Predicted lean body mass, fat mass, and all cause and cause specific mortality in men: prospective US cohort study
BMJ 2018; 362 doi: https://doi.org/10.1136/bmj.k2575 (Published 03 July 2018) Cite this as: BMJ 2018;362:k2575All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
I calculated 459 confidence intervals (CI), 3 exposures, 5 primary outcomes and at least 8 covariates in study by Lee and coworkers (1). The authors used 0.05 threshold for statistical significance which is counterintuitive. Interestingly, a highest number of reported associations in a single observational study was 264 based on a survey published in 2004 (2).
Furthermore, Lee and coworkers described possibility of confounding and instantly explained it away in the Discussion: "...However, the homogeneity of the study population and comprehensive data on the risk factors minimized potential confounding..." (1). A recent survey of observational studies showed that only few (2/120) studies warranted cautious interpretation due to confounding (3).
E-value has been proposed for sensitivity analyses due to unmeasured confounding in observational studies (4,5). The definition of E-value is: "The E-value is the minimum strength of association, on the risk ratio scale, that an unmeasured confounder would need to have with both the treatment and outcome, conditional on the measured covariates, to explain away a treatment–outcome association." (4).
For example, Lee et al reported hazard ratio of 1.35 (95% CI 1.26 to 1.46) for all-cause mortality when subjects with highest fifth of predicted fat mass were compared to subjects with lowest fifth (1). E-values based on the aforementioned hazard ratio and lower limit of CI are 1.76 and 1.63, respectively, if one ignores other (potential) biases and multiplicity of analyses.
References
1. Lee DH, Keum N, Hu FB, et al. Predicted lean body mass, fat mass, and all cause and cause specific mortality in men: prospective US cohort study. BMJ 2018;362:k2575.
2. Pocock SJ, Collier TJ, Dandreo KJ, et al. Issues in the reporting of epidemiological studies: a survey of recent practice. BMJ 2004;329:883.
3. Hemkens LG, Ewald H, Naudet F, et al. Interpretation of epidemiologic studies very often lacked adequate consideration of confounding. J Clin Epidemiol 2018;93:94-102.
4. VanderWeele TJ, Ding P. Sensitivity Analysis in Observational Research: Introducing the E-Value. Ann Intern Med 2017;167:268-274.
5. Mathur MB, Ding P, Riddell CA, VanderWeele TJ. Website and R Package for Computing E-Values. Epidemiology 2018 Jun 14. doi: 10.1097/EDE.0000000000000864.
Competing interests: I like statistics.
Dr Dong Hoon Lee and colleagues have shown elegantly that reduced lean body mass is the principle reason for any increase in mortality in thin people as assessed BMI. The results might have been even more striking had they used the natural rather than conventional measures of body mass and its components.
The natural scaling for mass is by l3 for objects of the same shape. BMI is a special case. It is a misnomer, not being strictly mathematically an index, as it is proportional to height. As excess fat is the risk factor it works well over a limited range of height because it effectively assumes that the body is a cylinder of fixed height, a particularly appropriate assumption for abdominal obesity, and possibly in addition at the lower end of the range because, being homoeothermic, smaller individuals benefit from greater body mass.
Scaling by l2 is clearly inappropriate for other measurements. It would be of great interest to look at the relative effects of ponderal index, fat index and lean body mass index all scaled by the natural l3 on mortality. The results might not be different but they would truly reflect the effects of the different components over the full range of height of the participants.
Competing interests: No competing interests
Lee et al (1) estimated the association of fat and lean mass with mortality using data from the Health Professionals Follow-up Study (HPFS). They had no data on fat and lean mass, so they imputed values for those quantities for the participants in HPFS. For the imputation, they used equations that they had previously developed (2) with National Health and Nutrition Examination Survey (NHANES) data. Their equations had been developed using measured weight, measured height and waist circumference, measured just above the iliac crest, to predict dual-energy X-ray absorptiometry (DXA)-derived fat and lean mass in NHANES.
None of the anthropometric variables used in the prediction equations were available in the HPFS. The HPFS contains only data on self-reported weight, self-reported height, and waist circumference measured at the umbilicus, a different anatomical site. Since the authors do not have the correct data to use in the prediction equations, the equations can’t be applied correctly. Calling the equations “validated” is misleading because they were not validated for the anthropometric variables available in the HPFS.
It is well-documented that self-reported weight and self-reported height have systematic errors related both to their true values and to other characteristics of the individual respondent (3). Misreporting is affected by sociodemographic factors, age, sex, ethnicity, and educational levels, by true measured levels, by mode of administration and presumably by other individual factors such as social desirability effects. These factors do not affect errors in measured data, which has minimal errors in any case. Thus the use of self-reported data in the HPFS is adding additional residual confounding by uncertain factors that may also affect mortality. The high correlations and similar mean values between self-reported and measured weight and height data, sometimes cited erroneously as evidence that self-reported data are reasonably accurate (4), mask the distortions introduced by self-reported weight and height data, which include high levels of misclassification and a narrowing of the whole distribution (5). Neither of these can be detected by only considering correlation coefficients and mean values. It is also well documented that there are differences in waist circumferences measured at different anatomical sites (6).
When data are imputed, the imputation procedure adds variability to the data. The authors are using the prediction equations to impute values of lean and fat mass for each individual participant and then using those values in further analyses. Analyses of such data need to use appropriate statistical methods that reflect the standard errors of the model coefficients. The additional variance introduced should be taken into account in the standard errors of the hazard ratios because predicted values should not be treated as exact values with no error.
It would be unfortunate if this article leads other researchers to believe that they can easily and accurately examine the relation of body composition to mortality or other outcomes even though they have no data on body composition. Prediction equations used to supply missing values need to be used with the correct variables. Statistical analysis of imputed data needs to take into account the variance that the imputation adds. When, as in the Lee et al article (1), all the values of a variable are missing and have to be imputed, it may be preferable to simply include the predictive variables directly rather than carrying out an imputation. Previous analyses of adiposity and mortality in the HPFS were able to examine the relation of weight, height and waist circumference to mortality (7) without resorting to fixed combinations of those variables to create new imputed variables.
1. Lee DH, Keum N, Hu FB, Orav EJ, Rimm EB, Willett WC, Giovannucci EL. Predicted lean body mass, fat mass, and all cause and cause specific mortality in men: prospective US cohort study. BMJ 2018;2362:k2575 p.
2. Lee DH, Keum N, Hu FB, Orav EJ, Rimm EB, Sun Q, Willett WC, et al. Development and validation of anthropometric prediction equations for lean body mass, fat mass and percent fat in adults using the National Health and Nutrition Examination Survey (NHANES) 1999-2006. Br J Nutr. 2017;118(10):858-866.
3. Cawley J, Maclean JC, Hammer M, Wintfeld N. Reporting error in weight and its implications for bias in economic models. Econ Hum Biol. 2015;19:27-44.
4. Di Angelantonio E, Bhupathiraju SN, Hu FB, Danesh J, Peto R, Lewington S. Body-mass index and all-cause mortality - Authors' reply. Lancet. 2017;389(10086):2285-2286.
5. Flegal KM, Kit BK, Graubard B. Bias in hazard ratios arising from misclassification by self-reported weight and height in observational studies of body mass index and mortality Am J Epidemiol. 2017.
6. Mason C, Katzmarzyk PT. Variability in waist circumference measurements according to anatomic measurement site. Obesity (Silver Spring). 2009;17(9):1789-1795.
7. Baik I, Ascherio A, Rimm EB, Giovannucci E, Spiegelman D, Stampfer MJ, Willett WC. Adiposity and mortality in men. Am J Epidemiol. 2000;152(3):264-271.
Competing interests: No competing interests
Re: Predicted lean body mass, fat mass, and all cause and cause specific mortality in men: prospective US cohort study
Dr. Flegal raises an interesting point regarding the potential measurement errors of using anthropometric prediction equations developed from the National Health and Nutrition Examination Survey (NHANES) to estimate body composition in the Health Professionals Follow-up Study (HPFS).
Dr. Flegal’s concern mainly arises from the difference between anthropometric measures in the NHANES and the HPFS. The prediction equations were developed using measured height, weight, and waist circumference in the NHANES, but the HPFS had information on self-reported height and weight and (self) measured waist circumference.(1, 2) Dr. Flegal points out that self-reported height and weight have systematic errors due to factors such as age, sex, race, and education. We agree that misreporting could be affected by those factors. However, the HPFS cohort included only men, predominantly White health professionals with similar socioeconomic status and (high) education level. Age was well controlled in all our analyses. Thus, it is highly unlikely for systematic errors to have occurred by these factors in the self-reported measures in the HPFS that would have significantly affected our results. This assertion is supported by our previous validation study conducted using the sample of HPFS.(3) The correlations between self-reported and technician measured weight and waist circumference were 0.97 and 0.95, respectively. The mean (standard deviation) of self-reported and technician measured values were 174 (23.8) and 177 (25.7) for weight (lb) and 37.3 (3.6) and 37.0 (3.5) for waist circumference (inch), respectively. Self-reported and measured height has already been shown to have high validity with minimal differences in previous validations.(4, 5)
Dr. Flegal also mentions the different anatomical sites of waist circumference measured in the NHANES (iliac crest) and the HPFS (umbilicus). However, a previous study on primarily White samples showed very high correlation and approximately the same means and standard deviations between the measurements from the two different anatomical sites in men.(6) The correlation between iliac crest and umbilicus was 0.99 and the mean (standard deviation) was 97.8 (12.8) for iliac crest (cm) and 98.3 (12.6) for umbilicus (cm). No significant difference was found between these two sites and also across four different anatomical sites including iliac crest, umbilicus, midpoint, and minimal waist among men. Moreover, different anatomical sites of waist circumference were shown to have no meaningful effect on the association of waist circumference with major chronic diseases and mortality.(7)
Overall, the remarkably high r-squares and very similar means and standard deviations between self-reported and measured anthropometric measures ensure that our reliance on self-reported measures did not affect the results in any meaningful way.
The overall contribution of our study should be weighed by all of the factors that may affect validity. Our study had a large sample size with 25 years of follow-up that allowed time lag analysis, repeated measures, detailed covariate data, and essentially complete follow-up. A study with a "perfect measure" of lean body mass and fat mass but small sample size, short follow-up, single measure, and no detailed confounder information would have more problems than what the slightly more accurate measure could offset. We believe our study can provide important insights into the relationship between obesity (including body composition) and mortality.
Finally, Dr. Flegal noted that using weight, height, and waist circumference directly would be preferable over an imputation, as had been done in the HPFS previously. It is not clear to us why a variable accounting for two factors only (e.g., height and weight in BMI) is superior to one additionally accounting for waist circumference. For example, as had been shown in the NHANES previously, in men addition of waist circumference adds immensely to the estimation of fat mass than BMI alone.(1, 8) Intentionally or not, BMI is used to estimate adiposity, so why not try to improve on it if feasible?
1. Lee DH, Keum N, Hu FB, Orav EJ, Rimm EB, Sun Q, et al. Development and validation of anthropometric prediction equations for lean body mass, fat mass and percent fat in adults using the National Health and Nutrition Examination Survey (NHANES) 1999–2006. British Journal of Nutrition. 2017;118(10):858-66.
2. Lee DH, Keum N, Hu FB, Orav EJ, Rimm EB, Willett WC, et al. Predicted lean body mass, fat mass, and all cause and cause specific mortality in men: prospective US cohort study. Bmj. 2018;362:k2575.
3. Rimm EB, Stampfer MJ, Colditz GA, Chute CG, Litin LB, Willett WC. Validity of self-reported waist and hip circumferences in men and women. Epidemiology. 1990;1(6):466-73.
4. Palta M, Prineas RJ, Berman R, Hannan P. Comparison of self-reported and measured height and weight. American journal of epidemiology. 1982;115(2):223-30.
5. Pirie P, Jacobs D, Jeffery R, Hannan P. Distortion in self-reported height and weight data. Journal of the American Dietetic Association. 1981;78(6):601-6.
6. Mason C, Katzmarzyk PT. Variability in waist circumference measurements according to anatomic measurement site. Obesity. 2009;17(9):1789-95.
7. Ross R, Berentzen T, Bradshaw AJ, Janssen I, Kahn HS, Katzmarzyk PT, et al. Does the relationship between waist circumference, morbidity and mortality depend on measurement protocol for waist circumference? Obesity reviews. 2008;9(4):312-25.
8. Lee WS. Body fatness charts based on BMI and waist circumference. Obesity. 2016;24(1):245-9.
Competing interests: No competing interests