ELSEVlER Relative Weight and Mortality in U.S. Blacks and Whites: Findings from Representative National Population Samples RAMdN DURAZO-ARVIZU, PIID, RICHARD S. COOPER, MD, AMY LUKE, PHL?, T. ELAINE PREWITT, DRPH, YOULIAN LIAO, MD, AND DANIEL L. MCGEE, Pr-rLl PURPOSE: To examine the impact of relative weight on mortality in black and white men and worncn. METHODS: Two representative national populations samples were used: the NHANES-I Epidemto- logic Follow-up Study (NHEFS), and the National Health Interview Survey (NHIS). The principal analysis focused on 13,242 participants in the NHEFS and 114,954 in the NHIS. Minimum mortality was estimated from both categorical analysis and a logistic model. RESULTS: Minimum mortality ranged from a body mass index (BMI) of 25 to 32 kg/m?. The m&cl- estimated BMI of minimum mortality for NHEFS was 27.1 (24.8-29.4, 95% CI), 26.8 (24.7-28.9, 95% CI), 24.8 (23.8-25.9, 95% Cl) and 24.3 (23.2-25.4,95% CI); for black men, black women, white men and white women, respectively, whereas for NWiS the corresponding values were 30.2 (24.&-35.6, %% CI) 26.4 (24.2-28.7, 95% CI), 27.1 (25.5-28.7, 95% CI), and 25.6 (24.2-27.0, 95% CI). In all groups the shape of the relative risk curve was virtually identical and a broad range of BMI values in the middle of the distribution was associated with low relative mortality risk. Averaging the results tram both surveys, the observed BMI of minimum risk was 3.1 kg/m’ higher in black men and 1.5 kg/m., higher in black women than in their white counterparts; when adjusted for covariates these differences were only of borderline statistical significance, however. CONCXUSIONS: Because of the wide range of RMI values associated with low risk, and the consisteni:v of the point of the up-turn in risk, group specifc definitions of optimal values do not app~:~a: to be warranted. Ann E~~~~~o~ f997;7:383-395. 0 1997 Elsevier Science Inc. KEY WORDS: obesity, Mortality, Race. INTRODUCTION Relative weight has long been recognized to make an impor- tant contribution to mortatity risk (l-20). In the great majority of previous studies an increase in risk among both the lean and the overweight has been observed, with a broad range of values compatible with minimum risk across the center of the population distribution (1, 10, 15, 21-28). Relatively little information is available on racial groups other rhan whites, however. in particular, inconsistent find- ings have emerged from the studies carried out among U.S. blacks--either the risk relationship has been found to be relatively flat, or comparable white samples were not avaii- able to permit direct comparisons (24, 29-31). Increased Fn~m rtw Ikparnnrnt of Prevenrive Medicine and ~piderni~l~~, Loyola University Switch School of Medicine, Maywood, IL 60153. Address reprint requesrs to: Dr. McGee, Department of Preventive Medicine and Epidemiology, Loyola Stritch School of Medicine, 2160 S. First Avenue, Maywood IL 60153. Dr. Durazo-Arvizu’s current address is Mayo Clinic, ikpartment of Health Sciences Research. Section of Biosca- tlsrics, Plummer 4. 200 First Street SW, Rochester, MN 55905. Received Ocrtther 16, 1996; revised April 24, 1997; accepted May 1. l’W7. mortality risk among the obese can be ascribed primarily to cardiovascular disease, diabetes, and possibly scme cancers (28). The reasons for the upturn in mortality at the lower end of the distribution remain much less wet1 understood. Given these uncertainties, the Location crt‘values associated with the best overall health outcomes is a subjecr of contin- ued debate (l-3). Clearly data on various racial sub-poputa- tions are needed for this debate to be fully informative. While randomized trials on the benet;ts of weight loss would provide the most valid evidence of the risk associated with obesity (32), they would be excee&ngly difficult to carry out. Prospecrive epidemiologic studies therefore re- main the sole source of information un this question, Severat requirements f(lr these cohort studies milke it Jifficui~ LC) obtain clear-cut answers. First, the cohorts must bc very large, given the interest in events in thy extremes of the distribution. Second, to provide the most secure inferences the samples should be representative of the general pop&- tion. This is particularly true since the porential role of confounding is at the center of this cimtu.jversg, and differ- ent populations may vary in the frcquencv and distribution of confounding factors. Third, potential c.(~~~founding factors must be characterized, and their impact e~rimat-ed. Finally, statisrical methods need to be avaIlable &ich promote, as