Jump to
Abstract
Unlike the general population, among hemodialysis patients body mass index (BMI) is related to blood pressure (BP) and mortality inversely. To explore the reasons for this risk-factor paradox, the cross-sectional association of obesity with the following factors was examined: the prevalence of hypertension, its control, and echocardiographic left ventricular mass index (LVMI). Longitudinal follow-up explored the relationship of BMI with all-cause mortality. Furthermore, it explored whether poorer survival in leaner individuals was related to either high BP or greater LVMI. Among 368 hemodialysis patients, both the prevalence of hypertension and its poor control were inversely related to BMI. BMI was also inversely associated with evidence of excess extracellular fluid volume, but adjustment for this variable did not completely remove the inverse relationship between BP and BMI. Over 1122 patient-years of cumulative follow-up (median: 2.7 years), 119 patients (32%) died. In the first 2 years of follow-up, the mortality hazard for the lowest BMI group was increased; thereafter, the survival curves were similar. Adjusting for several risk factors including BP and LVMI did not remove the inverse relationship of BMI with mortality. In conclusion, leaner patients on dialysis have a higher prevalence of hypertension, poorer control of hypertension, more LVMI, and greater evidence of extracellular fluid volume excess. However, volume explains the greater prevalence or poorer control of hypertension only partially. Leaner patients have an accelerated mortality rate in the first 2 years; this is not completely explained by BP, LVMI, or other cardiovascular or dialysis-specific risk factors.
- body mass index
- epidemiology
- hemodialysis
- ambulatory blood pressure
- left ventricular hypertrophy
- survival
Introduction
See Editorial Commentary, pp 989–990
The association of body mass index (BMI) with mortality in the general population is well recognized.1 There are >400 000 patients on long-term dialysis in the United States. Among these people, many epidemiological studies indicate that, unlike the general population, BMI is not inversely but directly associated with survival.2–8 The reasons for this paradoxical observation, termed risk-factor paradox or reverse epidemiology, are not completely understood.
Unlike the general population, obese people on dialysis also have lower blood pressure (BP).9,10 The reasons underlying this paradoxical association are also unclear. One reason could be that individuals with higher BMI are better able to sequester excess extracellular fluid volume. If so, markers of extracellular fluid volume excess among these individuals would be inversely related to BMI. Also, if these markers lie in the causal pathway, correction for these markers would diminish the relationship between BP and BMI. Furthermore, if BMI is truly associated with lower BP, then it must also be associated with less target organ damage measured as left ventricular mass index (LVMI).11 Whether this is so is also unknown.
The survival studies relating obesity to mortality often do not have extended follow-ups; some of them have been carried out only over 1 year.5,6 Notably, studies in the general population have excluded deaths in the first 5 years of follow-up to exclude reverse causation.1 The limited duration of these studies leaves open the possibility that leaner individuals have an accelerated death rate. Although this hypothesis has been proposed, so far there have been few data to support it.12 Finally, whether BP and LVMI are mediators of poorer survival among leaner individuals is also unknown.
In this study we examined the association of the prevalence and control of BP, as well as echocardiographic LVMI, with obesity. We sought the association of echocardiographic markers of volume excess with obesity and asked the question of whether excess volume is a mediator of prevalence and control of hypertension. In longitudinal follow-up of ≤8 years we explored the relationship of all-cause mortality and obesity and whether poorer survival in leaner individuals is related to high BP and LVMI.
Methods
Participants
Portions of this cohort have been described previously.11,13 Patients ≥18 years of age who had been on chronic hemodialysis for >3 months and were free of vascular, infectious, or bleeding complications within 1 month of recruitment who were dialyzed 3 times per week at 1 of the 4 dialysis units in Indianapolis affiliated with Indiana University were enrolled in the study. Those who missed ≥2 hemodialysis treatments over 1 month, abused drugs, had chronic atrial fibrillation, or had BMI of ≥40 kg/m2 at screening visit were excluded. Patients who had a change in dry weight or antihypertensive drugs within 2 weeks were also excluded. The study was approved by the institutional review board of Indiana University and the research and development committee of the Roudebush Veterans' Affairs Medical Center, and all of the subjects gave written informed consent.
Definition of BMI and Obesity
BMI was calculated as postdialysis weight (in kilograms) divided by height (in meters squared). Obesity was defined by BMI according to the World Health Organization14 (the ranges used to classify are shown in Table 1).
Descriptive Characteristics of the Study Population
Measurements
Ambulatory BP Monitoring and Definitions of Hypertension
Ambulatory BP monitoring was performed either after the first or midweek hemodialysis session for 44 hours. Ambulatory BP was recorded every 20 minutes during the day (6:00 am to 10:00 pm) and every 30 minutes during the night (10:00 pm to 6:00 am) using a SpaceLabs 90207 ambulatory BP monitor (SpaceLabs Medical Inc, Redmond, WA) in the nonaccess arm, as reported previously.15 In this study, patients who had <8 hours of ambulatory BP recordings were noted to have inadequate measurement and were excluded.
Hourly average of ambulatory BP was first computed. These averages were then averaged over the 44 hours of recording to yield the overall ambulatory BP. Ambulatory BP values ≥135/85 mm Hg were considered hypertensive.16 Also, any patient on antihypertensives was considered to be hypertensive. If the ambulatory BP was ≥135/85 mm Hg, then the patient was considered to be poorly controlled.
Echocardiograms
2D guided M-mode echocardiograms were performed by research echocardiographic technicians, 30 to 60 minutes after dialysis, in the dialysis unit with a digital cardiac ultrasound machine (Cypress Acuson, Siemens Medical) within 10 days of ambulatory BP recording, as reported previously.11
The left atrial diameter indexed for body surface area and the inferior vena cava diameter in expiration also indexed for body surface area were imaged as described previously.17 They have been shown previously to be markers of volume and, therefore, chosen as extracellular fluid volume markers for this analysis.17
Data Analysis
Descriptive statistics for demographic and clinical variables related to the categories of BMI were provided. Race was combined into 2 categories, black and nonblack. Dialysis vintage was categorized into 3 groups, dialysis <1 year, dialysis 1 to 4, and dialysis >4 years. The number of antihypertensives was capped at 4, because generally few patients were on >4 medications.
Odds ratio (OR) for prevalence was calculated by logistic regression. BMI categories were used as determinants, with the highest category used as reference. Given the small number of patients, normal, underweight, and severely underweight BMI categories were merged, as were grade 2 or 3 obesity categories. A multivariable model was then used to adjust for the following covariates: sex, smoking, urea reduction ratio, aspirin use, diabetes mellitus, and serum albumin. These covariates were selected based on their association with BMI. Stepwise forward-selection logistic regression was performed with factors added at the 0.15 level of significance. Similar models were fitted for poor control of hypertension.
Next, logistic regression models were constructed with each of the 2 markers of volume (inferior vena cava diameter and left atrial diameter) used separately to predict the prevalence of (models 3 and 4 in Table 2) or the lack of control of hypertension (models 4 and 5 in Table S1, available in the online Data Supplement). Sensitivity analyses were performed using multiple imputation for missing data for all of the logistic models using the “mi” set of commands in Stata.
Odds Ratios for Prevalence of Hypertension by Ambulatory Blood Pressure Monitoring
A linear regression model was used to predict echocardiographic LVMI (model 1, Table 3). This model was further adjusted for the following covariates: sex, smoking, urea reduction ratio, aspirin use, diabetes mellitus, and serum albumin using stepwise multivariable regression with covariates added at the 0.15 level of significance (model 2, Table 3). This model was further adjusted for interdialytic ambulatory systolic BP (model 3, Table 3). Sensitivity analyses were performed using multiple imputation as above.
Association of Body Mass Index With Left Ventricular Mass Index
Survival was analyzed by Kaplan-Meier methods (Figure 3) and Cox proportional hazards regression (Table 4). BMI was used as a continuous variable in the Cox model (model 1, Table 4). The proportionality assumption was violated; therefore, all of the subsequent models contained the BMI × time interaction term. Introducing this interaction term showed no violation of the proportionality assumption for other covariates (as tested by the Schoenfeld residuals). The model was further adjusted for covariates shown in Table 4.
Effects of Body Mass Index on Survival
All of the analyses were conducted using Stata 11.0 (Stata Corp, College Station, TX). The P values reported are 2 sided and taken to be significant at <0.05.
Results
Of the 441 patients who consented, 1 was missing BMI, 7 had inadequate ambulatory BP recordings, and 65 had none. These 368 patients who had measurements of both BMI and interdialytic ambulatory BP formed the study cohort: 316 (86%) of these also had echocardiographic data (please see Figure S1).
Table 1 shows the characteristics of the patients by BMI categories. Overall, we studied 368 patients with a mean BMI of 27.7, age 55 years, two thirds were men, one third were active smokers, and 85% were black. The etiology of end-stage renal disease was diabetes mellitus in 35% and hypertensive nephrosclerosis in nearly half. BMI was associated with sex (more obesity among women), smoking (smokers were leaner), diabetes mellitus, urea reduction ratio, serum albumin, and aspirin use (more aspirin use among obese). There was more evidence of volume expansion by echocardiographic criteria among leaner patients.
Table 2 shows the OR for the prevalence of hypertension by interdialytic ambulatory BP monitoring. The unadjusted model (model 1) showed a significant inverse relationship between the odds of being hypertensive and BMI categories. In a separate model, BMI used as a continuous variable also showed a similar inverse relationship (OR: 0.95; P=0.04). Model 2 shows ORs for only the significant determinants in which BMI categories were adjusted in a stepwise forward logistic regression for the following variables: sex, smoking, urea reduction ratio, aspirin use, diabetes mellitus, and serum albumin. Diabetes mellitus and serum albumin emerged as significant predictors of prevalence of hypertension. However, BMI categories still remained significantly and inversely associated with prevalent hypertension. As a continuous variable, BMI adjusted for diabetes mellitus and serum albumin remained significantly and inversely associated with hypertension (OR: 0.95; P=0.03). Further adjustment of model 2 for left atrial diameter (model 3) reduced the strength of the statistical association of BMI categories (P=0.07) and BMI (OR: 0.96; P=0.1) with hypertension. However, imputing for missing data showed that the statistical association of BMI categories (P=0.04) and BMI (OR: 0.95; P=0.07) with hypertension were both improved. Adjustment of model 2 for inferior vena cava diameter in expiration (model 4) removed the strength of the statistical association of BMI (OR: 0.94; P=0.06) but not BMI categories (P=0.03) with hypertension. Thus, BMI categories were significantly and inversely related to hypertension even after accounting for markers of volume.
Table S1 shows the OR for the control of hypertension. The unadjusted model (model 1) showed a significant inverse relationship between the odds of being poorly controlled hypertensive and BMI categories. BMI used as a continuous variable also showed a similar inverse relationship (OR: 0.96; P=0.01). Model 2 shows ORs for only the significant determinants in which BMI categories were adjusted in a stepwise forward logistic regression for the variables noted in methods. Only serum albumin emerged as a weak direct determinant of control of hypertension. However, BMI categories remained significantly and inversely associated with lack of control of hypertension. BMI adjusted for serum albumin remained significantly and inversely associated with hypertension (OR: 0.96; P=0.02). Further adjustment of model 2 for number of antihypertensive medications (model 3) reduced the statistical association of BMI categories (P=0.09) and BMI (OR: 0.97; P=0.1) with lack of control of hypertension. Imputation for missing data did not alter the results. Adjustment of model 3 for left atrial diameter indexed for body surface area (model 4) increased the statistical association of BMI (OR: 0.96; P=0.05) and BMI categories (P=0.04) with the lack of control of hypertension. Adjustment of model 3 for inferior vena cava diameter in expiration (model 5) removed the statistical association of BMI (OR: 0.96; P=0.1) but not BMI categories (P=0.05) with the lack of control of hypertension.
The distribution of BMI and the prevalence and lack of control of hypertension are shown in Figure 1. The solid lines represent the unadjusted estimates. The dotted lines are either prevalence adjusted for diabetes mellitus and serum albumin or lack of control adjusted for serum albumin.
Distribution of body mass index (BMI), prevalence of hypertension, and lack of control of hypertension. The solid lines represent the unadjusted estimates. The dotted lines are either prevalence adjusted for diabetes mellitus and serum albumin or lack of control adjusted for serum albumin. Prevalence of hypertension or lack of control are shown on the right y axis.
We next evaluated the association of BMI with target organ damage (LVMI). Table 3 shows a significant and inverse relationship of BMI with unadjusted LVMI (model 1). The model intercept was 172.8 g/m2. Stepwise forward multivariable linear regression for the following variables: sex, smoking, urea reduction ratio, aspirin use, diabetes mellitus, and serum albumin, removed the statistical association of BMI with LVMI. The significant determinants of LVMI are shown in model 2. Further adjustment of model 2 for interdialytic ambulatory systolic BP removed the association of BMI with LVMI nearly completely (model 3). Figure 2 shows the association of unadjusted LVMI with BMI (solid line), multivariate adjusted for variables shown in model 2 (dashed line), and further adjusted for interdialytic ambulatory systolic BP (dotted line).
Association of unadjusted left ventricular mass index (LVMI) with body mass index (BMI). Unadjusted estimate is shown by the solid line, multivariate adjusted (for variables shown in model 2, Table 3) by the dashed line. and further adjusted for interdialytic ambulatory systolic blood pressure (BP) by the dotted line. The relationship between BMI and LVMI was not significant after being adjusted for interdialytic ambulatory systolic BP.
Cumulative follow-up for 1122 patient-years of 368 patients with a median follow-up of 2.7 years culminated in 119 deaths (32%). Of 138 normal or underweight patients, 49 (36%) died. Of 110 overweight patients, 34 (31%) died. Of 71 mildly obese patients (BMI: <35), 23 (32%) died. Of 49 moderately or severely obese patients (BMI: ≥35), 13 (27%) died.
Figure 3A shows the survival according to BMI categories. In the first 2 years of follow-up, the mortality hazard for the lowest BMI group was increased (Figure 3B, which shows the enlarged version of the dotted area of Figure 3A); thereafter, the survival curves were similar. These graphs indicate violation of the proportionality assumption; therefore, Table 4 shows the hazard ratio for the unadjusted model without accounting for the proportionality assumption (model 1) and after interacting the predictor with time (model 2). We next added the following variables to the model: age, sex, race, serum albumin, history of diabetes mellitus, history of cardiovascular disease, dialysis vintage, LVMI, and systolic ambulatory BP (model 3). Even adjusting for these variables neither mitigated the strength of the association between BMI and mortality nor removed the statistical significance of this inverse association. Serum creatinine in dialysis patients is a proxy for muscle mass. We added this term to the model to further explore the relationship between BMI and mortality (model 4). Adding serum creatinine also did not explain the increased mortality associated with a lower BMI. However, adjusting for serum creatinine in those with low or normal body mass index removed the association of mortality with BMI (data not shown).
A, Survival by body mass index (BMI) categories. B, Enlarged version of the dotted area of A. During the first 2 years of follow-up, the mortality hazard for the lowest BMI group was increased; thereafter, the survival curves were similar.
Discussion
An inverse relationship between both the prevalence of hypertension and its poor control was associated with BMI. Those who were the leanest had the greatest prevalence of hypertension and also the poorest control. Similarly, an inverse relationship between LVMI and BMI was found. Thus, those who were leaner had a greater LVMI. The excess prevalence of hypertension and poor control of hypertension among lean hemodialysis patients may be because of the following reasons. First, obese patients may sequester excess fluid volume in the extracellular space more effectively than lean people and, therefore, not get hypertensive. Although echocardiographic evidence of volume excess among leaner patients (increased left atrial diameter or inferior vena cava diameter both indexed for body surface area) was found, these markers were by themselves insufficient to obliterate the inverse association between hypertension and BMI. Second, increased muscle mass may be associated with increased renalase expression in those with higher BMI.18 Renalase, a catecholamine-metabolizing enzyme, is expressed in skeletal muscle and can reduce the circulating catecholamine levels.18,19 This, in turn, would be associated with less prevalence of hypertension and better control. Neither muscle mass nor plasma renalase concentration was measured in our patients, so we are unable to confirm or refute this possibility.
The association of increased BMI with lower mortality was first reported in the Diaphane collaborative study in younger French dialysis patients treated with long-term dialysis during the 1970s.2 With one notable exception,20 these observations have now been confirmed in several cohorts.3–8 This study not only confirms the inverse association of BMI with mortality, it also provides some mechanistic insights. For example, adjustment for numerous explanatory variables for mortality, including ambulatory systolic BP and LVMI, did not remove the association of excess mortality and lower BMI. More importantly, the relationship of BMI and mortality was not constant over time. Leaner individuals showed accelerated mortality in the short term. However, after the first 2 years, the mortality curves were similar. Also notable was the lack of graded relationship between BMI and mortality. Thus, having a BMI of <25 (normal or underweight) was associated with increased mortality; however, being overweight or progressively increasing levels of obesity was not associated with increasing mortality. Thus, obesity by itself does not appear to confer a survival advantage; in contrast, being underweight or normal weight on hemodialysis confers an increased risk for mortality. These patients may have a greater burden of illness. In this cohort they were more often smokers and also had a lower serum albumin. The results support the hypothesis proposed by Beddhu.12 These investigators proposed that lean (low muscle mass) people have an accelerated rate of death. People who are obese have a high mortality rate but one that does not reach that of the leaner individuals. Our data provide direct support for this hypothesis. Notably, in a cohort of patients followed for 5 years, the effect of BMI on survival was also found to be time dependent; as noted in our study, the highest risk of death attributable to undernutrition or low BMI was noted to be in the first 2 years.3
There is an additional potential mechanism to explain these observations. The risk of increased mortality in this group of normal or underweight individuals could be because of misclassification of obesity. We have reported previously that the negative predictive value of BMI to detect obesity in chronic kidney disease is only 45%.21 Thus, a normal BMI does not rule out obesity. Although body composition was not measured in this cohort, it is unlikely that hemodialysis patients who are normal or underweight have low body fat. In fact, it is this group of patients who is more likely to have sarcopenia (and, therefore, body fat proportion that may be comparable to those with overt obesity). As an example, among end-stage renal disease patients in Sweden, protein energy wasting was measured by the subjective global assessment of nutrition.22 This condition was equally prevalent in patients with low, normal, and high BMI, lending support to the condition of “obese sarcopenia.” In this cohort, BMI, per se, did not predict mortality. However, for each BMI group, protein-energy malnutrition was associated with increased death risk. Serum creatinine is a surrogate for muscle mass in hemodialysis patients and has been inversely and independently associated with mortality.8 Adjusting for serum creatinine in those with low or normal body mass index removed the association of mortality with BMI. This suggests that low muscle mass (sarcopenia) may confer excess mortality risk among hemodialysis patients. This is further supported by subjective global assessment of nutrition among hemodialysis patients; malnutrition as assessed by subjective global assessment is associated with a remarkably increased early mortality.23
Our study has the following limitations: we did not measure body composition or 24-hour urine creatinine excretion. Assessed by 24-hour urine creatinine excretion, Beddhu et al24 have demonstrated previously that normal or increased muscle mass confers a survival advantage of high BMI among chronic kidney disease patients on long-term dialysis. We also did not measure change in BMI over time, so we are unable to assess the impact of change in BMI on mortality. However, others have shown previously that a drop in BMI over 6 months4 or postdialysis weight loss is associated with an increased risk for mortality.8 Other metrics of increased visceral fat, such as waist circumference, were not measured. However, it has been noted previously that a higher waist circumference even after adjusting for BMI is associated with a higher all-cause and cardiovascular mortality.25 Strengths of our study include the measurement of interdialytic ambulatory BP and echocardiographic LVMI as mediators of the inverse relationship between BMI and mortality.
Perspectives
This study shows that leaner patients on dialysis have a higher prevalence of hypertension, poorer control of hypertension, and greater evidence of extracellular fluid volume excess. However, the latter only partially explains the greater prevalence or poorer control of hypertension. Leaner patients also have evidence of more LVMI mostly because of higher interdialytic ambulatory BP. Leaner patients have an accelerated mortality rate in the first 2 years. Subsequently, the mortality rate among these patients matches ones with higher BMI. The accelerated mortality rate is not completely explained by hypertension, left ventricular hypertrophy, or other cardiovascular or dialysis-specific risk factors. Further research to explain the mechanistic relationship between low BMI and increased mortality is needed.
Sources of Funding
This work was supported by grant 2RO1-NIDDK062030-07 from the National Institutes of Health-National Institute of Diabetes and Digestive and Kidney Diseases.
Disclosures
None.
- Received July 23, 2011.
- Revision received August 14, 2011.
- Accepted October 7, 2011.
- © 2011 American Heart Association, Inc.
References
This Issue
Jump to
Article Tools
Share this Article
- Body Mass Index-Mortality Paradox in HemodialysisPermalink:
