Body Mass Index, Diabetes, Hypertension, and Short-Term Mortality: A Population-Based Observational Study, 2000–2006

Discussion

It is instructive to interpret the study findings in the context of previous broadly representative studies examining the association of BMI and mortality across the adult life spectrum (eg, not limited to the elderly), all of which employed BMI data collected more than a decade ago. In this study, overweight was associated with decreased mortality risk, whereas prior studies were mixed in this regard, with most finding either no association of overweight with mortality or increased risk.^{6⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓–19} Although all but one¹⁶ of the prior studies found milder obesity (BMI 30-<35 kg/m²) was associated with increased mortality risk,^{6⇓⇓⇓⇓⇓⇓⇓⇓–15,17⇓–19} in this study, obesity of this degree was not significantly associated with mortality in a model unadjusted for diabetes and hypertension and was not significantly associated with decreased mortality after adjusting for these conditions. Regarding severe obesity, as in the relatively fewer prior studies examining this category separately,^{15⇓⇓⇓–19} this study found it to be associated with significantly increased mortality risk without adjusting for diabetes or hypertension. However, severe obesity was no longer significantly associated with mortality after adjusting for these conditions, something not examined in the prior studies. Considered in the context of prior studies, these findings suggest that the mortality risk of above-normal BMI, at least in the short term, may be lower in the current era than in the past. This apparent temporal change in the association between BMI and mortality was more marked when we employed a normal BMI reference category of 18.5-<25 kg/m², as used in prior studies. Including persons with a BMI <20 as the normal reference category reduces the apparent mortality risk of above-normal BMI. Although this study did not directly examine secular changes in BMI-associated mortality, a prior US study found evidence for early stages of a temporal decline in BMI-associated mortality beginning in the late 1970s.⁸ The findings of this study are consistent with the continuation of this trend.

The study also helps to clarify the contemporary influence of BMI on mortality risk among individuals with diabetes or hypertension. Regarding diabetes, prior work employing data collected before widespread obesity found a higher mortality risk among diabetics with higher versus lower BMIs.²⁰ However, the mortality risk of diabetes in the current study was lower among individuals with higher versus lower BMIs, further supporting the notion of a secular decline in obesity-related, short-term mortality risk. By contrast, consistent with most prior studies examining this issue (all employing data collected 30–40 years ago),²¹ there was no evidence of a significant interaction between hypertension and BMI, indicating their effects on all-cause mortality are independent of each other at all levels of BMI examined. Increasing surveillance bias influencing the identification of diabetes may partly account for the finding of lower mortality risk among diabetics with higher BMI. In the current era, an increasing proportion of type 2 diabetics are severely obese.²⁸ The diagnosis of relatively mild diabetes may be more likely among such individuals because of heightened clinician suspicion of (and related increased testing for) the disease, and milder cases of diabetes may be associated with relatively low mortality risk. Also relevant is the lowering of the fasting plasma glucose threshold for diagnosing diabetes to ≥126 mg/dL in the 1990s.²² Partly because of that change, the age-adjusted incidence of diabetes has nearly doubled.²⁹ By contrast, the diagnostic threshold defining hypertension (≥140/90 mm Hg) has not changed since the early 1980s.

Strengths of this study included the use of nationally representative BMI data collected within the past decade and determination of BMI-mortality associations both without and with adjustment for hypertension and diabetes. The study also had some limitations. The analyses were observational, so causal associations cannot be inferred. BMI in our study was derived from self-reported height and weight. A number of studies also employed BMI determined from self-reported height and weight data collected before the year 2000 and found BMI-mortality associations similar to those found in contemporaneous studies of measured BMI.^{7,11⇓–13,30} Prior studies suggest a complex relationship between self-reported and objectively measured BMI, with differences in BMI category misclassification resulting from use of self-report data of sociodemographic characteristics (eg, country of residence, sex, race), and BMI category (eg, tendency to underestimate BMI among those with higher BMI vs overestimated BMI among individuals with lower BMI).^{31⇓⇓⇓–35} The net effects of such relationships on the BMI-mortality associations observed in a diverse, nationally representative sample such as ours are uncertain. However, in 2 supplementary analyses employing BMI values derived by applying formulas aimed at adjusting for self-report bias, we noted findings suggesting less risk of mortality associated with higher BMI categories. Thus, it seems unlikely that the relatively attenuated associations between BMI and mortality we observed compared with earlier studies simply reflect biased self-reporting of height and weight.

Diabetes and hypertension also were self-reported in our study. Although studies suggest that self-report of these conditions is reasonably accurate,^36,37 misreporting leading to respondent misclassification could have obscured true BMI-mortality relationships, especially if misreporting varied by BMI category. We also were unable to distinguish type 1 from type 2 diabetes in our data set. However, given the epidemiology of the conditions, it is likely that type 1 diabetics were relatively over-represented in the lower BMI categories. At first glance, this might seem to help explain the apparent protective effect of higher BMI on mortality that we observed in diabetics; people with type 1 diabetes generally may have had the diagnosis longer than those with type 2 diabetes and so might be posited to have developed more risk factors for premature mortality (eg, renal impairment manifested by microalbuminuria). However, at least 3 lines of evidence suggest this is unlikely. First, type 1 diabetes only accounts for an estimated 5% to 10% of all diagnosed cases of diabetes in the United States,³⁸ so our mortality findings likely were driven primarily by the association of BMI with mortality among type 2 diabetics. Second, some work suggests that the prevalence of risk factors for death caused by cardiovascular disease seems to be higher in type 2 versus type 1 diabetics with similar durations of diagnosis.³⁹ Third, 5-year adjusted mortality risk seems to be 3- to 4-fold higher for type 2 versus type 1 diabetics.⁴⁰ In other words, the true protective effect of overweight and nonsevere obesity on mortality in type 2 diabetes may be somewhat larger than the effect observed here. Nonetheless, further studies employing recently collected, objectively verified BMI and health condition data are needed to verify our findings.

The study also employed a relatively short follow-up period (up to 6 years) for mortality ascertainment, with only about one third of respondents having more than 5 years of follow-up. An increased mortality risk of obesity could require longer exposure to emerge. A prior review suggested at least 5 years' follow-up was needed to demonstrate the effects of obesity on mortality.⁴¹ However, the recommendation was derived largely from studies of small local or regional cohorts, with variation among studies in the duration of obesity before study enrollment. Furthermore, some large national studies of data collected before the era of prevalent obesity employed follow-up periods as short as or shorter than ours^7,11 and nonetheless observed significant effects of obesity on mortality. MEPS nonresponse also may have produced some bias, with uncertain applicability of our findings to nonresponders. Whether the current findings are applicable to countries and territories other than the United States requires further study.

Given the substantial burden of morbidity entailed by overweight and obesity,^2,3 the study findings should not be construed to mean that overweight and obesity no longer represent major threats to individual and public health. Rather, the findings may prove useful in guiding research aimed at developing effective clinical and population-level efforts to address overweight and obesity. For example, they suggest that efforts to reduce mortality among the overweight and obese might be targeted most productively at individuals with coexisting diabetes, hypertension, or both. The findings also raise the question of whether efforts to reduce mortality among diabetics should be targeted primarily to commonly co-occurring health risks associated with increased mortality risk, such as physical inactivity and smoking,^42,43 rather than weight. Given the observational nature of the current analyses, these notions represent hypotheses to be tested in randomized controlled trials.