Jump to
Abstract
The need for left ventricular mass (LVM) normalization to body size is well recognized. Currently used allometric exponents to normalize LVM may not account for the confounding effect of sex. Because sex is a strong determinant of body size and LVM, we hypothesized that these are subject to potential bias. We analyzed data from 7528 subjects enrolled in the Asklepios Study (n=2524) and the Multiethnic Study of Atherosclerosis (limited access data set; n=5,004) to assess metric relationships between LVM and body size, generate normative data for indexed LVM, and compare the ability of normalization methods to predict cardiovascular events. The allometric exponent that adequately described the LVM-body height relationship was 1.7 in both studies and significantly different from both the unity and 2.7, whereas the LVM-body surface area relationship was approximately linear. LVM/height2.7 consistently demonstrated important residual relationships with body height and systematically misclassified subjects regarding the presence of LVH. LVH defined by LVM/height1.7 was more sensitive than LVM/body surface area to identify obesity-related LVH and was most consistently associated with cardiovascular events and all-cause death. In contrast to current assumptions, LVM/height2.7 is not an adequate method to normalize LVM for body size. We provide more appropriate normalization methods, normative data by 2D echocardiography and gradient-echo cardiac MRI, and cutoffs for defining LVH, along with prognostic validation data.
Left ventricular mass (LVM) is a powerful predictor of cardiovascular risk.1–4⇓⇓⇓ LVM strongly relates to body size, indicating the need for appropriate normalization.5,6⇓ Because relations between body size and dimensions of organs are often nonlinear, allometric approaches are required,5,6⇓ in which LVM is divided by a body size variable raised to a scalar exponent intended to describe the unique relationship between that variable and LVM.5,6⇓
Sex is a strong correlate of body size and LVM. Therefore, appropriate assessments of allometric LVM relationships must account for the effect of sex to test for the presence of common exponents for men and women and to satisfy the group difference principle.7 According to the latter, in the presence of body size–independent sex differences in LVM, larger body sizes among men result in biased estimated exponents influenced by both sex and body size (therefore, not representing true allometric relationships), unless adjustment for sex is performed in statistical models to obtain such exponents. However, the most widely used allometric powers in adults were originally obtained without accounting for the confounding effect of sex.6 Therefore, contrary to common assumptions, it is unclear whether these powers appropriately normalize LVM for body size.
In this study, we aimed to do the following: (1) assess the allometric powers that account for the normal relationship between LVM and commonly used measures of body size and examine the confounding effect of sex in large population-based adult samples; (2) establish normative values of normalized LVM in adults measured by 2D echocardiography and cardiac MRI; (3) provide sex-/ethnic-specific normative data for normalized LVM; and (4) quantify and compare the prognostic value of LVM indexed by different methods.
Methods
Study Population
We analyzed data from the Asklepios Study and the Multi-Ethnic Study of Atherosclerosis (MESA), both of which recruited adults free of overt cardiovascular disease.8,9⇓ The Asklepios Study recruited a cohort of 2524 community-dwelling adults aged 35 to 55 years from 2 Belgian communities.9 MESA enrolled 6814 white, black, Hispanic, and Chinese adults aged 45 to 84 years from 6 US communities.8 For these analyses, we used data from the MESA limited-access data set, provided by the National Heart, Lung, and Blood Institute. Both studies were approved by the ethical committees/institutional review boards of participating centers. Subjects provided informed consent.
To provide reference standards for normal body size-LVM relationships, we selected reference subsamples of normal-weight adults (body mass index: 18 to 25 kg/m2) free of overt cardiovascular disease at baseline who did not meet any of the following criteria: (1) hypertension (systolic blood pressure ≥140 mm Hg, diastolic blood pressure ≥90 mm Hg, or antihypertensive drug treatment); (2) current smoking; (3) diabetes mellitus (fasting blood glucose ≥126 mg/dL or antidiabetic medication use); (4) low-density lipoprotein cholesterol >160 mg/dL; (5) triglycerides ≥150 mg/dL; (6) high-density lipoprotein cholesterol <40 mg/dL (men) or <50 mg/dL (women); and (7) use of lipid-lowering medication(s). Reference samples consisted of 523 MESA participants (265 white, 133 Chinese, 65 black, and 60 Hispanic) and 637 white Asklepios Study participants. The age range among reference samples from both studies was identical to the range of the overall study populations. After defining allometric powers, data from all of the subjects with available covariates were used to assess the effect of obesity on LVM within each study (Asklepios Study: n=2524; MESA: n=5004). Among MESA participants, we prospectively assessed the effect of LVM on the risk of cardiovascular events (CVEs) and all-cause death during follow-up.
LVM Measurements
In the Asklepios Study, Doppler-echocardiography was performed (Vivid-7 platform, Vingmed),9 and LVM was estimated from 2D left ventricular dimensions using the American Society of Echocardiography formula.10 In MESA, LVM was measured from end-diastolic short-axis gradient-echo cardiac-MRIs as follows: (epicardial-endocardial contour)×(slice thickness+image gap)×1.05 g/mL,11 excluding papillary muscles from LVM calculations.
Statistical Analyses
We estimated that 123 subjects were needed to achieve 80% power to demonstrate a sex-independent partial correlation coefficient ≥0.25 between body size parameters and LVM, considered quantitatively significant enough to warrant allometric normalization. The following equation was used: y=axbgcε, where “x” is a measure of body size; “g” is sex; “a,” “b,” and “c” are parameters; and “ε” is a random error term. Models without sex were also constructed to assess differences in the results. Measures of body size included height, weight, and body surface area (BSA) estimated with the Gehan method (BSA=0.0235×height in centimeters0.42246×weight in kilograms0.51456).12 We used log-linear regression to assess for interactions between sex and measures of body size.7 Definitive estimations of allometric powers were performed using nonlinear regression, although results from both methods were very similar. Because of the sample size of the black and Hispanic MESA reference subsamples, only the white and Chinese subsamples from MESA and all of the white reference participants in the Asklepios Study were included in analyses aimed at obtaining allometric powers. All of the reference subsamples were used to define normative data for indexed LVM. Ninety-fifth percentiles of LVM in reference subsamples were considered as upper-normal limits for various sex/ethnic strata. The effect of overweight/obesity on LVH prevalence was assessed using logistic regression. Cox regression was used to assess the performance of indexed LVM as predictor of hard CVEs, all CVEs, and all-cause death. Goodness-of-fit of different models were compared with the Akaike information criterion, with smaller values indicating better fit. Agreement for classification of presence/absence of LVH by different methods was assessed with the κ statistic. Analyses were performed using SPSS for Windows version 17 (SPSS Inc). A detailed description of our statistical methods is available in the online Data Supplement (please see http://hyper.ahajournals.org).
Results
Baseline characteristics of subjects included in both studies are shown in Table S1 in the online Data Supplement (available at http://hyper.ahajournals.org).
Physiological Relation of LVM to Body Height
Table 1 shows the relationship of LVM to body size in reference samples. Figure 1 shows 95% CIs for allometric powers describing such relationships. If CIs for these powers do not cross the unity, a linear relationship is rejected and a nonlinear relationship is demonstrated. We found no significant height-sex interactions as predictors of LVM, allowing the use of common exponents for both sexes. Exponents that described the LVM-height relationship were 1.67, 1.63, and 1.68 in Asklepios and MESA white and Chinese reference participants, respectively. In all 3 of the samples, exponents were significantly different from the unity and the currently recommended value of 2.7 (Table 1 and Figure 1).
Table 1. Relations of LVM to Body Height, Body Weight, and BSA in Reference Participants Before and After Allometric and Ratiometric Indexation
Figure 1. Point estimates and 95% CIs for allometric powers describing the relationship between LVM and body height (top), weight (middle), or BSA (bottom). The dashed line indicates the first power.
Residual relationships between indexed LVM and body height (Table 1) were absent when appropriate powers were used (R≤0.01; P>0.05), but ratiometric indexation (LVM/height) resulted in significant residual correlations with body height (R=0.11 to 0.18; P<0.05). Normalization to height2.7 resulted in even stronger residual relationships with body height (R=−0.20 to −0.29; P<0.001).
Physiological Relation of LVM to Body Weight and BSA
In Asklepios reference participants, the exponent describing the weight-LVM relationship was 0.91 (Figure 1 /Table 1). A similar exponent was found among men in MESA, whereas exponents of 0.51 and 0.74 were found among white and Chinese female MESA participants, respectively. Allometric (unlike ratiometric) normalization of LVM for body weight consistently eliminated sex-independent relations of indexed LVM to body weight (R≤0.02; P>0.05).
In Asklepios reference participants, LVM related to BSA in a minimally nonlinear fashion (power: 1.40). BSA linearly related to LVM among MESA participants. Normalization of LVM for BSA using appropriate powers eliminated the sex-independent relations of indexed LVM to BSA (R≤0.01; P>0.05). However, ratiometric indexation was also effective in eliminating the residual LVM-BSA relationship in both studies (Table 1).
Sex, Ethnicity, and Normative LVM Values
Distributions of LVM indexed by different methods in Asklepios and MESA reference participants are summarized in Table 2. Sex demonstrated a significant relationship with LVM after adjustments for body size. In MESA, we found no significant differences in indexed LVM between white and black men or women; therefore, normative distributions are combined for these ethnic groups. LVM/BSA was significantly lower (P<0.001) among Chinese men compared with white (P=0.017), black (P<0.001), and Hispanic (P=0.04) men. LVM/height1.7 was significantly lower (P<0.001) among Chinese men compared with white (P=0.005), black (P<0.001), and Hispanic (P=0.008) men. Among women, LVM/height1.7 differed significantly among ethnic groups (P=0.002) because of significantly lower LVM/height1.7 among Chinese compared with Hispanic women (P=0.001). Finally, LVM/height1.7 was significantly higher in Hispanic compared with white women (P=0.01).
Table 2. Normalized LVM Percentile Values in Asklepios and MESA Reference Samples
Indexed LVM in Lean, Overweight, and Obese Subjects
An important increase in LVM/body height1.7 and the prevalence of LVH defined by LVM/height1.7 were observed from normal weight to obesity. In contrast, overweight and obesity were not associated with increased LVM/BSA in either study (Table S2).
Results Obtained When Confounding by Sex Was Not Considered
In models that did not account for sex, LVM related to body height in an approximately cubic fashion. Exponents for body height from Asklepios and white and Chinese MESA reference participants were 3.19 (95% CI: 2.90 to 3.47), 3.32 (95% CI: 2.92 to 3.72), and 2.83 (95% CI: 2.34 to 3.32), respectively. Exponents for body weight obtained from Asklepios and white and Chinese MESA reference participants were 1.41 (95% CI: 1.30 to 1.53), 1.30 (95% CI: 1.14 to 1.44), and 1.15 (95% CI: 0.96 to 1.35), respectively. Respective exponents for BSA were 2.14 (95% CI: 1.97 to 2.31), 2.00 (95% CI: 1.77 to 2.21), and 1.81 (95% CI: 1.52 to 2.10).
Figure 2 shows an example of the profound bias derived from estimating allometric exponents while neglecting the sex effect, demonstrating how the exponent describing the LVM-body height relationship was markedly overestimated in the Asklepios reference sample. Continuous gray and black lines show the mildly nonlinear LVM-body height relationship among men and women, respectively. When men and women were analyzed together without accounting for the significantly higher LVM observed in men for any given body height, the nonlinearity was exaggerated to an approximately cubic relationship (black dashed line).
Figure 2. Body height-LVM relationship in Asklepios reference participants assessed with nonlinear regression with and without accounting for the confounding effect of sex. The continuous gray line represents the body height-LVM relationship in men. The continuous black line represents the body height-LVM relationship in women. The dashed black line represents the exaggeration of nonlinearity in the height-LVM relationship when the confounding effect of sex is neglected.
LVM/height2.7 failed to demonstrate differences between Chinese and white women (28.7 versus 28.9 g/m2.7; P=0.99) or between Chinese and white men (32 versus 33.6 g/m2.7; P=0.66), despite the fact that, in nonlinear models, Chinese ethnicity was an independent predictor of lower LVM after adjustment for height and sex (P=0.002).
LVM/height2.7 demonstrated an artificial negative relationship with body height not only in the reference sample but also in the entire study population (R=−0.25; P<0.0001). Importantly, in MESA, shorter body height independently predicted hard CVE in fully adjusted models (data not shown). It follows that LVM/height2.7 artificially captures prognostic information associated with body height itself.
Finally, there was substantial disagreement between LVM/height1.7 and LVM/height2.7 in identifying individuals with LVH in MESA (κ=0.79). In analyses stratified by height, it was clear that LVM/height2.7 grossly overestimated the prevalence of LVH in men and women with shorter body height while underestimating its prevalence in those with greater body height (Figure 3). Similar results were obtained in the Asklepios Study (data not shown).
Figure 3. Prevalence of LVH defined by LVM/height1.7 and LVM/height2.7 among male (A) and female (B) MESA participants stratified by body height.
LVM Index as a Predictor of CVE
Table 3 shows hazard ratios (HRs) associated with a 1-SD increase in LVM/height1.7, LVM/height2.7, or LVM/BSA as predictors of CVE and death. During a median follow-up of 4.8 years, 214 subjects had a CVE, 134 had a hard CVE, and 151 subjects died. In fully adjusted models, all of the forms of indexation performed similarly when treated as continuous variables. However, only LVH defined by LVM/height1.7 demonstrated a consistent relationship with all CVE, hard CVE, and all-cause death. LVH defined by LVM/height2.7 or LVM/BSA failed to predict all-cause death. Adjustment for body mass index resulted in negligible changes in the hazard ratios associated with LVM or the goodness-of-fit of models containing LVM index and other predictors.
Table 3. Cox Models Assessing the Performance of LVM Indexed for Body Height and BSA Predicting CVEs and All-Cause Death in MESA (n=4890)
Discussion
We report on the normal sex-independent relationship of LVM determined by 2D echocardiography and cardiac MRI to body size and on the ability of indexed LVM to detect obesity-associated LVH and predict CVE and death among middle-aged/older adults. We determined allometric exponents for LVM with and without adjustment for sex and found that the larger body size among men is an important confounder of the body size-LVM relationship, because sex is an independent determinant of LVM. The exponent that described the body height-LVM relationship was 1.7 (rather than 2.7, as described previously based on analyses that did not account for sex).6 LVM/height2.7 demonstrated important residual relationships with body height (which were even stronger than those seen with ratiometric normalization) and systematically misclassified individuals with long or short body heights, demonstrating that this is not an adequate indexation method. We present normative data for 2D echocardiography and cardiac–MRI-derived LVM. We demonstrate significant ethnic differences in indexed LVM. We show that LVM/height1.7 was most consistently associated with different outcomes, including all-cause death.
Given the impact of sex on LVM and body size, normalization must exclude this confounding effect. This concept, although not obvious for allometric models, is essentially no different than the concept applied for other inference procedures in which sex affects the variables at hand. Not only is the presence of a common exponent applicable to both sexes an important assumption that must be ascertained, but allometric normalization itself must not be influenced by the body size–independent sex effect. Only under these circumstances can an exponent be identified that is not affected by sex itself, ensuring that estimates of associations between LVH and CVE are not affected by male sex, which is associated with higher cardiovascular risk. In addition, we show that, when the sex effect is ignored, an important residual relationship between indexed LVM and body height occurs, which biases the relationship between indexed LVM and cardiovascular risk, because body height itself is associated with CVEs.13 Therefore, LVM/height2.7 artificially captures prognostic information associated with body height. Despite this artificial advantage, LVM/height2.7 was not a superior predictor of a CVE as a continuous variable, and LVH defined by LVM/height2.7 was less consistently associated with CVE and, in contrast to LVH defined by LVM/height1.7, failed to predict all-cause death. Furthermore, the prognostic disadvantage of LVM/height2.7 is likely to be even greater than is apparent from our relatively short-term follow-up, during which CVEs related to LVH are expected to occur predominantly among subjects with the most pronounced increases in LVM, likely to be classified as LVH regardless of the indexation method.
Previous data in which the sex effect was not considered suggested that the LVM-body weight relationship in adults is approximately linear and that the BSA-LVM relationship is described by a power of ≈1.5, whereas the body height-LVM relationship is approximately cubic.6 This was explained on the basis of assumed relationships among LVM (3D measure), body weight (related to body mass, a 3D measure), BSA (2D measure), and body height (1D measure). Such relationships were said to be expected based on the ratio of the measured dimensions. Therefore, LVM would be expected to relate linearly to the first power of weight (dimension ratio: 3/3), the 1.5th power of BSA (dimension ratio: 3/2), and the third power of height (dimension ratio: 3/1). However, such expectations are only valid if the relationship between body height and other body dimensions is isometric and the relative amount of body tissues with different densities is constant across the entire body height range. Clearly, such assumptions are untenable, as demonstrated by the body weight-height relationship, which is not cubic as expected by their dimension ratio (3/1) but approximately quadratic. The wide use of body mass index (weight/height2) to normalize body weight for height assumes such a relationship. The noncubic weight-height relationship is, therefore, inconsistent with previously reported allometric powers for LVM (particularly height2.7), which were confounded by the body size–independent sex effect on LVM. The widely assumed approximately cubic relationship between body height and LVM is also highly inconsistent with the reports by Lauer et al14 and Brumback et al,15 who found this relationship to be approximately quadratic, in agreement with our results. Our results are also highly consistent with theoretical physiological considerations related to mechanical myocardial load and circulatory dynamics, as discussed in the online Data Supplement.
The landmark contributions by de Simone et al6,16,17⇓⇓ represented important steps forward in understanding the importance of nonlinear LVM-body size relationships and the need for allometric LVM normalization. However, there is a need to improve normalization methods to avoid the bias introduced by sex, which results in an overestimation of allometric exponents (Figure 2), residual relationships with body size (Table 1), and important systematic misclassification of individuals (Figure 3).
We used data from 2 independent studies using different methods to measure LVM. Because different imaging methods provide different absolute LVM estimates, we did not attempt to combine data from these studies. Instead, we performed independent analyses in both data sets to test the consistency of our findings and to provide normative data for LVM assessed by both methods. Interestingly, equations describing the relationship between echocardiography-derived LVM and body height in the Asklepios Study were essentially identical to those obtained from white MESA participants, resulting in similar cutoff values to define LVH. However, LVM estimates are sensitive to the imaging method. In this regard, it is interesting to note that a slightly nonlinear exponent was obtained from Asklepios Study participants for BSA, whereas a linear relationship was observed in MESA. This was likely because of geometric LV variations with increasing body size. Specifically, the American Society of Echocardiography formula assumes a constant LV shape, whereas cardiac-MRI–derived LVM does not depend on geometric assumptions. Therefore, geometric changes associated with increasing body size (eg, those in LV sphericity) that would have been undetected by 2D echocardiography in the Asklepios Study likely resulted in a mild exaggeration of the BSA-LVM relationship (slightly higher exponent).
In addition to the methods presented and validated herein, Bluemke et al4 proposed an approach in which MRI-derived LVM was expressed as a ratio of expected:observed LVM, where expected LVM was computed from an individual’s body height and weight using sex-specific nonlinear equations. This height-weight method has been shown to be superior to nonindexed LVM to predict a composite end point of nonfatal and fatal coronary heart disease and stroke.15 Although these equations optimally fit the body size-LVM relationship among normal individuals, we found that the model fit between BSA and LVM was very similar to the fit of a model containing separate terms for height and weight in both studies (data not shown). Indexations for body height and BSA represent simple methods of normalization, particularly for clinical use, and are complementary to the combined weight-height method proposed by Bluemke et al.4 Of note, Bluemke et al4 reported that normalization for body height to the power of 1.9 did not fully eliminate the LVM-body height relationship, which is not inconsistent with our findings, because, as shown by our results, Chinese individuals have lower values of LVM indexed for height. Because Chinese individuals also tend to demonstrate shorter body heights compared with other ethnic groups, the LVM-height relationship in the overall sample is expected to be influenced by the effect of Chinese ethnicity. Our findings unequivocally demonstrate that normalization for body height1.7 fully eliminates the ethnic-independent, sex-independent relationship between LVM and body height in our reference samples (Table 1).
Our cutoffs to establish abnormal values to define LVH were derived from data distribution in reference samples of normotensive, normal-weight individuals without conventional risk factors and demonstrate the prognostic validity of this approach. However, factors other than blood pressure, obesity, and conventional risk factors may impact LVM. Furthermore, dichotomization was performed for operative definitions using an arbitrary percentile in reference samples. This approach may not establish optimal cut points to define LVH. In addition, the relationship between LVH and cardiovascular risk may be continuous rather than discrete. Clearly, further studies are required to assess optimal cutoffs to define LVH based on the magnitude of risk increase with increasing degrees of LVH. However, this will require larger sample sizes and/or number of events.
Our study has limitations. Given our strict exclusion criteria, we did not study an adequate number of healthy blacks and Hispanics. Consequently, normalization using powers derived from white and Chinese populations may not be optimal for these ethnic groups. Therefore, data from these populations should be considered preliminary. Inclusion of these ethnic groups in analyses evaluating the association between LVH and CVE may have influenced our results. Longer time to follow-up resulting in a greater number of CVEs will provide adequate power to assess ethnic-specific associations between LVH and CVEs. Because age influences LVM, data presented herein should be strictly applied according to the age distribution of the described populations. The Asklepios Study population was restricted to middle-aged adults (35 to 55 years). We did not determine methods for LVM normalization to the mass of tissues with different metabolic rates (lean body mass and fat mass). Previous studies demonstrated that LVM is more strongly related to fat-free compared with adipose mass18 and that indexation for fat-free mass eliminates sex-related differences in LVM.19 Although future research is needed to identify the prognostic implications of indexation to fat-free mass, we provide information for appropriate indexation to the most practical and commonly used body size indices in clinical and epidemiological settings.
Perspectives
We present normalization methods that account for the sex-independent LVM-body size relationship. These powers are equally valid both sexes, for sex-independent estimations of LVH correlates, and associations between LVH and CVEs. The ability to properly account for body size will allow for more accurate quantification of LVM in clinical and research settings.
Acknowledgments
Sources of Funding
The Asklepios Study was funded by Fonds voor Wetenschappelijk Onderzoek Vlaanderen grant G.0.838.10. MESA is conducted/supported by the National Heart, Lung, and Blood Institute in collaboration with MESA investigators.
Disclosures
None.
Footnotes
-
This article was prepared using a limited-access data set obtained from the National Heart, Lung, and Blood Institute and does not necessarily reflect the opinions or views of all of the MESA investigators or the National Heart, Lung, and Blood Institute.
- Received January 12, 2010.
- Revision received February 1, 2010.
- Accepted April 15, 2010.
References
This Issue
Jump to
Article Tools
Share this Article
- Left Ventricular Mass
