Fetal echocardiography: z-score reference ranges for a large patient population

Transcription

1 Ultrasound Obstet Gynecol 2010; 35: 2 3 Published online 15 December 2009 in Wiley InterScience ( DOI: /uog.73 Fetal echocardiography: z-score reference ranges for a large patient population W. LEE*, T. RIGGS, V. AMULA, M. TSIMIS, N. CUTLER, R. BRONSTEEN* andc.h.comstock* *Department of Obstetrics and Gynecology, William Beaumont Hospital, Department of Pediatrics, Beaumont Children s Hospital and Division of Biostatistics, Beaumont Research Institute, Royal Oak and Wayne State University School of Medicine, Detroit, MI, USA KEYWORDS: biometry; fetal echocardiography; fetal measurements; z-scores ABSTRACT Objectives The main goal was to develop new z-score reference ranges for common fetal echocardiographic measurements from a large referral population. Methods A retrospective cross-sectional study of 2735 fetuses was performed for standard biometry (biparietal diameter (BPD) and femoral diaphysis length (FDL)) and an assessment of menstrual age (MA). Standardized fetal echocardiographic measurements included aortic valve annulus and pulmonary valve annulus diameters at endsystole, right and left ventricular diameters at end-diastole, and cardiac circumference from a four-chamber view of the heart during end-diastole. Normal z-score ranges were developed for these echocardiographic measurements using MA, BPD and FDL as independent variables. This was accomplished by using first standard regression analysis and then weighted regression of absolute residual values for each parameter in order to adjust for inconstant variance. Results A simple, linear regression model was the best description of the data in each case and correlations between fetal cardiac measurements and the independent variables were excellent. There was significant heteroscedasticity of standard deviation with increasing gestational age, which also could be modeled with simple linear regression. After this adjustment, the residuals conformed to a normal distribution, validating the calculation and interpretation of z-scores. Conclusion Development of reliable z-scores is possible for common fetal echocardiographic parameters by applying statistical methods that are based on a large sample size and weighted regression of absolute residuals in order to minimize the effect of heteroscedasticity. These normative ranges should be especially useful for the detection and monitoring of suspected fetal cardiac size and growth abnormalities. Copyright 2009 ISUOG. Published by John Wiley & Sons, Ltd. INTRODUCTION Current practice guidelines for fetal echocardiography are largely based on a qualitative assessment of heart structures using standardized procedures that may include anatomic measurements for suspected growth problems 1 3. Previous investigations have established normal percentile ranges, based on menstrual age (MA), for evaluating the size and growth of cardiac structures 6. This approach, however, poses practical challenges for healthcare professionals who need to interpret these measurements in growth restricted or macrosomic fetuses. An alternative approach for interpreting pediatric heart biometry by standardizing the distribution of parameter measurements has been proposed 7,. The z-score quantifies the degree to which an individual measurement lies above or below the mean value for a given population. From a practical perspective, it is much easier to interpret an aortic valve diameter that is 2.5 standard deviations (SDs) below the mean as opposed to simply knowing that a measurement is less than the 5 th percentile for somatic size or MA. This approach has been used to assess the size of fetal cardiac structures for congenital heart diseases with obstruction to ventricular outflow Earlier studies originally applied z-score measurements of the fetal heart on the basis of predicted values for MA,13. Schneider et al. 1 have addressed this limitation by developing prediction models for the computation Correspondence to: Dr W. Lee, Division of Fetal Imaging, Department of Obstetrics and Gynecology, William Beaumont Hospital, 3601 West Thirteen Mile Road, Royal Oak, MI , USA ( wlee@beaumont.edu) Accepted: 26 October 2009 Copyright 2009 ISUOG. Published by John Wiley & Sons, Ltd. ORIGINAL PAPER

2 z-score reference ranges 29 of z-scores in relation to fetal biometry. Natural log transformations were used for 17 different fetal heart measurements from 130 pregnancies at between 15 and 39 weeks MA. Regression models were based on fetal parameters that included femur length, biparietal diameter (BPD) and MA. Owing to the relatively small sample size they recommended caution when applying their regression results at the extremes of body size and gestation. The current investigation applied similar concepts to a much larger population of women with normal pregnancies. We have emphasized the use of specific statistical techniques to optimize regression models for the prediction of robust fetal cardiac z-scores. METHODS The study population was initially derived from a database of more than 5000 singleton fetuses that were scanned at William Beaumont Hospital, Royal Oak, MI from January 1993 to February From this number, we excluded any fetus with either cardiac or non-cardiac abnormalities, including fetal cardiac dysrhythmias, maternal diabetes, chromosomal abnormalities, fetuses that were small or large for MA and those having increased nuchal translucency thickness. Our exclusion criteria excluded a large number of fetuses, but we wanted to exclude all individuals with abnormalities even minor ones that potentially might affect fetal measurements. We also excluded any fetus with incomplete biometry. If more than one fetal echocardiogram was performed during a pregnancy, then only one of the scans was randomly selected for inclusion. Fetal echocardiographic studies were performed from standard scanning planes of the four-chamber view and arterial outflow tracts (Acuson Sequoia, Siemens Medical, Issaquah, WA; Voluson, GE Healthcare, Milwaukee, WI; and iu22, Philips Medical, Bothell, WA, USA). Electronic calipers were used to measure cardiac dimensions by experienced obstetrical sonographers and sonologists. Five cardiac parameters were measured. The diameters of the aortic (Ao) and pulmonary (Pa) valve annulus (inner edge to inner edge) were taken at end-systole. Figure 1 Composite image of fetal heart measurements, showing measurement of ventricular diameters (a), cardiac circumference (b), aortic valve annulus diameter (c) and pulmonary valve annulus diameter (d). End-diastolic measurements of the cardiac ventricles were obtained at the level of the mitral or tricuspid valve annulus. Cardiac circumference was also recorded at end-diastole. Aortic and pulmonary valve annulus diameters were taken at end-systole using inner edge-to-inner edge methodology.

3 30 Lee et al. (a) (b) Aortic valve annulus diameter (mm) Pulmonary valve annulus diameter (mm) (c) 20 (d) 20 Left ventricular diameter (mm) 16 Right ventricular diameter (mm) 16 (e) Cardiac circumference (mm) End-diastolic measurements of the right ventricle (RV) and left ventricle (LV) were obtained at the level of their respective atrioventricular valve annulus, and cardiac circumference (Circ) was also recorded at end-diastole (Figure 1). MA was based on the first day of the last normal menstrual period and confirmed by either first- or early second-trimester ultrasound scan. A normal last menstrual period was defined as regular cyclic menses without antecedent oral contraceptive use. Age estimates in the first trimester were based on crown rump length measurements 15. Age estimates in the second trimester were based on measurements of BPD, head circumference, abdominal circumference and femoral diaphysis length (FDL) Sonographic age was used Figure 2 Linear relationships between femoral diaphysis length (FDL) and aortic valve annulus diameter (a), pulmonary valve annulus diameter (b), left ventricular diameter (c), right ventricular diameter (d) and cardiac circumference (e) based on measurements of 2735 fetuses., predicted cardiac dimension as a function of FDL; , corresponding predictions for z-scores of ± 1, 2 and 3. to adjust MA if there was a discrepancy of more than 1 week between menstrual dating and sonographic assessment. Statistical analyses replicated the methodology outlined by Royston and Wright to determine the optimal model and z-scores for reference ranges of biological data 20,21. Cardiac dimensions were the dependent variables and MA, FDL and BPD were the independent variables. After plotting the data, we compared linear, quadratic, cubic and logarithmic models. If the more complex model did not result in significant improvement in the correlation coefficient, then the simpler model was used 22. The resulting residuals (differences between observed data and predicted values from the model) were examined and Kolmogorov Smirnov tests for normality were applied to determine whether they conformed to a normal, bellshaped distribution. If needed, the data were transformed and the new residuals were checked to determine whether they conformed to a normal distribution. Examination of residuals was crucial for the development of z-scores that would accurately predict the normal ranges. Data were also examined to determine whether the SD of the residuals varied across the range of values for the independent variable. Using the method suggested by

4 z-score reference ranges 31 Table 1 Linear models for prediction of fetal cardiac dimensions based on femoral diaphysis length, menstrual age and biparietal diameter Parameter Slope Intercept Correlation coefficient Femoral diaphysis length Aortic valve end-systolic diameter Pulmonary valve end-systolic diameter Left ventricular end-diastolic diameter Right ventricular end-diastolic diameter End-diastolic cardiac circumference Menstrual age Aortic valve end-systolic diameter Pulmonary valve end-systolic diameter Left ventricular end-diastolic diameter Right ventricular end-diastolic diameter End-diastolic cardiac circumference Biparietal diameter Aortic valve end-systolic diameter Pulmonary valve end-systolic diameter Left ventricular end-diastolic diameter Right ventricular end-diastolic diameter End-diastolic cardiac circumference The data apply for all measurements in mm except for menstrual age, which is given as weeks. Equations are linear, e.g. aortic valve end-systolic diameter = (0.119 FDL) 0.577; r = FDL, femoral diaphysis length. Aortic valve diameter residuals (mm) Femoral disphysis length (mm) Frequency Residuals Figure 3 Residuals (differences between observed data and value predicted from the model described in Figure 2) plotted along the range of femoral diaphysis length (FDL) values. There is a slight increase in variation as FDL increases; the method of weighted linear regression using absolute residuals was used to calculate how SD increased. The lines demonstrate one SD above and below the predicted value of aortic diameter. Aortic SD (mm) = (0.005 FDL); r = Altman 23 residuals were examined across the range of the independent variable, by creating five or six groups. If Levine s test showed a significant degree of inconstant variance, or heteroscedasticity, then weighted regression of absolute residuals was used to adjust the SD. RESULTS A total of 2735 subjects met the inclusion criteria. Their mean MA was 21.7 ±.2 (range, 17 1) weeks. A simple linear regression model was the best description of the data in each case; cubic and quadratic models did Figure Plot of residuals in Figure 3 compared with a hypothetical normal distribution. There is no significant difference, validating direct construction of z-scores without transformation of the data. Similar plots stratified by femoral diaphysis length category also conformed to a normal distribution. Kolmogorov Smirnov test accepts normal distribution: P = not improve upon linear models. All of the independent variables (MA, BPD, and FDL) strongly correlated with all cardiac parameters. Figure 2 demonstrates the close, linear relationship between FDL and Ao, Pa, and ventricular dimensions. The correlation between fetal biometry and echocardiographic parameters was excellent (Table 1). Figure 3 shows the residuals plotted as a function of FDL; the dispersion appears symmetrical above and below the midline. A slight, but significant increase in variation was noted along the range of FDL. By Levine s test for homogeneity, there was significant variation in SD for each echocardiographic parameter as FDL, MA or BPD increased, so the data were stratified into groups and weighted linear regression of the absolute

5 32 Lee et al. Table 2 Linear models for predicted standard deviations of fetal cardiac dimensions based on femoral diaphysis length, menstrual age and biparietal diameter Parameter Slope Intercept Correlation coefficient Femoral diaphysis length Aortic valve end-systolic diameter Pulmonary valve end-systolic diameter Left ventricular end-diastolic diameter Right ventricular end-diastolic diameter End-diastolic cardiac circumference Menstrual age Aortic valve end-systolic diameter Pulmonary valve end-systolic diameter Left ventricular end-diastolic diameter Right ventricular end-diastolic diameter End-diastolic cardiac circumference Biparietal diameter Aortic valve end-systolic diameter Pulmonary valve end-systolic diameter Left ventricular end-diastolic diameter Right ventricular end-diastolic diameter End-diastolic cardiac circumference The data apply for all measurements in mm except for menstrual age, which is given as weeks. Equations are linear, e.g. SD of aortic valve end-systolic diameter = (0.005 FDL) ; r = Estimates of SD are based on weighted linear regression of absolute residuals. FDL, femoral diaphysis length. residuals was used to determine the relationship between SD and each independent variable 23.Table2summarizes how SDs from the predicted values in Table 1 vary with each independent variable. Figure demonstrates the normal distribution of residuals a prerequisite for the development of reliable z-scores. Emphasis was placed on the FDL parameter because it is easy to obtain, is not sensitive to normal variations in head shape and it allows one to avoid using MA if dating criteria are not reliable. Hence, for the sake of brevity, corresponding figures for MA and BPD are provided as Supporting Information (Figures S1 S2) on the Journal website. Z-score nomograms and automated calculations are available on the Internet at DISCUSSION Fetal cardiac size and growth abnormalities may occur in a wide range of congenital heart diseases. This problem can be particularly pertinent in cases of severely obstructive cardiac lesions that have ductal dependency. Echocardiographic measurements can provide important objective information to complement both the anatomic evaluation and Doppler flow studies. In this report, new z-score reference ranges were used to characterize the relationship between an observed value and a reference standard as the number of SDs above or below the observed value with respect to a predicted value. This approach implies a normal distribution in which 2z (or SD) above or below the mean includes 95% of the population 2. Based on our results, the z-score can easily be calculated from an FDL measurement and a corresponding fetal echocardiographic parameter (e.g. Ao, Pa, RV, LV, or Circ) (see Appendix). The development of reliable z-score reference ranges depends on careful examination of the data to verify that variation above and below the predicted values either conforms to a normal distribution or is transformed mathematically so that it does conform in this manner. In the present study, we carefully compared the population sample with several linear regression models. The residuals (differences between the predicted and observed values) were analyzed to calculate SDs and then tested to determine whether their residuals conformed to a normal distribution. Our large sample size was associated with a high level of confidence in the accuracy of these new z-score prediction models. Schneider et al. 1 used a logarithmic transformation to model the relationship between cardiac dimensions and FDL, menstrual age or BPD. Our data set is considerably larger and appears to refute the non-linearity proposed by their results. Additionally, the multiplier factor in their model is very close to (and not significantly different from) 1.0 in many cases. Exponentiation of a log log equation shows that if the multiplier value is 1.0, then a log log model is mathematically equivalent to simple linear regression. A log log model constructs SDs that are constant on a logarithmic scale, but a constant value on a logarithmic scale represents a multiplying factor for SD as the independent variable increases. For example, as femur length increases from 2 to mm, Schneider s model 1 predicts a four-fold increase in SD for the aortic diameter, while in our model, it does not even double. More importantly, whatever model is used, analysis of residuals is crucial, that is, data should be evaluated across the range of FDL to determine whether and how SD varies, otherwise the denominator used in the calculation of a z-score may be inaccurate. Although there may be

6 z-score reference ranges 33 more biological variation in cardiac dimensions later in gestation, measurement error that is proportionately larger when the cardiac measurements are a few mm at 20 weeks gestation rather than 9 or 10 mm near term may modulate this trend. That is, the sum of biological variability and measurement error may be relatively constant throughout gestation, explaining the relatively slight increase in SD that we observed. Despite the absence of strict guidelines for estimating the required sample size for constructing a reference range, Royston 20 suggested restricting the standard error of the limits of the reference range to 10% of the SD. For reference ranges of ± 1, 2 or 3z (SD), the corresponding sample sizes were 150, 300 and 550. For our sample size of 2735, the standard error of the limits for a reference range of ± 2z would be 3.3%. By comparison, Schneider et al. 1 previously studied a smaller sample size of 0 pregnant women with a standard error that was nearly five times greater, or 15.%. We have previously described linear growth of the Ao and Pa with MA, with regression coefficients very similar to those in the present study 25. Previous studies by Steed et al. 26 yielded linear regression equations for LV, RV, Ao and Pa in terms of MA that are in very close agreement with ours (note that their measurements are in cm, so there is scaling factor of 10 ). Ruano et al. 27 constructed nomograms for pulmonary artery (PA) growth from 19 to 0 weeks gestation and their linear regression model (MPA = (0.30 MA)) for main pulmonary artery diameter in terms of MA is almost identical to ours (PA = (0.3 MA)). Shapiro et al. 2 derived fetal cardiac measurements from 1 weeks gestation to term and found a linear relationship between MA and both Ao and PA, with regression coefficients very similar to those in the present study. For RV and LV dimensions, they found significant non-linearity and used a regression model with a quadratic term. Their earlier evaluations and longer range for MA may account for the non-linearity in growth; their model had steep growth before 20 weeks gestation, although the slope became more linear thereafter. An important limitation to this retrospective study design was the logistical problem of ascertaining that all fetuses had been born with a normal heart, since pediatric echocardiography is not typically performed in all newborns. However, fetal echocardiography is highly correlative with postnatal and neonatal echocardiographic findings 29. Selection bias is always a potential concern in using data to establish reference standards, but we used a strict definition of normality and excluded many subjects with minor abnormalities (e.g. isolated premature beats). Out of an initial cohort of 5000 fetuses, only nine were excluded on the basis of being small-for-gestational age and only 10 were excluded on the basis of being large-for-gestational age. It is unlikely that these small numbers would materially alter the regression model or residual analyses. To exclude potential bias from serial measurements, information from each subject was used only once. To summarize, common fetal echocardiographic measurements were highly correlated to MA, BPD and FDL in our large population sample. Contrary to previously reported z-score ranges for fetal Ao, Pa, RV, LV and Circ measurements, significant linear relationships were found between these parameters and FDL. Weighted regression of absolute residuals, based on the properties of a half-normal distribution, was used to address the problem of inconstant variance for these measurements over time. This was a key step for minimizing the effect of heteroscedasticity on the development of robust z-score standards for fetal echocardiography. APPENDIX Clinical example for z-score calculation Femoraldiaphysislength(FDL)= 30 mm Aortic valve (Ao) diameter = 3.5 mm From Table 1 Predicted Ao diameter = ( ) = 3.0 mm From Table 2 Predicted SD = ( ) = 0. mm z = Ao observed Ao predicted Predicted SD = = 1.25 The aortic diameter z-score is SDs above the predicted mean for this femur length and therefore is within the normal range. REFERENCES 1. Rychik J, Ayres N, Cuneo B, Gotteiner N, Hornberger L, Spevak PJ, Van Der Veld M. American Society of Echocardiography guidelines and standards for performance of the fetal echocardiogram. J Am Soc Echocardiogr 200; 17: Lee W, Allan LD, Carvalho JS, Chaoui R, Copel J, Devore G, Hecher K, Munoz H, Nelson T, Paladini D, Yagel S. ISUOG consensus statement: what constitutes a fetal echocardiogram? Ultrasound Obstet Gynecol 200; 32: Wood D, Respondek-Liberska M, Puerto B, Weiner S. Perinatal echocardiography: protocols for evaluating the fetal and neonatal heart. JPerinatMed2009; 37: Devore GR, Siassi B, Platt LD. Fetal echocardiography. V. M- mode measurements of the aortic root and aortic valve in second- and third-trimester normal human fetuses. Am J Obstet Gynecol 195; 152: Sharland GK, Allan LD. Normal fetal cardiac measurements derived by cross-sectional echocardiography. Ultrasound Obstet Gynecol 1992; 2: Tan J, Silverman NH, Hoffman JI, Villegas M, Schmidt KG. Cardiac dimensions determined by cross-sectional echocardiography in the normal human fetus from 1 weeks to term. Am J Cardiol 1992; 70: Rimoldi HJA, Lev M. A note on the concept of normality and abnormality in quantitation of pathologic findings in congenital heart disease. Pediatr Clin North Am 1963; 10: Pettersen MD, Wei D, Skeens ME, Humes RA. Regression equations for calculation of z scores of cardiac structures in a large cohort of healthy infants, children, and adolescents: an echocardiograhic study. J Am Soc Echocardiogr 200; 21:

7 3 Lee et al. 9. Hornberger LK, Sanders SP, Rein AJ, Spevak PJ, Parness IA, Colan SD. Left heart obstructive lesions and left ventricular growth in the midtrimester fetus. A longitudinal study. Circulation 1995; 92: Hornberger LK, Lipshultz SE, Easley KA, Colan SD, Schwartz M, Kaplan S, Starc TJ, Ayres NA, Lai WW, Moodie DS, Kasten-Sportes C, Sanders SP. Cardiac structure and function in fetuses of mothers infected with HIV: the prospective PCHIV multicenter study. Am Heart J 2000; 10: Makikallio K, McElhinney DB, Levine JC, Marx GR, Colan SD, Marshall AC, Lock JE, Marcus EN, Tworetzky W. Fetal aortic valve stenosis and the evolution of hypoplastic left heart syndrome: patient selection for fetal intervention. Circulation 2006; 113: Salvin JW, McElhinney DB, Colan SD, Gauvreau K, del Nido PJ, Jenkins KJ, Lock JE, Tworetzky W. Fetal tricuspid valve size and growth as predictors of outcome in pulmonary atresia with intact ventricular septum. Pediatrics 2006; 11: e15 e DeVore G. Opinion. The use of z-scores in the analysis of fetal cardiac dimensions. Ultrasound Obstet Gynecol 2005; 26: Schneider C, McCrindle BW, Carvalho JS, Hornberger LK, McCarthy KP, Daubeney PE. Development of z-scores for fetal cardiac dimensions from echocardiography. Ultrasound Obstet Gynecol 2005; 26: Hadlock FP, Shah YP, Kanon DJ, Lindsey JV. Fetal crown rump length: reevaluation of relation to menstrual age (5 1 weeks) with high-resolution real-time US. Radiology 1992; 12: Hadlock FP, Deter RL, Harrist RB, Park SK. Fetal biparietal diameter: a critical re-evaluation of the relationship to menstrual age by means of real-time ultrasound. J Ultrasound Med 192; 1: Hadlock FP, Deter RL, Harrist RB, Park SK. Fetal head circumference: relation to menstrual age. AJR Am J Roentgenol 192; 13: Hadlock FP, Deter RL, Harrist RB, Park SK. Fetal abdominal circumference as a predictor of menstrual age. AJR Am J Roentgenol 192; 139: Warda A, Deter RL, Rossavik IK, Carpenter RJ, Hadlock FP. Fetal femur length: a critical re-evaluation of the relationship to menstrual age. Obstet Gynecol 195; 66: Royston P. Constructing time-specific reference ranges. Stat Med 1991; 10: Royston P, Wright EM. Goodness-of-fit statistics for agespecific reference intervals. Stat Med 2000; 19: Meng X-L, Rosenthal R, Rubin DB. Comparing correlated correlation coefficients. Psychol Bull 1992; 111: Altman DG. Construction of age-related reference centiles using absolute residuals. Stat Med 1993; : Silverwood RJ, Cole TJ. Statistical methods for constructing gestational age-related reference intervals and centile charts for fetal size. Ultrasound Obstet Gynecol 2007; 29: Comstock CH, Riggs T, Lee W, Kirk J. Pulmonary-to-aorta diameter ratio in the normal and abnormal fetal heart. Am J Obstet Gynecol 1991; 165: Steed RD, Strickland DM, Swanson MS, Hannon DW, McConnell ME, Dombroski RA, Webb SR. Normal fetal cardiac dimensions obtained by perpendicular imaging. Am J Cardiol 199; 1: Ruano R, de Fátima Yukie Maeda M, Niigaki JI, Zugaib M. Pulmonary artery diameters in healthy fetuses from 19 to 0 weeks gestation. J Ultrasound Med 2007; 26: Shapiro I, Degani S, Leibovitz Z, Ohel G, Tal Y, Abinader EG. Fetal cardiac measurements derived by transvaginal and transabdominal cross-sectional echocardiography from 1 weeks of gestation to term. Ultrasound Obstet Gynecol 199; : Berkley EMF, Goens MB, Karr S, Rappaport V. Utility of fetal echocardiography in postnatal management of infants with prenatally diagnosed congenital heart disease. Prenat Diagn 2009; 29: SUPPORTING INFORMATION ON THE INTERNET The following supporting information may be found in the online version of this article: The supplementary reference for this z-score calculator and graphing utility can be found at: Figure S1 Linear relationships between menstrual age and the aortic valve annulus (Aorta), pulmonary valve annulus (PA), left ventricular diameter (LV), right ventricular diameter (RV), and cardiac circumference (Circ) are shown for 2735 patients. The solid red lines show the predicted cardiac dimension as a function of menstrual age, while the dashed lines show the corresponding predictions for z-scores of ± 1, 2 and 3. Figure S2 Linear relationships between biparietal diameter (BPD) and the aortic valve annulus (Aorta), pulmonary valve annulus (PA), left ventricular diameter (LV), right ventricular diameter (RV), and cardiac circumference (Circ) are shown for 2735 patients. The solid red lines show the predicted cardiac dimension as a function of BPD, while the dashed lines show the corresponding predictions for z-scores of ± 1, 2 and 3.