Impaired Fasting Glucose and Body Mass Index as Determinants of Mortality in ALLHAT: Is the Obesity Paradox Real?

Abstract

Emerging literature suggests that obesity may be “protective” against mortality and cardiovascular outcomes, while dysglycemia may worsen outcomes regardless of obesity. The authors measured the association of weight, smoking, and glycemia with mortality in the Antihypertensive and Lipid‐Lowering Treatment to Prevent Heart Attack Trial (ALLHAT). Among 5423 ALLHAT participants without established diabetes or cardiovascular disease, 3980 (73%) had normal fasting glucose and 1443 (27%) had impaired fasting glucose (IFG) levels at study entry. After a median of 4.9 years follow‐up, 554 (10%) had died (37% cardiovascular). IFG was associated with higher all‐cause mortality (adjusted hazard ratio [HR], 1.23; 95% confidence interval [CI], 1.02–1.50), while obesity was associated with lower all‐cause mortality (adjusted HR, 0.76; 95% CI, 0.60–0.96). However, after excluding underweight individuals (body mass index [BMI] <22 kg/m²) and smokers, neither obesity nor IFG was associated with all‐cause mortality, but IFG identifies individuals at greater risk in the nonobese population. Although obesity appeared protective against mortality, this association was not significant in never‐smokers or after exclusion of BMI <22 kg/m². The obesity paradox may result from confounding by a sicker, underweight referent population and smoking.

A growing percentage of the adult population in Europe and the United States is overweight or obese,1 making the effects of excess adiposity on health a matter of public health concern. Consistently, excess weight has been associated with incident cardiovascular (CV) disease2, 3, 4 and diabetes,5 prompting medical and public health societies to counsel maintenance of optimal weight.6 To date, the measure most commonly used for measuring adiposity is the body mass index (BMI), for which a dose‐dependent relationship with mortality has been previously demonstrated.7, 8 Incongruously, a recent large study suggests that people with an elevated BMI have better survival than normal weight individuals,9 an observation that has been noted in various other studies.10, 11 Two hypotheses have been put forward to explain this apparent obesity paradox. First is the supposition that BMI itself (as an aggregate of lean and fat mass) is inadequate to estimate excess adiposity,12, 13 and that alternative measures (such as waist‐to‐hip ratio14 or visceral adiposity15) may be more representative of clinical risk. A second explanation is that reference ranges used to define “normal” BMI allow for residual confounding by chronic conditions, in which inclusion of leaner individuals who may have chronic medical conditions associated with lower BMI (smoking, frailty, or other12, 13) artificially dilutes the risks of an elevated BMI.

Amidst this debate, there is increasing evidence that dysglycemia (impaired fasting glucose [IFG] defined as fasting glucose of 100–125 mg/dL) may be more important than BMI in defining CV risk.16, 17, 18, 19 Inflammation, insulin resistance, and oxidative stress may be common underlying factors for the constellation of clinical abnormalities termed the metabolic syndrome, with glucose intolerance as a surrogate for the pathophysiologic processes.20, 21 Emerging data support the hypothesis that metabolic health status may stratify prognosis in individuals across BMI, identifying individuals at highest risk for CV events and poorer survival. A link between dysglycemia and mortality may thus provide an explanation for the obesity paradox.

To address the role of BMI on mortality and impact of dysglycemia, we investigated 5423 participants in the Antihypertensive and Lipid‐Lowering Treatment to Prevent Heart Attack Trial (ALLHAT) without diabetes or manifest CV disease at study entry. We measured the univariable‐ and multivariable‐adjusted association of BMI classes and IFG with all‐cause and CV mortality to test the hypothesis that the obesity paradox may be explained by excess risk in underweight persons, and IFG (but not BMI) would be associated with poorer prognosis, thereby separately assessing frailty and chronic illness to improve our current understanding of BMI and the obesity paradox.

Methods

Study Participants

The rationale, design, and main results of the ALLHAT study have been published.22 Briefly, ALLHAT enrolled 42,418 participants aged 55 years and older with hypertension and at least one additional coronary heart disease risk factor, including previous myocardial infarction or stroke, left ventricular hypertrophy, type 2 diabetes mellitus (T2DM), ongoing cigarette smoking, high‐density lipoprotein (HDL) concentration <35 mg/dL, and/or documentation of any other atherosclerotic CV disease. Participants with clinically overt heart failure and/or known left ventricular ejection fraction <35% were excluded, as well as those with comorbid illnesses likely to lead to non‐CV death. All study participants gave written informed consent, and all centers obtained institutional review board approval. Study participants underwent double‐blinded randomization to lisinopril, amlodipine, doxazosin, or chlorthalidone (active control), and were followed for CV and other major clinical events and mortality, including all‐cause and CV mortality. Mortality was ascertained by clinical report and by assessment from the National Death Index and the Social Security Administration databases in conjunction with a death certificate. T2DM was defined as a clinical diagnosis or fasting blood glucose ≥126 mg/dL on a single visit determination.

For the purposes of this study, we restricted our analyses to participants without baseline known T2DM and without prior evidence of atherosclerotic CV disease (defined as myocardial infarction, peripheral arterial disease, or stroke) to reduce the risk for incomplete adjustment for CV disease in multivariable models. In addition, we excluded individuals randomized to doxazosin, given their disproportionate follow‐up as a result of early discontinuation of this arm. We defined an overweight BMI as between 25 kg/m² and 30 kg/m² and considered a BMI >30 kg/m² obese. Additionally, in light of the controversy regarding the protective association of obesity/overweight and whether this relates to an overly broad “normal” BMI (18.5–25 kg/m²) category, including individuals with comorbid illness, we constructed a fully adjusted multivariable model incorporating IFG, BMI, and clinical risk factors, excluding individuals with a BMI <22 kg/m², a priori defined based on the J‐shaped relationship between BMI and mortality reported in multiple other large cohorts.23 The “normal” referent category for this secondary analysis was thus defined as a BMI 22 kg/m² to 25 kg/m². We also conducted an analysis in a never‐smoker cohort.

Statistical Analysis

Data are summarized as means and standard deviations for continuous variables and number of patients and percentage for categorical variables. The Cox proportional hazards model was used to determine hazard ratios and 95% confidence intervals (CIs). Heterogeneity of IFG effects on outcomes across BMI strata were examined by testing for BMI‐IFG interaction in Cox models using a P value <.05. Interaction terms that were not statistically significant were removed from the fully adjusted models. If the interaction term was statistically significant, linear combinations of the Cox model were used to ascertain coefficients and statistical significance in the subgroups. Subgroup and interaction analyses were performed without adjustment for multiple hypothesis testing. All statistical analyses were carried out using STATA version 12.0 (StataCorp, College Station, TX).

Results

Baseline Characteristics

The derivation of 5423 participants comprising the study sample is shown in Figure 1. After a median follow‐up of 4.95 years (interquartile range, 4.32–5.88 years), a total of 554 participants (10%) had died, of which 204 (36.8%) were caused by CV causes. The study population had a balanced sex (55% male) and ethnic distribution (45% white non‐Hispanic, 32% black non‐Hispanic, and 17% Hispanic) and a mean age of 65 years (Table 1). There were very few Asian participants. A total of 3980 (73%) of the participants had normal fasting glucose level and a total of 1443 (27%) participants had IFG at the start of the trial.

Figure 1.

CONSORT diagram depicting inclusion and exclusion criteria for the composition of the study cohort. ALLHAT indicates Antihypertensive and Lipid‐Lowering Treatment to Prevent Heart Attack Trial; BMI, body mass index; IFG, impaired fasting glucose; T2DM, type 2 diabetes mellitus.

Table 1.

Baseline Characteristics of the Study Population

	Overall Group (N=5423)
Age, mean (SD), y	65.18 (7.31)
Male, %	55.26
Race, %
White	58.58
Black	35.64
American Indian/Alaskan Native	0.18
Asian/Pacific Islander	1.16
Other	4.43
Ethnicity, %
White non‐Hispanic	45.05
Black non‐Hispanic	31.68
White Hispanic	13.53
Black Hispanic	3.96
Other	5.77
Education, mean (SD), y	11.21 (4.01)
Taking antihypertensive treatment prior to enrollment, %	87.11
Treatment group, %
Chlorthalidone	45.79
Amlodipine	26.46
Lisinopril	27.75
SBP, mean (SD)	146.07 (15.38)
DBP, mean (SD)	85.87 (9.73)
SBP (treated prior to enrollment), mean (SD)	144.58 (15.20)
DBP (treated prior to enrollment), mean (SD)	85.07 (9.59)
SBP (untreated prior to enrollment), mean (SD)	156.17 (12.57)
DBP (untreated prior to enrollment), mean (SD)	91.28 (8.95)
Total cholesterol, mean (SD)	216.39 (41.62)
HDL, mean (SD)	47.92 (15.78)
LDL, mean (SD)	137.71 (36.03)
Triglycerides, mean (SD)	161.21 (117.71)
Glucose, mean (SD)	93.24 (11.66)
In lipid‐lowering trial, %	32.64
Ever smoked, %	72.82
LVH on ECG, %	33.78
Estimated GFR, mean (SD)	78.48 (17.84)
Taking aspirin, %	23.75
Taking lipid‐lowering therapy, %	9.53
In lipid‐lowering arm and taking pravastatin therapy, %	16.49
BMI, mean (SD)	29.19 (5.91)
BMI by category, No. (%)
18.5≤BMI<22.0	356 (6.56)
22.0≤BMI<25.0	906 (16.71)
25.0≤BMI<30.0	2149 (39.63)
30.0≤BMI<35.0	1252 (23.09)
BMI ≥35	760 (14.01)

Abbreviations: BMI, body mass index; DBP, diastolic blood pressure; HDL, high‐density lipoprotein; LDL, low‐density lipoprotein; LVH, left ventricular hypertrophy; SBP, systolic blood pressure; SD, standard deviation. Electrocardiographic (ECG) LVH by Minnesota code 3 to 1 or 3 to 3 with ST/T‐wave changes (Minnesota codes 4 and 5) as determined by the central ECG laboratory.

Association Between Weight Class and All‐Cause Mortality

Regarding weight class and all‐cause mortality in a univariable model, obese (hazard ratio [HR], 0.56, 95% CI, 0.45–0.69; P<.001) and overweight status (HR, 0.80; 95% CI, 0.65–0.97; P=.024) were associated with lower risk of all‐cause mortality (Table 2, Figure 2A). After multivariable adjustment (for baseline IFG, overweight BMI, obese BMI, age, sex, race, smoking status, systolic and diastolic blood pressure, low‐density lipoprotein [LDL] and HDL levels, fasting triglycerides, treatment group, left ventricular hypertrophy on electrocardiography, and lipid, aspirin, or hypertensive medication use at baseline), obesity remained statistically significant (model 3: HR, 0.76; 95% CI, 0.60–0.96; P=.024) while overweight did not.

Table 2.

Risk Factors of All‐Cause Mortality

Baseline Characteristics	Univariable Cox Regression (n=5412)		Multivariable Cox Regression Modelc (n=5067)		Multivariable Cox Regressiond Model Excluding BMI <22 (n=4720)		Multivariable Cox Regressiond Model Excluding Smokers (n=1361)		Multivariable Cox Regressione Model Excluding BMI <22 and Smokers (n=1307)
	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value
Overweight vs normal weight	0.80 (0.65–0.97)	.024	0.91 (0.73–1.12)	.369	0.96 (0.76–1.23)	.770	1.40 (0.77–2.54)	.275	1.85 (0.90–3.82)	.094
Obese vs normal weight	0.56 (0.45–0.69)	<.001	0.76 (0.60–0.96)	.024	0.81 (0.62–1.06)	.123	1.44 (0.78–2.68)	.246	1.89 (0.90–3.98)	.092
Impaired fasting glucose	1.23 (1.03–1.47)	.025	1.23 (1.02–1.50)	.033	1.18 (0.96–1.44)	.113	1.44 (0.93–2.24)	.103	1.40 (0.88–2.22)	.152
Age	1.07 (1.06–1.08)	<.001	1.07 (1.06–1.09)	<.001	1.07 (1.06–1.09)	<.001	1.10 (1.07–1.13)	<.001	1.10 (1.07–1.13)	<.001
Male	1.45 (1.21–1.72)	<.001	1.32 (1.08–1.61)	.006	1.41 (1.14–1.74)	.002	1.81 (1.15–2.85)	.010	1.80 (1.12–2.90)	.015
Ever smoked	1.65 (1.34–2.03)	<.001	2.02 (1.59–2.56)	<.001	2.02 (1.57–2.58)	<.001	–	–	–	–
Black	1.34 (1.13–1.58)	.001	1.30 (1.07–1.57)	.007	1.28 (1.04–1.56)	.020	1.01 (0.64–1.59)	.983	1.13 (0.70–1.81)	.625
Aspirin use	1.01 (0.83–1.22)	.954	0.87 (0.71–1.08)	.208	0.87 (0.69–1.08)	.209	0.73 (0.43–1.24)	.244	0.74 (0.43–1.27)	.269
Taking antihypertensive medication prior to enrollment	1.01 (0.78–1.31)	.928	0.97 (0.74–1.27)	.832	1.07 (0.79–1.45)	.648	1.25 (0.59–2.67)	.563	1.08 (0.50–2.33)	.840
Amlodipine vs chlorthalidonea	1.01 (0.82–1.23)	.961	1.01 (0.82–1.25)	.904	1.01 (0.80–1.27)	.927	0.82 (0.49–1.38)	.463	0.89 (0.52–1.52)	.662
Lisinopril vs chlorthalidoneb	1.11 (0.91–1.35)	.302	1.19 (0.97–1.46)	.098	1.25 (1.01–1.55)	.042	1.01 (0.63–1.63)	.969	1.02 (0.62–1.68)	.949
Taking lipid‐lowering therapy	0.67 (0.48–0.93)	.018	0.80 (0.57–1.13)	.207	0.76 (0.53–1.09)	.137	1.04 (0.51–2.12)	.908	0.97 (0.46–2.06)	.946
LVH on ECG	1.20 (1.01–1.42)	.036	0.99 (0.82–1.20)	.938	1.03 (0.84–1.26)	.810	1.41 (0.90–2.21)	.135	1.42 (0.89–2.27)	.145
SBP, mm Hg	1.00 (1.00–1.01)	.090	1.00 (1.00–1.01)	.413	1.00 (1.00–1.01)	.311	1.00 (0.99–1.02)	.605	1.00 (0.99–1.02)	.887
DBP, mm Hg	0.99 (0.98–0.99)	.001	1.00 (0.99–1.01)	.350	1.00 (0.98–1.01)	.418	1.00 (0.98–1.02)	.917	1.00 (0.98–1.02)	.935
LDL, mg/dL	1.00 (1.00–1.00)	.079	1.00 (1.00–1.00)	.644	1.00 (1.00–1.00)	.731	1.00 (1.00–1.01)	.321	1.00 (1.00–1.01)	.148
HDL, mg/dL	1.01 (1.00–1.01)	.009	1.00 (1.00–1.01)	.200	1.00 (0.99–1.01)	.609	1.00 (0.99–1.02)	.824	1.00 (0.98–1.01)	.600
Fasting triglycerides, mg/dL	1.00 (1.00–1.00)	.004	1.00 (1.00–1.00)	.640	1.00 (1.00–1.00)	.673	1.00 (1.00–1.00)	.623	1.00 (1.00–1.00)	.588

Abbreviation: CI, confidence interval; DBP, diastolic blood pressure; HDL, high‐density lipoprotein; HR, hazard ratio; LDL, low‐density lipoprotein; LVH, left ventricular hypertrophy; SBP, systolic blood pressure. Electrocardiographic (ECG) LVH by Minnesota code 3 to 1 or 3 to 3 with ST/T‐wave changes (Minnesota codes 4 and 5) as determined by the central ECG laboratory.

^aAmlodipine vs chlorthalidone treatment groups.

^bLisinopril vs chlorthalidone treatment groups.

^cAdjusted for the following baseline characteristics excluding the covariate of interest: impaired fasting glucose, overweight BMI, obese BMI, interaction between overweight BMI and impaired fasting glucose, and interaction between obese BMI and impaired fasting glucose.

^dAdjusted for the following baseline characteristics excluding the covariate of interest: impaired fasting glucose, overweight BMI, obese BMI, age, sex, race, smoking status, SBP, DBP, LDL, HDL, fasting triglycerides, treatment group, lipid medication use at baseline, aspirin use at baseline, antihypertensive medication use prior to enrollment, and LVH on ECG.

^eAdjusted for the following baseline characteristics excluding the covariate of interest: impaired fasting glucose, overweight BMI, obese BMI, age, sex, race, SBP, DBP, LDL, HDL, fasting triglycerides, treatment group, lipid medication use at baseline, aspirin use at baseline, antihypertensive medication use prior to enrollment, and LVH on ECG.

Figure 2.

Kaplan‐Meier estimates for all‐cause mortality stratified by the presence and absence of impaired fasting glucose (IFG) across body mass index (BMI) (A). Kaplan‐Meier estimates for cardiovascular mortality stratified by the presence and absence of IFG across BMI (B). BMI <25 kg/m² refers to 18.5 kg/m² to 25 kg/m².

Since smokers and underweight individuals may confound obesity‐related “protection,”12 we investigated the associations of BMI with mortality excluding ever‐smokers and using a higher normative BMI. After excluding participants who ever smoked, neither overweight nor obesity was significantly associated with all‐cause mortality (Table 2). To investigate whether the use of a different referent standard (22–25 kg/m²) for “normal” BMI would eliminate confounding by unrecognized illness in underweight individuals, we excluded individuals with a BMI <22 kg/m². In the remaining 4720 participants, neither overweight nor obesity was significantly associated with all‐cause mortality (Table 2).

Impact of Dysglycemia on All‐Cause Mortality

In addition to factors traditionally associated with increased mortality risk (age, male sex, smoking, black race, left ventricular hypertrophy), IFG (HR, 1.23; 95% CI, 1.03–1.47; P=.025) was associated with higher risk of all‐cause mortality in a univariable model (Table 2, Figure 2A). In fully adjusted models, IFG was associated with mortality independent of BMI (Table 2; model 3: HR, 1.23; 95% CI, 1.02–1.50; P=.033).

Given that patients with IFG have a higher risk of developing T2DM at follow‐up,16 we also evaluated the impact of IFG on mortality by obesity class. Significant associations between IFG and all‐cause mortality and CV mortality were found for patients with BMI <25 kg/m² (Figure 2A, B). We also included development of T2DM as a time‐dependent covariate in multivariable models for all‐cause mortality. After full adjustment, incident T2DM during follow‐up was not associated with mortality within the timeframe of the study. Finally, we performed log‐rank tests to compare all‐cause mortality between patients randomized to chlorthalidone vs other randomized therapy within ALLHAT stratified by follow‐up dysglycemia status (eg, normal fasting glucose staying normal throughout study, transitioning to IFG at any point, or transitioning to T2DM at any point). There was no significant difference between chlorthalidone vs other therapies with regard to all‐cause mortality in any of these subgroups.

Association Between Weight Class, IFG, and CV Mortality

In univariable Cox regression models, age, black race, history of smoking, fasting triglycerides, and electrocardiographic left ventricular hypertrophy were associated with CV death, whereas baseline LDL, HDL, blood pressure, IFG, and BMI status were not (Table S1). After multivariable adjustment, age (HR, 1.08; 95% CI, 1.06–1.10; P<.001), left ventricular hypertrophy (HR, 1.43; 95% CI, 1.05–1.94; P=.022), diastolic BP (HR, 1.02; 95% CI, 1.00–1.04; P=.036), and a history of smoking (HR, 2.32; 95% CI, 1.57–3.41; P<.001) were associated with CV death. While IFG was associated with CV death in normal weight individuals in unadjusted analysis (Figure 2B), neither weight class nor the presence of IFG was associated with CV death after full multivariable adjustment (Table S1).

In addition, we investigated the associations of BMI and IFG with CV mortality excluding smokers and using a higher normative BMI to address smoking and underweight‐related confounding. Neither BMI nor IFG were associated with CV mortality in individuals with a BMI >22 kg/m². Separately, when ever‐smokers were excluded, we found a significant interaction between obesity and IFG on CV mortality (Table S2). Specifically, in normoglycemic but not dysglycemic individuals who never smoked, obesity trended toward an association of a greater risk of CV death than normal weight (HR, 3.48; P=.06), inconsistent with an obesity paradox. On the other hand, in normal weight nonsmokers (but not overweight nor obese) individuals, IFG was associated with a higher hazard of CV death than normoglycemia (HR, 5.84; P=.033).

Discussion

In this study of nondiabetic hypertensive individuals without clinical CV disease, we found that overweight (BMI 25–30 kg/m²) and obesity (BMI >30 kg/m²) were associated with lower all‐cause mortality in unadjusted analyses. However, only obesity remained associated with lower mortality after adjustment. When we restricted our analyses to participants with BMI >22 kg/m², the lower all‐cause mortality risk associated with obesity was no longer significant. This latter finding suggests that the protective effect of obesity may be due to a comparison that includes a more chronically ill subgroup within the referent population. Finally, when we excluded smokers from the cohort, we again found no protective association of obesity on either all‐cause mortality or CV mortality. Collectively, these results suggest that inclusion of individuals who are underweight and/or who have ever smoked may at least partly explain the “obesity paradox” in individuals with hypertension.

In a recent meta‐analysis of 2.88 million adults, Flegal and colleagues9 reported that an overweight BMI conferred a 6% lower hazard of all‐cause mortality, with no increased risk for grade 1 obesity (BMI 30–35 kg/m²).9 Likewise, in a study of 2625 individuals with T2DM collected from community‐based cohorts in the United States, Carnethon and colleagues24 reported a paradoxically higher mortality rate in normal weight individuals with T2DM relative to overweight/obese individuals with T2DM.24 To address these findings, some have proposed that markers of metabolic risk—in addition to BMI—must be included when identifying those at highest risk for mortality.25, 26 Our data support this hypothesis. We demonstrate an association between IFG and increased all‐cause mortality both in unadjusted models and independent of BMI after adjustment for other risk factors, especially in patients with BMI <25 kg/m², potentially identifying those with IFG as “metabolically unwell” lean individuals. Other investigators have suggested that residual confounding—the notion that lower BMI and smoking may track with comorbid conditions that modify prognosis—explains the paradoxical findings.12 In addition, the use of 22 to 25 kg/m² as a reference range (as opposed to the World Health Organization's “normal” 18.5–25 kg/m² BMI range) stems from published data from large cohorts suggesting a J‐shaped association between BMI and mortality, with the lowest mortality in 22 kg/m² to 25 kg/m².23 Our data support the hypothesis of confounding by inclusion of sicker individuals within the normative range as the protective association of overweight/obesity with mortality is not maintained (1) when the normative bounds of BMI are increased to eliminate participants with an underweight BMI possibly indicative of chronic illness and (2) when patients with any history of smoking were excluded.

Our analysis additionally suggests that statistical adjustment for smoking in multivariable regression may not be sufficient and that exclusion of ever‐smokers may be necessary to fully capture smoking‐related risk: the obesity‐mediated protection against all‐cause mortality is not present after exclusion of smokers, as opposed to adjustment. This phenomenon has been suggested by other authors.12 Ultimately, these observations lend credence to the hypothesis that residual confounding by both a lower “normative” BMI category currently accepted by the World Health Organization (18.5–25 kg/m²) and inclusion of smokers may drive obesity‐related protection. In addition, independent of the influence of these confounders, our results suggest that IFG—an easily assessed point estimate of prediabetes—influences outcome independent of BMI and may be a fruitful, earlier target across all BMI categories for intervention to improve outcome.

In an analogous recent report from the Avoiding Cardiovascular Events Through Combination Therapy in Patients Living With Systolic Hypertension (ACCOMPLISH) study (a prospective, blinded study to compare antihypertensive therapy), Weber and colleagues27 investigated major adverse CV events by different antihypertensive therapy in different BMI categories (with normal weight <25 kg/m²). While the primary analysis was aimed at investigating effects of different antihypertensive therapy on CV events by BMI category, the authors also reported that compared with normal weight patients, obese and overweight individuals did not have a higher risk of events. This observation (that patients with lower BMI may be at higher risk) is not new in hypertension studies.28 In contrast, our analysis found no such protection for all‐cause mortality or CV mortality in overweight individuals after adjustment. These discrepant findings may be due to our exclusion of participants with baseline T2DM and/or CV disease, as well as our adjustment for baseline dysglycemia, ever‐smoking, blood pressure, and lipid profiles. In particular, smoking—a well‐known correlate of lower BMI—was not included in the report from ACCOMPLISH, and smoking represented a significant all‐cause and CV mortality risk in our study, with a near 2‐fold higher hazard of death.

Study Limitations

The findings of our study need to be viewed in the context of its design. First, the cohort used for these analyses was exclusively hypertensive. While this may limit the generalizability of our findings, excess weight, IFG, and hypertension are common comorbidities in individuals with CV disease. ALLHAT excluded individuals with a high likelihood of non‐CV death (eg, known malignancy), unlike other investigations.9 While we did impose restrictive inclusion criteria from the parent ALLHAT cohort, the study population included only individuals without established CV or diabetes at baseline to provide a clearer description of associations among BMI, IFG, and outcome. A single measure of fasting glucose above threshold was used to diagnose diabetes so it is possible some cases would not have been diagnosed on repeat testing. While our findings may be limited by the lack of balanced stratification by obesity grade and resulting limited power for multivariable adjustment in each stratum, we were nevertheless able to provide insights on conflicting data regarding overweight protection observed in prior work.29, 30 Finally, the use of a reference range of 22 kg/m² to 25 kg/m² led to a decreased sample size, and may have led to loss of statistical power to examine the association between this BMI range and mortality. However, the exclusion of smokers resulted in an HR for obese vs normal BMI that had higher risk of mortality (HR, 1.44 when smokers were excluded; HR, 1.89 when smokers and BMI <22 kg/m² were excluded) compared with when smokers were included (HR, 0.76 with no exclusions). Thus, while lack of statistical power may explain the results, HR point estimates progressively change from protective to increased risk and the removal of sicker individuals (lower BMI and/or ever‐smokers) may also add to the explanation.

Conclusions

We did not find a protective association of an overweight or obese BMI with all‐cause mortality or CV events after exclusion of underweight individuals or those who ever smoked. This is contrary to studies suggesting that overweight is associated with improved survival. Additionally, IFG may identify metabolically unfit lean individuals at increased risk of all‐cause and CV mortality. Ultimately, the observation that a higher BMI may be “protective” against mortality may be confounded by comparisons with underweight individuals and smokers, an observation critical to future epidemiologic research in obesity and CV risk.

Disclosures

This study was supported by contracts NO1‐HC‐35130 and HHSN268201100036C with the National Heart, Lung, and Blood Institute and an American Heart Association Post‐Doctoral Research Award to Dr Shah. The ALLHAT investigators acknowledge study medications contributed by Pfizer, Inc (amlodipine and doxazosin), AstraZeneca (atenolol and lisinopril), and Bristol‐Myers Squibb (pravastatin) and financial support provided by Pfizer, Inc.

Supporting information

Table S1. Risk factors for cardiovascular death.

Table S2. Risk factors for cardiovascular death in Antihypertensive and Lipid‐Lowering Treatment to Prevent Heart Attack Trial (ALLHAT) participants who have never smoked.

J Clin Hypertens (Greenwich). 2014;16:451–458. DOI: 10.1111/jch.12325. ©2014 Wiley Periodicals, Inc.