Abstract
The purpose of this study was to determine the optimal index for normalizing left ventricular (LV) echocardiographic dimensions for differences in body size. Mmode echocardiograms defined LV internal dimension at end diastole (LVIDD) and LV wall thickness (LVWT) in 107 adults (59 male, 48 female). Allometric relations were assessed between cardiac dimensions (Y) and body size variables (X) of fatfree mass (FFM), height (H), body surface area (BSA), and fat mass (FM). Further to confirmation of homogeneity of regression slopes, size exponents common to both genders were fitted by a loglinear model: lnY = lna +c ⋅ gender +b ⋅ lnX, wherea is the proportionality coefficient,b is the size exponent, andc is the gender coefficient. For LVIDD, mean body size exponents (95% confidence interval) were FFM^{0.35}(0.22–0.47),H ^{0.68}(0.32–1.03), and BSA^{0.44}(0.26–0.62). For LVWT, the derived exponents were FFM^{0.43} (0.20–0.65),H ^{0.65}(0–1.3), and BSA^{0.56}(0.23–0.89). Body fatness (expressed by FM) had no influence on LV dimensions, with exponents not different from zero (P > 0.05). The rootmeansquares error from the separate regression models indicated that the FFM index was the optimal solution. Indexation of LV dimensions byH was associated with the greatest error. Because the 95% confidence interval for the FFM exponents included 0.33, we recommend that linear LV dimensions be indexed by the cube root of FFM. In the absence of FFM data, the root of BSA was found to be the best surrogate index.
 allometric relations
 heart size
 loglinear models
despite the advent of two and even threedimensional echocardiography (32), Mmode echocardiography continues to be a useful tool in the diagnosis and management of cardiovascular disease (26). In addition, Mmode techniques have been widely applied in the exercise and sport sciences to examine such issues as gender differences in cardiac dimensions (3) and the “athletic heart syndrome” (16,40). Because of the strong relationship between heart size and body size, cardiac dimensions must be scaled for body size differences to establish reference standards for normality (35) and to permit meaningful intersubject or intergroup comparisons (3).
In the cardiology literature, a variety of different methods of normalizing heart size to account for the influence of body size have been proposed. Traditionally in clinical practice, cardiac dimensions have been divided by body surface area (BSA), although this “cardiac index” has been criticized on theoretical (19) and mathematical (44) grounds. To address these concerns, a number of crosssectional studies have modeled echocardiographic dimensions using a general allometric equation (e.g., see Refs. 710, 23, 24, 30, 45, 47). The allometric model assumes a nonlinear relationship between the body size and heart size variables of the form Y= aX^{b} ε (whereY is the cardiac dimension,X is the indicator of body size,b is the “dimensionless” size exponent, a is the proportionality coefficient, and ε is the multiplicative residual error term). Derivation of b permits the construction of a power function ratio standard (Y/X^{b} ), which is allegedly size independent.
In studies of these allometric relations, “height” has been the most popular indicator of general body size because it is simple to use and allegedly a surrogate of lean body mass (24). In modeling left ventricular (LV) mass, a variety of different height exponents have been reported, including height^{1.97} in young, “apparently healthy” men and women from the Framingham Heart Study (23), height^{2.7} in a large population study of normotensive children and adults (9), and height^{3} in children and adolescents aged 6–17 yr (7). For linear echocardiographic dimensions, Lauer et al. (24) reported mean height exponents in males and females of 0.57 and 0.50, respectively, for LV internal dimension and 0.84 and 0.41, respectively, for LV wall thickness [LVWT; sum of posterior free wall (PWT) and interventricular septum (ST) thicknesses]. It is clear that the majority of studies have focused on echocardiographically predicted LV mass, with less attention paid to LVWT and LV internal dimensions. Allometric relations between body size and linear echocardiographic dimensions, however, are of critical importance in establishing normal limits for quantification of specific disease states (35) and in comparative studies when attempting to differentiate between “concentric” (due to increased LVWT) and “eccentric” (due to increased LV internal dimension) LV hypertrophy (16).
Despite the apparent popularity of indexing cardiac dimensions according to their allometric relations with height, the most appropriate indicator of body size is yet to be defined and remains controversial. It has been suggested that height is superior to BSA normalization, because the latter masks the observed independent influence of obesity on heart size (22, 25). Others have suggested that the impact of obesity on cardiac dimensions is negligible (2, 8, 17). We believe that the controversy relates to how obesity is operationally defined. Lauer et al. (22, 24) and others (25) have represented obesity by threshold values for body mass index (BMI). It is well known, however, that BMI is influenced to an almost equal degree by the lean and fat compartments of the body (21), indicating that this ratio is unable to distinguish muscularity from adiposity (36). Because a strong link between cardiac and skeletal muscularity has been assumed (16), it is possible that the observed positive relationships between BMI and LV mass (22, 24, 25) are due to the influence of fatfree mass (FFM) rather than to obesity per se. On the basis of this critique, to assess the influence of overfatness or obesity on heart size, we suggest that it is more appropriate to model the influence of estimated fat mass (FM) on cardiac dimensions. Indeed, in contrast to the observed influence of BMI on LV hypertrophy, FM has been found to be of only minor importance in determining cardiac dimensions in children and adolescents (8) and in older adults (2).
Recently, Daniels et al. (8) and Roman (35) have suggested that FFM may indeed represent the optimal parameter for allometric normalization of cardiac dimensions. Estimated FFM has been found to be the strongest predictor of heart size in children and adolescents (7, 47), older adults (2), and younger adults (3, 11, 20, 29). Criterion methods for assessing body composition, such as dualenergy Xray absorptiometry (DEXA), clearly offer greater precision and accuracy. Indeed, DEXA may be superior to the traditional gold standard of hydrodensitometry, because it is capable of separating the FFM into bone and bonefree compartments (28). Estimation of body composition using DEXA thus circumvents two key assumptions of the twocomponent model, that the density of the FFM is known and constant and that the components of the FFM normally exist in constant proportions (21).
In clinical practice, however, concerns have been voiced about the practicality of obtaining accurate measurements of FFM using criterion techniques that may be expensive and cumbersome and require highly skilled technicians (35). Methods that are safe, rapid, and acceptably precise and accurate are required for clinical and field testing. For the busy clinician, the choice falls between anthropometry and bioelectrical impedance analysis (BIA). In the field, percent body fat can be estimated using skinfold equations with acceptable accuracy [Ref. 27; standard error of the estimate (SEE) 3–4% fat]. Estimation of body composition using BIA, although rapid and noninvasive, may have limited general applicability because of the lack of appropriately crossvalidated prediction equations (15). In a recent report (15), the prediction of body fat percentage using skinfolds (SEE 2.6% fat compared with the criterion derived from densitometry) was found to be more accurate than any of 10 published BIA equations in a population of young adult males. The authors considered that the BIA equations developed by Guo et al. (18) (SEE 2.9% fat) and Segal et al. (39) (SEE 3.3% fat) were acceptable alternatives to the skinfold method. However, both of these equations require additional anthropometry, along with resistance measures, to improve the predictive accuracy of the model. Indeed, the Segal fatnessspecific BIA equations (39) require the a priori estimation of body fat percentage from skinfolds to determine which equation to apply. Moreover, results of multiple regression analyses including anthropometric and BIA variables (15, 18) indicate that anthropometric measures account for the majority of the variance in the prediction of body fat percentage. The addition of resistance measures from BIA contributes little to the explained variance and has a marginal effect on the SEE. On the basis of these findings, we believe that, despite the inherent and welldocumented limitations (31), anthropometric estimates of FFM can be an extremely useful indicator of general body size. Recently, we have demonstrated (3) the utility of FFM, estimated by skinfolds, in allometric scaling of LV mass in adult males and females, and this is the preferred field method for the current study. In the absence of FFM estimates, an accurate body size surrogate of FFM is required to appropriately normalize cardiac dimensions. To date, insufficient attention has been paid to the determination and critical evaluation of these surrogates.
The aim of the current study was to determine the most appropriate and practical method of normalizing linear echocardiographic dimensions. We modeled the influence of general body size (FFM, BSA, and height) and “fatness” (FM) on LVWT and LV internal dimension. The best potential surrogate of FFM was predicted from a dimensional analysis of the allometric relations between the different body size indicators.
METHODS
Subjects.
One hundred and seven Caucasian adults [59 males, age 23.5 ± 3.2 (mean ± SD) yr and 48 females, age 23.7 ± 2.7 yr] volunteered for the study. Subjects were screened medically with a standard laboratory questionnaire, and all were found to be apparently healthy, asymptomatic, and free from cardiovascular disease and major risk factors for coronary heart disease. Exclusion criteria also included the chronic use of medications that may influence resting echocardiographic dimensions. Previous testing in the same laboratory had revealed no evidence of resting or exertional hypertension or electrocardiographic abnormalities. Subject characteristics are displayed in Table 1. Institutional ethics approval for the project and written informed consent from all subjects were obtained.
A simple, “global” selfreport assessment of habitual physical activity was obtained through personal interview. The instrument was modified from that used in the Allied Dunbar National Fitness Survey (1), with the frequency and intensity of ≥20min exercise or activity sessions in the previous 4 wk documented. All subjects were found to be moderately recreationally active, with 49% of males and 46% of females reporting an average of three ≥20min sessions per week at a “vigorous” exercise intensity (≥7.5 kcal/min). The remainder of the sample reported an equivalent frequency of “moderate” intensity activity (∼5 kcal/min). Male and female activity levels were not significantly different (P> 0.05). There was no relationship (P > 0.05) between physical activity status and LV dimensions. Moreover, no subjects reported physical activity levels of sufficient frequency, intensity, and duration to be associated with concentric or eccentric LV hypertrophy.
Anthropometry.
Stature (height) to the nearest 0.005 m was measured using a Harpenden stadiometer. Body mass (in light underwear) was assessed to the nearest 0.1 kg using Avery beam balance scales. Height (H) and body mass measurements were used to predict BSA according to the formula of DuBois and DuBois (13). Body density was estimated by a generalized age and genderspecific regression equation, using the log of the sum of bicep, tricep, subscapular, and suprailiac skinfolds (14). Body density was converted to percent fat using the formula of Siri (41). All skinfolds were measured, using Harpenden calipers, by an investigator who had previously demonstrated technical errors of measurement (TEM) and intraclass correlations (R) for repeated measures on each site ranging from TEM = 0.6 mm for tricep, with an R of 0.98, to TEM = 1.4 mm for suprailiac (R = 0.96) skinfolds. The median of three measurement rotations that agreed within 10% was used for subsequent analyses. Total body mass and fat percent were used to partition body mass into its FM and FFM components. BMI data revealed that 15 males (25%) and 7 females (15%) could be classified as “overweight” according to the criteria of the Royal College of Physicians report on obesity (Ref. 37; BMI 25.1–29.9 for men and 23.9–28.5 for women). One male and one female (∼2% of the sample) were classified as “obese” (BMI >30 for men and >28.6 for women).
Echocardiography.
A HewlettPackard (Andover, MA) Sonos 100 ultrasound imaging system (2.5MHz transducer) in sector twodimensional mode was used to image a longitudinal axis view of the left ventricle from the parasternal window. Mmode recordings were derived from a cursor line crossing the left ventricle just below the tips of the mitral valve leaflets. All echocardiograms were conducted and analyzed by a single experienced technician. The ST, PWT, and LVIDD measurements were made in centimeters according to the Penn convention (12). The ST and PWT measurements were then summed to form LVWT. All readings were obtained in held expiration, at the peak of a simultaneous electrocardiogram R wave, with subjects in a standardized left lateral decubitus position. Measurements represented an average of three to five heart cycles and met standard criteria of technical quality (12).
Allometric modeling.
All analyses were carried out by using the statistical package SPSS (release 6.0 for Windows; SPSS, Chicago, IL). Allometric relationships were derived from natural log transformations (base e) of the absolute data. The general curvilinear allometric equationY =aX^{b} ε can be linearized by taking natural logarithms of both sides: lnY = lna +bln X+ ln ε. The exponent b is simply the slope of the loglinear plot, and a is derived from the antilog of theYintercept. All exponents were calculated as mean point estimates, with 95% confidence intervals (CI). Statistical significance of the coefficients was tested at an αlevel of 0.05.
Allometric relations between body size indicators.
To predict the optimal surrogate of FFM, a dimensional analysis was conducted of the allometric relations between the different body size variables. Estimated FFM was modeled as the dependent variable (Y), withH and BSA included separately as independent variables (X). To check whether a single model common to both genders could be identified, homogeneity of regression slopes (43) was confirmed by including gender (G, coded “0” for males, “1” for females), and a G × lnBSA (or G × ln H) interaction term in a multiple loglinear regression model
Allometric relations between body size and cardiac dimensions.
To evaluate whether allometric normalization models can reduce the betweengender variability of cardiac dimensions, it is necessary to derive a common power function ratio standard (Y/X^{b}
). With this index, male and female cardiac dimensions can be compared in the same units of measurement. Clearly, this is possible only if homogeneity of regression slopes is confirmed, that is, there is no significant gender difference in the slopes of the loglinear relationships between body size indicators and cardiac dimensions. Therefore, allometric relations between the indicators of body size (X) and the echocardiographic dimensions (Y) were determined, as per the previous dimensional analysis
The best surrogate of the optimal body size parameter was determined by comparing the 95% agreement limits (5) between the LVIDD and LVWT values predicted by the obtained optimal model and the corresponding values predicted by the surrogate models.
RESULTS AND DISCUSSION
The physical and physiological characteristics of the subjects in the current study (Table 1) are remarkably consistent with those reported in the healthy, body sizerestricted sample of men and women from the Framingham Heart Study (24). This suggests that our sample is adequately representative of a normal adult population. The prevalence of overweight and obesity in men in the current study compares well with that reported in the Allied Dunbar National Fitness Survey (1) for the equivalent age group. For women, the data indicate that the subjects in our sample were slightly leaner than the national average. Clearly, therefore, caution must be exercised in extrapolating the findings of the current study to populations with different body size and composition characteristics. All body size and heart size variables demonstrated significant gender differences. On average, females were 0.11 m shorter and 15 kg lighter, with 18.9 kg less FFM. In absolute terms, females possessed ∼91% of the LVIDD and 79% of the LVWT of males.
KolmogorovSmirnov onesample tests revealed that the independent and dependent logtransformed variables, together with the allometric model residuals, were normally distributed (P > 0.1). In addition, no correlation was found between the absolute residuals and the predictor (independent) variables for any of the allometric models analyzed (P > 0.05), indicating that the model assumption of homoscedasticity had been satisfied.
Allometric relations between body size indicators.
For estimated FFM modeled by BSA, the following solution was obtained
This simple dimensional analysis of the allometric relations between indicators of body size suggests that BSA is superior toH as a surrogate of FFM in young, moderately active adults of averagetolean body fatness. This finding is in agreement with previous literature (42) in which stature was found to be an ineffective predictor of FFM when used alone. The BSA model is able to explain 12% more of the variance in FFM and has a much smaller rootmeansquares error. In addition, the width of the confidence interval around the mean body size exponent is much broader for H, suggesting that theH exponent is relatively less stable.
The mean exponent for BSA of 1.32 (95% CI 1.20–1.43) is representative of slight “negative allometry.” Simple dimensionality theory (38) predicts that FFM (a 3dimensional construct) would be proportional to BSA (a 2dimensional construct) to the power of 1.5, to maintain dimensional consistency or “isometry.” The upper limit of the 95% CI for the BSA exponent, however, excludes 1.5.
The mean exponent for H of 1.63 (95% CI 1.21–2.06) again indicates negative allometry. Dimensional consistency would require that FFM be proportional to the cube of height (a linear dimension), a value precluded in the current analysis. The 95% CI for the H exponent includes 2, indicating that in this sample FFM is approximately proportional to height squared. This result is remarkably consistent with previous findings in adults. In a population study of >3,000 adult males and females (age 16–64 yr), Nevill and Holder (33) reported that estimated FFM was proportional to the square of height in all groups except female subjects aged ≥55 yr (where theH exponent was <2). In contrast, Daniels et al. (7) reported that FFM, determined by DEXA, was proportional to height cubed in children and adolescents. Such differences in the reported allometric relations between FFM andH in different populations may well explain the wide range of H exponents obtained in allometric modeling of LV mass. Previously reportedH exponents for LV mass of 1.97 in adults (23) and 3.0 in children and adolescents (7) are entirely consistent with the foregoing analysis. Clearly, these findings lend indirect support to the pivotal role of FFM in determining echocardiographic dimensions.
Allometric modeling of LVIDD and LVWT.
The findings from the specific allometric models are presented in Tables 2 (LVIDD) and 3 (LVWT). All allometric models were successful in providing dimensionless size exponents, with no residual correlations evident (P > 0.05) between the power function ratioscaled cardiac dimension variable (Y/X^{b} ) and the body size variable (X). In addition, examination of the model residuals revealed no (linear or curvilinear) sizerelated distributional patterns (P > 0.05), indicating that the loglinear model is correctly specified, with the residuals randomly scattered about zero.
The derived body size exponents are consistent with the foregoing dimensional analysis. If LV mass (a 3dimensional construct) is assumed to be directly proportional to the first power of FFM, as we have previously reported (3), then it follows that linear echocardiographic dimensions should relate to the cube root of FFM (FFM^{0.33}). The 95% CI surrounding the FFM exponents for LVIDD and LVWT both include 0.33 (Tables 2 and 3). Indeed, the point estimate for the FFM exponent for LVIDD is very close to 1/3. The dimensional analysis revealed that FFM in this sample was proportional to BSA to the power of 1.32, with the upper boundary of the CI just excluding the value of 1.5 predicted from simple dimensionality theory. If linear cardiac dimensions are proportional to FFM^{0.33}, it would be predicted that LVIDD and LVWT would be related to BSA to the power of 0.44 (1.32 × 0.33). For LVIDD (Table 2), the mean BSA exponent is exactly 0.44. For the LVWT model (Table 3), the 95% CI for the BSA exponent includes this value. In practice, however, because the 95% CI for the exponent also includes the value of 0.5, it would be more convenient to scale LVIDD and LVWT by the square root of BSA, as recommended by Gutsegell and Rembold (19). Simple dimensionality theory would predict that linear cardiac dimensions should be indexed by the first power of height. However, this would only be so if FFM and LV mass were related to the cube of height. Because we have stated that in adults these variables appear to be proportional to height squared, it follows that LVIDD and LVWT should scale with height to the power of 2/3. Consistent with this prediction, Tables 2 and 3 show that the meanH exponents for both LVIDD and LVWT are very close to 2/3. Within 95% confidence limits, however, these height exponents are not different from unity (P > 0.05) or from previously reported values for the equivalent cardiac dimensions in a large population sample (24).
For both linear echocardiographic dimensions, the model including FFM as the body size indicator provides the highestR ^{2} and lowest rootmeansquares error. This finding supports the welldocumented observation that FFM is the strongest univariate predictor of heart size (e.g., see Refs. 3, 8, 29, 47). Conversely, modeling echocardiographic dimensions by height results in the lowestR ^{2} and highest rootmeansquares error of the three general body size indicators. All size exponents are statistically significant, however, and the differences in variance of cardiac dimension accounted for are relatively modest. Additional insight can be gained from the width of the CI surrounding the size exponent. The relatively wide CI surrounding the mean H exponent, especially for the LVWT model, suggests that this point estimate is less stable and potentially less generalizable to other populations than either the FFM or BSA exponents.
Modeling of cardiac dimensions by FM sheds light on the influence of body fatness on LV dimensions. Tables 2 and 3 reveal that FM was not a significant determinant of LVIDD or LVWT in this sample, withb exponents not significantly different from zero. The magnitude of effect of FM on LV dimension can be quantified by an analysis of the mean exponents. The βcoefficient for FM signifies the predicted increase in the specific LV dimension, associated with a 1unit increase in log e FM. Because the backtransformation of a logtransformed variable provides a ratio (4), a 1unit increase in log e FM represents a 2.718fold increment (antilog of 1 = 2.718, base e). For LVIDD, the mean βcoefficient of 0.02 indicates that a 2.718fold increase in FM is associated with only a 2% increase in the cardiac dimension (antilog of 0.02 = 1.02). For LVWT, the equivalent increase in FM is associated with only a 6% increment in the dependent variable (antilog 0.06 = 1.06). In contrast, the corresponding predicted increments in LV dimensions associated with a 2.718fold increase in FFM are 42% for LVIDD (antilog 0.35 = 1.42) and 54% for LVWT (antilog 0.43 = 1.54). Modeling of echocardiographic dimensions by gender and FM was associated with the lowestR ^{2} and highest rootmeansquares error overall. It would appear that in this sample of young, healthy adults, body fat has a negligible impact on LV hypertrophy, as has previously been reported in children and adolescents (8) and older adults (2). We urge caution in extrapolating this finding to larger and diverse population samples, however, in which a greater degree and a higher prevalence of overweight or obesity may influence the results.
As detailed in methods, the relative reduction in betweengender variability due to different scaling models can be evaluated by calculating the separate proportionality coefficients (a) in the allometric equations for males and females, assuming a common size exponent (b). For comparison, findings are presented for FFM and H (the body size indicators associated with the lowest and highest rootmeansquares error, respectively) in modeling LVIDD and LVWT. For modeling of LVIDD by FFM, in males LVIDD (cm) = 1.25 ⋅ FFM (kg)^{0.35} and in females LVIDD (cm) = 1.28 ⋅ FFM (kg)^{0.35} (female/male percentage, 102%, P = 0.34). For modeling of LVIDD by H, in males LVIDD (cm) = 3.60 ⋅ H(m)^{0.67} and in females LVIDD (cm) = 3.41 ⋅ H(m)^{0.67} (female/male percentage 95%, P = 0.002). For modeling of LVWT by FFM, in males LVWT (cm) = 0.33 ⋅ FFM (kg)^{0.43} and in females LVWT (cm) = 0.30 ⋅ FFM (kg)^{0.43} (female/male percentage 91%, P = 0.03). For modeling of LVWT by H, in males LVWT (cm) = 1.33 ⋅ H(m)^{0.65} and in females LVWT (cm) = 1.08 ⋅ H(m)^{0.65} (female/male percentage 81%, P = 0.0000).
The findings confirm that indexation of cardiac dimensions by FFM, raised to its allometric power, is most effective in reducing betweengender variability. For LVIDD, significant absolute gender differences (females 91% of male value) disappear when the data are appropriately normalized for differences in FFM. For LVWT, although the differences in FFMadjusted values remain significant, betweengender variability is considerably reduced. Females possess 91% of the male FFMadjusted value compared with 79% in absolute terms. These results are concordant with previous findings reported in the literature. Devereux et al. (11) reported that scaling LV mass by lean body mass, estimated by 24h urinary creatinine excretion, eliminated gender differences in absolute LV mass. Furthermore, we have previously demonstrated the ability of an allometric model using FFM, estimated by skinfolds, to reduce betweengender variability in LV mass (3). In contrast, the present results for heightadjusted cardiac dimensions illustrate that betweengender variability remains almost as high as the unadjusted absolute values, especially for LVWT.
FFM surrogates: BlandAltman agreement limits.
The preliminary dimensional analysis suggested that BSA would be the best surrogate of FFM for modeling echocardiographic dimensions. The LVIDD and LVWT predicted from the allometric models including gender and BSA, and gender and H as predictor variables, were compared with the predicted values from the criterion (optimal) model (including gender and FFM). BlandAltman plots (5) revealed that for LVIDD, BSA was a more effective surrogate, with a 95% agreement limit of 98–102% of the FFMpredicted value. The corresponding agreement limits forH were 94–106%. For LVWT, a similar picture emerged. Agreement limits for BSA were 96–104% compared with 94–106% forH.
In the current study, the high values for goodness of fit, the low rootmeansquares error, the relatively narrow CIs and stable point estimates, and the demonstrably superior ability to reduce betweengender variability suggest strongly that estimated FFM is the optimal body size parameter for modeling echocardiographic dimensions. Therefore, despite the welldocumented limitations (31), we believe that a simple, anthropometric estimate of FFM may be more appropriate than height in accounting for the influence of body size on heart size. Future research, on larger samples and diverse populations, is obviously required to substantiate this assertion. Anthropometric variables can be measured rapidly, with acceptable precision, with relatively little specialized training (34), making estimation of FFM viable in clinical practice. If FFM estimates are unavailable or impractical, however, allometric indexation by BSA appears to be the most effective surrogate in young adults of averagetolean body composition. Scaling by various powers of height has a negligible impact on betweengender variability in cardiac dimensions and demonstrates relatively weaker agreement with criterion modeling by FFM.
Because of the assumed close links between skeletal and cardiac muscularity, with testosterone as the primary signal messenger (16), it is likely that measures of “skeletal muscle mass” (MM) would represent an improvement on FFM for modeling echocardiographic dimensions. Indeed, analyses from the Brussels Cadaver Study (6) have revealed that the proportion of adipose tissuefree mass that is composed of MM and bone mass demonstrates considerable intersubject variability. In 25 cadavers, bone mass ranged from 16.3 to 25.7%, and MM from 41.9 to 59.4% of adipose tissuefree mass. If this variability were present in the current sample, an important assumption of the existing allometric model, that MM represents a constant fraction of FFM, would be violated. Hence, in the current study, even indexation of cardiac dimensions by various powers of FFM may be prone to considerable error because of violation of the assumptions of the twocomponent (FM and FFM) body composition model. Unfortunately, although methods such as DEXA and computerized axial tomography show great promise, criterion methods for the in vivo measurement of whole body MM have not been adequately developed and validated (46). At present, it would seem that measures of FFM, preferably by DEXA or alternatively by anthropometric techniques if a simple estimation is required in clinical practice, are the best compromise available.
In conclusion, based on the current findings, together with those of our previous study (3), we recommend indexation of LV mass by FFM to the first power (LVM/FFM) and of linear Mmode echocardiographic dimensions by FFM^{0.33}. Alternatively, for indexation by BSA, the power function ratio standards LV mass/BSA^{1.5} and LVIDD/BSA^{0.5} (or LVWT/BSA^{0.5}) seem most applicable, as originally proposed by Gutsegell and Rembold (19). Further research is required to test the applicability of these findings to diverse population samples, with different age, racial/ethnic origin, and body size and composition characteristics.
Footnotes

Address for reprint requests: A. Batterham, School of Social Sciences, University of Teesside, Middlesbrough TS1 3BA, UK. Email:A.Batterham{at}tees.ac.uk.
 Copyright © 1998 the American Physiological Society