are provided ( Tables 4A–4D ). A typical curve illustrating the quality of fit for models is shown in Figure 8 .
| Structure | Root MSE | Best fitting equation |
|---|---|---|
| AoVA | 0.1082 | ln(AoVA) = −5.4177 +(0.3589) × EGA + (−9.75e-03) × EGA 2 +(9.75e-05) × EGA 3 |
| PVA | 0.1222 | ln(PVA) = −5.1029 + (0.3475) × EGA + (−9.33e-03) × EGA 2 + (9.25e-05) × EGA 3 |
| TVA | 0.1399 | ln(TVA) = −4.7678 + (0.3596) × EGA + (−9.42e-03) × EGA 2 + (8.92e-05) × EGA 3 |
| MVA | 0.1303 | ln(MVA) = −5.0632 + (0.3932) × EGA + (−1.08e-02) × EGA 2 + (1.05e-04) × EGA 3 |
| MPA | 0.1248 | ln(MPA) = −5.7996 + (0.4392) × EGA + (−0.0131) × EGA 2 + (1.43e-04) × EGA 3 |
| Asc Ao | 0.1417 | ln(Asc Ao) = −4.8807 +(0.3400) × EGA + (−9.94e-03) × EGA 2 +(1.11e-04) × EGA 3 |
| Dsc Ao | 0.1397 | ln(Dsc Ao) = −4.1357 + (0.2572) × EGA + (−6.73e-03) × EGA 2 + (6.89e-05) × EGA 3 |
| AoI | 0.0339 | AoI = −0.0556 +(0.0141) × EGA |
| PDA | 0.0605 | PDA = −0.6004 + (0.0884) × EGA + (−2.73e-03) × EGA 2 + (3.51e-05) × EGA 3 |
| LPA | 0.0335 | LPA = −0.0366 + (0.0134) × EGA |
| RPA | 0.0383 | sqrt(RPA) = 0.2409 + (0.0123) × EGA |
| LVL | 0.0847 | sqrt(LVL) = −0.2551 + (0.0898) × EGA + (−8.62e-04) × EGA 2 |
| RVL | 0.0888 | sqrt(RVL) = −0.2694 + (0.0856) × EGA + (−7.80e-04) × EGA 2 |
| Structure | Root MSE | Best fitting equation |
|---|---|---|
| AoVA | 0.1110 | ln(AoVA) = −3.9474 + (0.9568) × BPD + (−0.1089) × BPD 2 + (4.91e-03) × BPD 3 |
| PVA | 0.1251 | ln(PVA) = −3.7118 + (0.9500) × BPD + (−0.1084) × BPD 2 + (4.91e-03) × BPD 3 |
| TVA | 0.1431 | ln(TVA) = −3.2043 + (0.9325) × BPD + (−0.1016) × BPD 2 + (4.37e-03) × BPD 3 |
| MVA | 0.1351 | ln(MVA) = −3.3248 + (0.9699) × BPD + (−0.1069) × BPD 2 + (4.47e-03) × BPD 3 |
| MPA | 0.1207 | ln(MPA) = −3.8461 + (1.0486) × BPD + (−0.1278) × BPD 2 + (6.06e-03) × BPD 3 |
| Asc Ao | 0.1454 | ln(Asc Ao) = −3.5165 + (0.9028) × BPD + (−0.1113) × BPD 2 + (5.55e-03) × BPD 3 |
| Dsc Ao | 0.1441 | ln(Dsc Ao) = −2.8563 + (0.5808) × BPD + (−0.0569) × BPD 2 + (2.51e-03) × BPD 3 |
| AoI | 0.0359 | AoI = −0.1064 + (0.1147) × BPD + (−0.0127) × BPD 2 + (7.50e-04) × BPD 3 |
| PDA | 0.0616 | PDA = −0.2274 + (0.2229) × BPD + (−0.0286) × BPD 2 + (1.66e-03) × BPD 3 |
| LPA | 0.0351 | LPA = −0.1345 + (0.1386) × BPD + (−0.0176) × BPD 2 + (1.05e-03) × BPD 3 |
| RPA | 0.0392 | sqrt(RPA) = 0.2934 + (0.0416) × BPD |
| LVL | 0.0936 | sqrt(LVL) = 0.2949 + (0.2176) × BPD + (−5.14e-03) × BPD 2 |
| RVL | 0.0995 | sqrt(RVL) = 0.2634 + (0.2067) × BPD + (−4.31e-03) × BPD 2 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
