Normalization of echocardiographically derived paediatric cardiac dimensions to body surface area: time for a standardized approach

The quantification of cardiac dimensions derived from echocardiography is important in paediatric cardiac practice. Evaluation of the size and growth of cardiac chambers, valves, and great vessels plays a key role in the management of congenital heart disease, from the initial decision-making in the neonatal period to the nature and timing of subsequent interventions. It may also be important in the assessment and risk stratification of children with ‘acquired’ heart disease such as hypertrophic cardiomyopathy or coronary artery involvement in Kawasaki disease. Body size and cardiac dimensions change dramatically during normal growth and development. Therefore, it is necessary to place in context the measured size of a given cardiac structure by correcting for body size, through the process of normalization. The paper by Neilan et al.1 in this issue of the Journal is a timely addition to the literature on normalization in paediatric populations, and raises a number of important issues. Various algebraic formulae to describe the relationship between cardiac dimensions ( y ) and body size ( x ) have been suggested. Initial studies assumed a linear relationship, expressed as y = mx or y = mx + c (where c is the value of y -intercept). 2 , 3 However, such linear relationships have not been found in childhood. More recent studies have used allometric models, in which a non-linear relationship is assumed. 4–6 These relationships can be expressed in the form of y = ax b (where a is the scaling factor and b the scaling exponent) or the natural logarithmic form ln( y ) = ln( a ) + ( b ) ln( x ). There is now good evidence, supported by the study in this issue, that the latter models most accurately represent the relationship between body size and cardiac dimensions, without heteroscedasticity (non-constant variance). Sluysman and Colan 6 suggested in 2005 that a scaling exponent of 0.5 provides the best index for cardiac and vascular diameters, a prediction that appears to hold true in other studies. 4 Interestingly, however, they found that a linear relationship best predicted cardiac and vascular areas. 6 An issue that remains to be resolved is which is the most appropriate measure of body size to use in normalization equations? Several parameters have been suggested, including weight, height, body surface area 4–10 (estimated from published formulae), 11 , 12 body mass index, 1 and, in the fetus, femur length or biparietal diameter. 13 Most paediatric studies have found that body surface area correlates most closely with cardiac dimensions. Neilan et al. 's study has suggested that correcting for weight provides the most accurate correlation with left atrial size, 1 but it remains to be seen whether this holds true for other cardiac dimensions. Normalization allows clinicians to quantify not only whether a given cardiac dimension falls within the normal range but also to monitor whether temporal changes are related to normal body growth or caused by pathological processes. When using normalized values, it would seem most appropriate that these are derived using the same modality and techniques. Therefore, for echocardiographic parameters, normal values derived from echocardiographic studies should be used, as these differ, for example, from normal values derived from cardiac magnetic resonance studies. This is particularly important in congenital heart disease patients, as the normal values most commonly used by cardiac surgeons to assess the size of cardiac and vascular structures intra-operatively are derived from formalin-fixed heart specimens examined post-mortem. 14 As virtually all cardiac structures shrink during formalin fixation, there is a significant discrepancy between these values and those derived in vivo using non-invasive imaging techniques. 15 As a result, it has been suggested that in the current era, echocardiographically derived normal values should be used as the gold standard on which to base management decisions. 4 Long-term follow-up studies based on such values are required to confirm whether this approach is appropriate. Cardiac dimensions have been expressed as a percentage of the mean normal value, but this does not take into account normal variation. Another commonly used method for expressing normal values is the use of centile charts. Although these are useful for determining where a given value lies within the population, values that lie above or below the uppermost or lowermost centile lines, respectively, cannot be quantified. This limitation is overcome by the use of z -scores (i.e. the number of standard deviations a measurement departs from the normal mean). The use of z -scores enables a more precise assessment of cardiac dimensions, which is particularly important in conditions associated with very small chamber dimensions (e.g. hypoplastic left heart syndrome, pulmonary atresia) or very large measurements (e.g. wall thickness in hypertrophic cardiomyopathy). Furthermore, in the research setting, z -scores are easily manipulated mathematically to allow statistical comparisons to be made between subgroups of patients. Z-scores can be expressed as nomograms or used to create computer spreadsheets that generate a z -score when the body size and cardiac dimension variables are entered, 16 providing a very simple means for comparing echocardiographic measurements with normal values. In recent years, there have been a number of studies reporting normal echocardiographic values in the paediatric population. 2–10 Although these studies provide useful normative data, they are limited by a number of problems. First, most of the reported studies, with the notable exception of Neilan et al. 's in this issue, are very small, with <200 patients in any individual study, ranging in age between birth and 18 years. As a result, many of the normal values are derived by extrapolation. Despite this, the relation between the mean value for a cardiac dimension and square root of the body surface area is remarkable constant across studies. 6 However, variation in the spread of the data in different studies can lead to differing standard deviations, hence differences in the unit ‘size’ of the z -score and consequently widely varying z -score values when utilized in disease states. 15 Furthermore, different studies use different techniques (e.g. M-mode vs. two-dimensional echocardiography) and not all studies provide normal values for all cardiac chambers, valves, and great vessels. Perhaps not surprisingly, there is considerable discrepancy between these individual studies. As a result, data from individual institutions may not be comparable. For an individual lesion, these discrepancies make the scientific evaluation of medical, interventional, and surgical management options problematic. It is beyond debate that echocardiographic parameters in the paediatric population must be normalized to body size. In order to obtain accurate and reproducible results, a robust set of normal values, derived from a large number of individuals and comprising standardized and complete measurements of all cardiac structures, using M-mode and two-dimensional echocardiography, must be obtained. The study published in this issue of the Journal 1 is a timely step in the right direction. What is needed now is for the paediatric cardiology scientific community to adopt a single study to allow z -score computation using an agreed body surface area formula (e.g. the Haycock formula 12 which may be more accurate at smaller surface areas). 6 The benefits of this approach are self-evident and would lead to better comparability of outcome studies in congenital heart disease, many of which are undermined by differences in z -score methodology. Achieving this would require a concerted and collaborative international approach involving the leading national and international congenital societies. Although not underestimating the challenge that this would present, we recommend that such discussions be initiated immediately to achieve the goal of a universal standard.