Diagnosing Obesity by Body Mass Index in Chronic Kidney Disease
An Explanation for the “Obesity Paradox?”
Although obesity is associated with poor outcomes, among patients with chronic kidney disease (CKD), obesity is related to improved survival. These results may be related to poor diagnostic performance of body mass index (BMI) in assessing body fat content. Accordingly, among 77 patients with CKD and 20 controls, body fat percentage was estimated by air displacement plethysmography (ADP), skinfold thickness, and body impedance analysis. Defined by BMI ≥30 kg/m2, the prevalence of obesity was 20% in controls and 65% in patients with CKD. Defined by ADP, the prevalence increased to 60% among controls and to 90% among patients with CKD. Although sensitivity and positive predictive value of BMI to diagnose obesity were 100%, specificity was 72%, but the negative predictive value was only 30%. BMI correctly classified adiposity in 75%. Regardless of the presence or absence of CKD, subclinical obesity (defined as BMI <30 kg/m2 but excess body fat by ADP) was often missed in people with low lean body mass. The adjusted odds ratio for subclinical obesity per 1 kg of reduced lean body mass by ADP was 1.14 (95% CI: 1.06 to 1.23; P<0.001). Skinfold thickness measurements correctly classified 94% of CKD patients, but bioelectrical impedance analyzer–assessed body fat estimation did so in only 65%. Air displacement plethysmography–, skinfold thickness–, and bioelectrical impedance analyzer–assessed body fat all provided reproducible estimates of adiposity. Skinfold thickness measurements may be a better test to classify obesity among those with CKD. Given the low negative predictive value of BMI for obesity, our study may provide an explanation of the “obesity paradox.”
- chronic kidney disease
- skinfold thickness
- body impedance analysis
- body composition assessment
One in 3 adult in the United States is obese, and the epidemic of obesity continues to grow.1 Obesity is associated with a variety of adverse health consequences, such as cardiovascular disease, dyslipidemia, diabetes mellitus, and shortened life span in the general population, yet among patients with chronic kidney disease (CKD), obesity is paradoxically associated with better outcomes.2 This paradoxical association has been mostly found among patients on hemodialysis, but data among people with CKD not yet on dialysis also point out the same paradoxical association.3 The “reverse epidemiology” of obesity has not been explained by conventional cardiovascular risk factors. Whether this is because of a true association or because of poor diagnostic performance of clinical methods to assess obesity is unclear.
The World Health Organization defines obesity as the presence of excess body fat.4 Body fat is considered excessive when it is >25% in men and >35% in women.4 Body fat is difficult to measure directly. Therefore, in clinical practice, body mass index (BMI) is commonly used to diagnose obesity. However, BMI can be influenced by muscle mass, and its ability to diagnose obesity can vary considerably by predictors of muscle mass, such as age, sex, and race. Among patients with CKD, who often are elderly and frail, lean body mass may be reduced. Furthermore, volume overload that often accompanies CKD by itself can influence BMI estimation. Among patients with CKD, therefore, BMI may not accurately reflect excess body fat. We hypothesized that BMI is a poor surrogate for adiposity among those with CKD. Furthermore, if BMI is found to have poor diagnostic performance, then alternative methods must be sought to better assess body fat content.
The gold standard for body composition assessment is air displacement plethysmography (ADP).5 ADP uses whole-body densitometry to determine body composition. It is based on the same reference standard operating principle as underwater weighing, except that air displacement, instead of water displacement, is used to provide quick and convenient results. In comparison with other body composition assessment methods, ADP has several advantages. For example, it eliminates the radiation exposure inherent in dual energy x-ray absorptiometry, as well as the difficulties associated with underwater submersion in hydrostatic weighing. However, ADP requires specialized expensive equipment and requires the subject to wear compression underwear to perform body composition assessment. ADP, therefore, remains largely a research technique and a reference standard.
Two other techniques are relatively easy to perform. Skinfold thickness (SFT) measurement is performed through calipers, and an experienced operator can perform this in a few minutes. Likewise, body impedance analysis (BIA) is simple and inexpensive. Accordingly we assessed these 2 research techniques to assess body fat content and compared it to the reference standard of ADP.
We hypothesized that BMI does not assess obesity well among patients with CKD. We reasoned that if BMI is a poor indicator of body fat, it would suggest that the association of BMI with outcomes may not be because of obesity. The purpose of our study was, therefore, to compare obesity assessed by BMI with the gold standard of ADP. To improve the assessment of obesity, we also assessed fat percentage assessed by SFT and BIA compared with ADP.
Materials and Methods
We studied patients with CKD stages 3 and 4 and blood pressure <140/90 mm Hg in the seated position in the clinic. Patients with overt proteinuria with stage 2 CKD were also included in the study. Patients were recruited from the Roudebush Veterans’ Administration Medical Center and Wishard Memorial Hospital (Indianapolis, IN). As a control group, 20 veterans without CKD were recruited from the medicine clinic of the Roudebush Veterans’ Administration Medical Center. To participate, these non-CKD controls had to be nonsmokers with BMI <40 kg/m2, estimated glomerular filtration rate >60 mL/min per 1.73 m2, and no history of CKD, myocardial infarction, stroke, diabetes mellitus, or hypertension. All of the body composition measurements were made after an overnight fast on the same day. To assess test-retest reliability, among 40 patients, all of the measurements were repeated after 4 to 8 weeks of the initial measurement. The study protocol was approved by the Institutional Review Board for protection of human subjects of the Indiana University and the research and development committee of the Roudebush Veterans’ Administration Medical Center. All of the study subjects gave written informed consent.
Air Displacement Plethysmography
ADP was measured using the BOD POD Gold Standard Body Composition Tracking System (Life Measurement, Inc). The ADP system consisted of the air plethysmograph, a digital scale, and computer software (BOD POD version 4.2+). Each participant was asked to change into compression shorts (for men) or a swimsuit (for women) and a swim cap and to remove any jewelry. Body mass was measured to the nearest 0.001 kg using the electronic scale before the body volume measurement. Height was taken using a Seca 222 measuring rod (Seca Group). The subject stood with his or her back to the measuring rod, and the measuring slide was pushed onto the head so that the measuring slide abutted without bending. The height was read to the nearest millimeter. The subject then entered the ADP chamber and was instructed not to move. There were 2 body volume readings, with the door being opened in between each reading. In the event of a discrepancy between the 2 readings, a third reading was taken. To measure thoracic gas volume, the subject, after a prompt, inhaled and exhaled multiple times while holding his or her nose and making a tight seal around the tube attached to the ADP system. They were then instructed to exhale lightly 3 times. After thoracic gas volume was measured, they exited the ADP system and were examined for the presence of pedal edema. Body fat percentage was calculated using sex- and race-specific equations, as follows: (1) for nonblacks we used the Siri equation6; (2) for black men we used the Shutte equation7; and (3) for black women we used the Ortiz equation.8
Skinfold Thickness Measurements
Skinfold thickness was measured using a Lange Skinfold Caliper (Beta Technologies) on both sides of the body in triplicate at 4 locations: biceps, triceps, subscapular, and supra iliac. The measurements were made by pinching the skin with the thumb and index finger, as follows. Biceps skinfold thickness was measured at the midpoint of the arm with the patient sitting with arms relaxed in the supine position resting on thighs. Triceps skinfold thickness was measured at the midpoint between the acromion and the olecranon process in the sitting position with arms crossed at a 90° bend, resting on thighs. The subscapular skinfold was measured with the patient standing with arms to the side. The shoulder blade was found and followed down to where it started to curve. The skin was pinched and measured with the calipers. The supra iliac skinfold was also measured with the patient standing. The skin above the right hipbone along the midaxillary line was measured. The average skinfold thickness measurement from each of the 4 sites was used in the Durnin and Womersely equation to predict the percentage of body fat.9
Using a measuring tape, waist and hip circumferences were measured to the nearest 0.1 cm. The tape was snug but not so tight that it compressed the underlying soft tissue. Waist circumference was measured with the subject standing comfortably with his or her weight distributed evenly with feet ≈25 to 30 cm apart. The measurement was taken midway between the inferior margin of the last rib and the crest of the ilium in a horizontal plane. Hip circumference was measured with the subject standing with his or her arms at the side and feet together. The person taking the measurement sat at the side of the subject so that the level of maximum extension of the buttocks could be seen. The measuring tape was placed around the buttocks in a horizontal plane at the maximum extension.
Body Impedance Analysis
Body impedance was measured using a bioelectrical impedance analyzer (BIA) (RJL Systems). Subjects were asked to lie down on the bed with their right shoe and sock removed. They were instructed to keep still with feet apart and their hands not touching their body. Electrodes were placed at 4 locations on the right side of the body as follows: (1) right wrist adjacent to an imaginary line bisecting the ulnar head; (2) base of the right middle finger; (3) right ankle adjacent to an imaginary line bisecting the inside ankle (medial malleolus); and (4) base of the index toe. Leads were attached to the electrodes as shown in the user manual and plugged into the BIA. Numbers corresponding to resistance, reactance, impedance, and phase angle were recorded in triplicate. Equations used to assess body composition from population-based studies were used to assess body fat.10
Using ADP measurements, obesity was defined as body fat percentage >25% in men and >35% in women according to the World Health Organization.4 Using these criteria, obesity was present in all but 6 of the patients. Given that obesity was nearly universal in our population, receiver operating characteristic curve analysis was not performed. We assessed the relationship between BMI and ADP body fat percentage by examining correlations between the 2 variables. We next examined the relationship between BMI and ADP lean mass. Lean mass was calculated by multiplying the percentage of lean mass by body weight.
We defined subclinical obesity as the presence of high body fat percentage but BMI <30 kg/m2. We defined overt obesity as the presence of high body fat percentage and BMI >30 kg/m2. To explore the determinants of subclinical obesity, we assessed potential explanatory variables (as shown in Table 3). We hypothesized that muscle mass may be reduced in patients with subclinical obesity. Accordingly, we used lean mass assessed by ADP to reflect muscle mass. Mid-arm circumference and triceps skinfold thickness were also used to calculate bone-free arm muscle area. Corrected arm muscle areas were calculated from triceps skinfold thickness and midarm circumference using the formulas used by Schmidt et al.11
To examine the relationship of SFT and BIA we used Bland-Altman plots and Lin concordance correlation coefficients. We finally examined the relationship between weight and body fat using a multivariate regression analysis.
Table 1 shows the baseline characteristics of the patients. The sample included predominantly men and was composed of older individuals. Table S1 (please see the online Data Supplement at http://hyper.ahajournals.org) shows baseline characteristics by CKD stage. Notably, compared with non-CKD controls and stage 3A CKD, the physical activity score was reduced at stages 3B and 4 CKD. Table 2 shows the body composition assessment by various techniques. Patients with CKD had a greater BMI and had greater adiposity. However, the lean body mass, on average, was comparable between controls and those with CKD. Table S2 shows body composition by CKD stage. Increasing adiposity was noted with higher CKD stages most clearly by ADP assessment and less so by BIA assessment.
BMI as a Screening Tool for Obesity
As defined by BMI ≥30 kg/m2, the prevalence of obesity in non-CKD controls was 20%. By comparison, the prevalence of obesity among CKD patients was 65%. However, applying the gold standard of ADP-measured body fat, the prevalence of obesity increased to 60% among non-CKD patients and to an astounding 90% among patients with CKD.
Figure 1 shows that ADP-assessed percentage of body fat or adiposity was related to BMI (r=0.67; P<0.01). At higher levels of adiposity there was greater variation in BMI. For example, at 40% body fat, BMI could vary from <30 kg/m2 to >40 kg/m2. No patient in our study was so muscular that BMI was increased and misclassified him or her as obese (n=0 in top left quadrant of Figure 1). Thus, BMI ≥30 kg/m2 had 100% specificity and 100% positive predictive value for obesity. However, using the same threshold of BMI to classify obesity, 30% of patients with CKD would not be classified as obese, although they would meet the definition of obesity by the gold standard test of obesity (negative predictive value of BMI 30%). Accordingly, BMI as a screening tool to detect obesity would miss 30% of the patients. BMI was 72% sensitive in detecting obesity and correctly classified adiposity in 75% of the patients.
Among people without CKD, as in patients with CKD, BMI ≥30 kg/m2 had 100% specificity and 100% positive predictive value for obesity. However, the sensitivity of this threshold to detect obesity was only 33% and negative predictive value only 50%. BMI correctly classified adiposity in 60% of the patients.
Relationship of BMI to Lean Body Mass and Comparison With SFT and BIA
Other than being related to percentage of body fat, BMI was also related to ADP-assessed lean body mass (Figure S1). The relationship between lean body mass and BMI (r=0.53; P<0.01) was similar to the relationship between percentage of body fat and BMI (r=0.67; P<0.01). In contrast to significant BMI and lean body mass relationship, the relationships of lean body mass to either SFT-assessed percentage of fat (Figure 2,middle) or BIA-assessed percentage of fat (Figure 2, bottom) were not significant.
Correlates of Subclinical Obesity
To explore the determinants of subclinical obesity, we assessed potential explanatory variables, as shown in Table S3. Bivariate analysis showed that patients with subclinical obesity were less often diabetic, but other demographic characteristics were well matched. We hypothesized that muscle mass may be reduced in patients with subclinical obesity. Compared with those with overt obesity, among those with subclinical obesity, ADP-assessed lean mass was 9.4 kg less (P<0.0001; Table 3). Similarly, arm circumference and bone-free arm muscle area were significantly different between groups. Adjusted for age, CKD, and diabetes mellitus, a multivariable logistic regression model revealed that, compared with overt obesity, the odds for subclinical obesity per 1 kg of reduced lean body mass by ADP were 1.14 (95% CI: 1.06 to 1.23; P<0.001). Similarly, the odds for subclinical obesity per 1-cm2 reduced bone-free arm muscle area were 1.08 (95% CI: 1.03 to 1.13; P=0.001).
Diagnostic Performance of SFT and BIA in Assessing Obesity
Because BMI performed poorly to detect obesity, we assessed the diagnostic performance of 2 simple and readily available tests, skin-fold thickness and BIA to assess adiposity.
The following were the diagnostic test performance results among non-CKD controls: sensitivity 100%, specificity 13%, positive predictive value 63%, and negative predictive value 100%, and 65% were correctly classified (Figure S2, right). For CKD patients, the diagnostic test performance results were as follows: sensitivity 99%, specificity 50%, positive predictive value 94%, and negative predictive value 80%, and 94% were correctly classified (Figure S2, left).
Compared with the line of identity, the relationship between SFT-assessed body fat percentage and adiposity was flatter (Figure 2, top left). Unlike the BMI-adiposity relationship shown in Figure 1, there was no megaphone shape to the scatter plot. Thus, the error in the assessment of adiposity was similarly distributed at all levels of body fat.
SFT-assessed body fat did not, on average, overestimate or underestimate body fat percentage (Figure 2, bottom left). The limits of agreement were wide; individual estimates could be off by 13%. At higher levels of adiposity, SFT-assessed body fat underestimated percentage of body fat. At lower levels of adiposity, it overestimated percentage of body fat.
The following were the diagnostic test performance results among non-CKD controls: sensitivity 42%, specificity 88%, positive predictive value 83%, and negative predictive value 50%, and 60% were correctly classified (Figure S3, right). For CKD patients, the results were as follows: sensitivity 63%, specificity 86%, positive predictive value 98%, and negative predictive value 20%, and 65% were correctly classified (Figure S3, left).
Compared with the line of identity, the relationship between BIA-assessed body fat percentage and adiposity was flatter (Figure 2, top right). Unlike the BMI-adiposity relationship shown in Figure 1, there was no megaphone shape to the scatter plot. Thus, the error in the assessment of adiposity was similarly distributed at all of the levels of body fat.
BIA-assessed body fat, on average, underestimated body fat percentage by 9% (Figure 2, bottom left). The limits of agreement were wide; individual estimates could be off by ≈11%. At higher levels of adiposity, BIA-assessed body fat underestimated percentage of body fat. At lower levels of adiposity, it overestimated percentage of body fat. Thus, the concordance correlation coefficient was lower (0.44) compared with SFT (0.56).
Test-Retest Reliability of ADP-, SFT-, and BIA-Assessed Body Fat
Among 40 participants, we repeated the assessment of body composition by all 3 of the techniques within 2 months. The results of the test-retest reliability are shown in Figure 3. Agreement between 2 measurements was best in the case of ADP (concordance correlation coefficient: 0.947), followed by SFT, and least for BIA.
The major findings of our study are the following: (1) the prevalence of obesity among patients with CKD is high but is even higher when measured directly; (2) normal BMI does not exclude obesity, although a high BMI rules it in; (3) regardless of presence or absence of CKD, subclinical obesity is often missed in people with low lean body mass; (4) skinfold thickness measurements can exclude obesity with a high degree of certainty; BIA-assessed body fat estimation does not rule out obesity; (5) SFT-assessed body fat percentage can detect the majority of subclinical obesity; and (6) ADP-, SFT-, and BIA-assessed body fat all provide reproducible estimates of adiposity.
Similar results were reported by Romero-Corral et al14 among patients with coronary artery disease. Using ADP-assessed body fat at the gold standard, BMI of ≥30 kg/m2 had excellent specificity (95%) and positive predictive value (97%) in detecting obesity. The authors reported poor sensitivity (43%) and negative predictive value (59%) of BMI in diagnosing obesity, which is in keeping with our findings. BMI could not distinguish reliably between fat mass and lean body mass in their study, as also noted in our study.
Our data also support the studies of Beddhu et al. These investigators demonstrated that the survival advantage of high BMI among CKD patients on long-term dialysis was limited to those with normal or increased muscle mass. Patients with high BMI and high body fat had increased all-cause and cardiovascular mortality. In contrast to Beddhu et al,15 who estimated muscle mass from 24-hour urine creatinine, we directly measured lean body mass by the gold-standard measurement of ADP. Our study not only supports their observations but extends them in calling attention to body fat excess among those with low BMI. Our study, therefore, calls into question the accuracy of BMI in predicting body fat. These results support the findings of Postorino et al,16 who found that, whereas BMI was inversely related to mortality among dialysis patients, surrogate measures of abdominal obesity and segmental fat distribution were directly associated with mortality. Skinfold thickness measurement appears to be an attractive and simple-to-implement technique when screening for obesity among those with CKD. It correctly classified adiposity in 94% of the patients.
Pathophysiological studies have revealed that, among patients with CKD, obesity is not an innocuous bystander and may directly or indirectly damage the kidney.17 Evidence for the direct damaging effect of obesity is the following. Because of heightened sympathetic activity, high levels of angiotensin II, and hyperinsulinemia, obesity is often accompanied by glomerular hyperfiltration and increased proximal tubular sodium resorption. Enhanced proximal salt reabsorption determines a reduced delivery of sodium to the macula densa. This then causes afferent vasodilatation and enhanced renin synthesis. As a result of high local angiotensin II levels, the efferent arteriole is constricted in the obese, and glomerulomegaly and focal glomerulosclerosis ensue. Evidence is emerging that fat cells may trigger inflammation in the kidney indirectly by producing inflammatory cytokines, which may further aggravate renal dysfunction.
There are several limitations of our study. We did not study a random sample of the CKD population but only those willing to participate in these studies. There were few women and mostly elderly patients in our sample. Given the cross-sectional nature of our study, we cannot infer causality. A larger cohort with longitudinal follow-up may begin to better address the questions of what might explain the reverse epidemiology of obesity in CKD.
Using BMI to detect obesity among those with CKD may miss a large number of people with excess body fat. Often these people have low muscle mass. Using a caliper to measure skinfold thickness may detect these patients in a more reliable way. Given the low negative predictive value of BMI for obesity, our study may provide an explanation of the “obesity paradox.” Perhaps better measurements of risk factors, such as blood pressure, by ambulatory or home BP monitoring18 and obesity by ADP can reverse the “reverse epidemiology”2 and provide support to the notion that neither being hypertensive nor being fat may be good, even for those with CKD.
Sources of Funding
This work was supported by a Veterans’ Administration Merit Review grant to R.A.
- Received August 4, 2010.
- Revision received August 21, 2010.
- Accepted August 31, 2010.
Kalantar-Zadeh K, Abbott KC, Salahudeen AK, Kilpatrick RD, Horwich TB. Survival advantages of obesity in dialysis patients. Am J Clin Nutr. 2005; 81: 543–554.
World Health Organization Expert Committee on Physical Status: the Use and Interpretation of Anthropometry. Physical Status: The Use and Interpretation of Anthropometry–Report of a WHO Expert Committee. Geneva, Switzerland: World Health Organization; 1995.
Fields DA, Goran MI, McCrory MA. Body-composition assessment via air-displacement plethysmography in adults and children: a review. Am J Clin Nutr. 2002; 75: 453–467.
Siri WE. Body composition from fluid spaces and density: analysis of methods. In: Brozek J, Henschel A, eds. Techniques for Measuring Body Composition. Washington, DC: National Academy of Science; 1961: 223–2944.
Schutte JE, Townsend EJ, Hugg J, Shoup RF, Malina RM, Blomqvist CG. Density of lean body mass is greater in blacks than in whites. J Appl Physiol. 1984; 56: 1647–1649.
Ortiz O, Russell M, Daley TL, Baumgartner RN, Waki M, Lichtman S, Wang J, Pierson RN Jr, Heymsfield SB. Differences in skeletal muscle and bone mineral mass between black and white females and their relevance to estimates of body composition. Am J Clin Nutr. 1992; 55: 8–13.
Sun SS, Chumlea WC, Heymsfield SB, Lukaski HC, Schoeller D, Friedl K, Kuczmarski RJ, Flegal KM, Johnson CL, Hubbard VS. Development of bioelectrical impedance analysis prediction equations for body composition with the use of a multicomponent model for use in epidemiologic surveys. Am J Clin Nutr. 2003; 77: 331–340.
Romero-Corral A, Somers VK, Sierra-Johnson J, Jensen MD, Thomas RJ, Squires RW, Allison TG, Korinek J, Lopez-Jimenez F. Diagnostic performance of body mass index to detect obesity in patients with coronary artery disease. Eur Heart J. 2007; 28: 2087–2093.
Beddhu S, Pappas LM, Ramkumar N, Samore M. Effects of body size and body composition on survival in hemodialysis patients. J Am Soc Nephrol. 2003; 14: 2366–2372.
Agarwal R. Blood pressure and mortality among hemodialysis patients. Hypertension. 2010; 55: 762–768.