The Effect of Age on the Shape of the BMI-Mortality Relation And
Total Page:16
File Type:pdf, Size:1020Kb
International Journal of Obesity (2015) 39, 530–534 © 2015 Macmillan Publishers Limited All rights reserved 0307-0565/15 www.nature.com/ijo ORIGINAL ARTICLE The effect of age on the shape of the BMI–mortality relation and BMI associated with minimum all-cause mortality in a large Austrian cohort RS Peter1,2, B Mayer2, H Concin1 and G Nagel1,2 BACKGROUND: It is unclear if the body mass index (BMI) associated with minimum all-cause mortality is constant throughout adult life or increasing with age. METHODS: We applied multivariable fractional polynomials to the data of the Vorarlberg Health Monitoring and Prevention Program to quantify the BMI associated with minimum mortality over age. The analysis included data of 129 904 never-smoking women and men (mean age: 45.4 years) who were followed for a median of 18.6 years. RESULTS: Optimum BMI in women increased with age, lying within the normal BMI category (according to the World Health Organization definition) from the age of 20 years (23.3 kg m−2, 95% confidence interval (CI): 22.2–24.3) to the age of 54 years and in the lower half of the overweight category from the age of 55 years onwards, reaching 26.2 kg m−2 (95% CI: 25.1–27.3) at the age of 69 years. In men, optimum BMI increased slightly from 23.7 kg m−2 (95% CI: 22.1–25.2) at the age of 20 years until the age of 59 years, reaching a BMI of 25.4 kg m−2 (95% CI: 24.8–26.0) and decreased afterwards to 22.7 kg m−2 (95% CI: 20.9–24.6) at the age of 80 years. CONCLUSIONS: Our results indicate that BMI associated with minimum all-cause mortality changes with age and that patterns differ by sex. Sex- and age-independent BMI recommendations might therefore be inappropriate. Further studies using flexible methods instead of predefined categories are necessary to revise BMI recommendations. International Journal of Obesity (2015) 39, 530–534; doi:10.1038/ijo.2014.168 INTRODUCTION category with the lowest mortality risk. Also, the number and The relationship between body mass index (BMI) and all-cause definition of categories differs between studies from World Health 4 1 mortality is known to be U-shaped. Mortality is elevated in the Organization categories of BMI, equally spaced categories, to the 7,8 lowest BMI categories as well as in higher BMI categories.1–3 Most use of quintiles, limiting the comparability of results. Some studies reported the BMI associated with minimum mortality to lie authors assumed a specific functional form (e.g. quadratic) to either in the normal range (BMI: 18.5–25 kg m−2)2 or in the assess the BMI associated with minimum mortality.13,14 However, overweight range (BMI: 25–30 kg m−2).4,5 Differences in the the relationship of BMI and mortality is usually not found to be optimum BMI between studies might be related to differences completely symmetrical and the optimum of a fitted quadratic in the participants’ baseline health status and age. Heiat et al.6 did curve therefore does not match the 'real' optimum. not find any evidence for increased all-cause mortality related to Our objective was to model the relation between BMI and overweight in older individuals while reviewing 13 studies mortality over adult life in a large cohort using a flexible data including up to 46 954 participants aged 65 years or older. Some adaptive method without categorizing or implying a specific studies in older subjects even found an inverse relation between functional form for BMI, age and their interactive effect on BMI and mortality.7,8 One very large study reported an increased mortality. risk for BMI ⩾ 27 kg m−2 in participants aged 65 to 74 years but no We therefore used the multivariable fractional polynomials 15 association in participants older than 74 years, despite a large (MFP) procedure invented by Royston and Altman, and modified 16 number of 5363 events in this age group.9 by Sauerbrei and Royston. MFP combines the determination of Populations are becoming older in western countries and the fractional polynomial functions of continuous variables with obesity prevalence is increasing even in old age. The level of BMI backward elimination. MFP has previously been used to model associated with lower mortality in older subjects is one of the the U-shaped relationship of hematocrit and mortality,17 maternal unresolved questions in obesity research.10 Whether the recom- age and blood pressure,18 BMI and all-cause mortality in patients mended BMI range shall be constant throughout adult life or be with established coronary artery disease,19 and BMI and risk of increased for higher age groups remains unclear. work disability due to cardiovascular disease or cancer.20 The majority of studies used categories of BMI to model the Wong et al.21 investigated the use of MFP to model the BMI all- relationship between BMI and mortality,1–4,7–9,11,12 making it cause mortality relation using data of national health surveys and difficult to assess an optimal BMI range as categories are defined concluded, 'the MFP method identified improvements in model fit a priori and the optimum might lie anywhere within or near the compared with other commonly used models that estimate the 1Agency for Preventive and Social Medicine, Bregenz, Austria and 2Institute of Epidemiology and Medical Biometry, Ulm University, Ulm, Germany. Correspondence: RS Peter, Institute of Epidemiology and Medical Biometry, Ulm University, Helmholtzstrasse 22, Ulm 89081, Germany, E-mail: [email protected] Received 21 April 2014; revised 26 August 2014; accepted 6 September 2014; accepted article preview online 12 September 2014; advance online publication, 7 October 2014 Age and BMI associated with minimum mortality RS Peter et al 531 BMI–mortality relationship, and is a robust method to determine each component test and thus provides some protection against the functional form for BMI'. overfitting.25 A prespecified nominal level of α = 0.05 was used for We focused on all-cause mortality as the criterion for defining a choosing the best-fitting MFP model. Variables were rescaled and healthy weight should be the weight range corresponding to centered in the selection process to provide numerical stability. In lowest all-cause mortality.22 the second step, we refitted the models identified by the MFP procedure using sandwich variance estimation to account for the correlated structure of multiple measurements within individuals, SUBJECTS AND METHODS and kept terms still significant on α = 0.05. In step three, we The Vorarlberg Health Monitoring and Prevention Program (VHM&PP) is a included possible interaction terms of BMI and age. The final risk factor surveillance program in Vorarlberg, the westernmost province of model was determined via backward elimination of interaction Austria. Health examinations were routinely performed by the Agency of terms. A summary of the model selection can be found in ⩾ Social and Preventive Medicine and addressed all adults (aged 19 years) Supplementary Tables 1 and 2. of the entire province. Participation in the health examination was voluntary and was conducted by local physicians. The program includes a physical examination, a blood test, and a consultation with a physician. BMI associated with minimum mortality Costs are covered by the participant’s (compulsory) health insurance. A BMI associated with minimum mortality as a function of age was 23,24 detailed description of the program can be found elsewhere. Height, determined by setting the derivative of the final model with weight and smoking status among other variables were documented. The respect to BMI to zero and solving subsequently for BMI. data set was linked to the Vorarlberg death index and to the Vorarlberg BMI categories recommended by the World Health cancer registry. During the study period of 20 years (from January 1985 to Organization26 were used for comparison with optimum BMI August 2005), the participants underwent an unequal number of fi examinations from 1 up to 21. More than 55% of the general population identi ed in our data. in the eligible age range participated in the program at least once.12 Height and weight were measured in a standardized manner; Hazard relative to optimum BMI participants did not wear shoes and had only light clothing. Height was For calculation of the hazard relative to BMI associated with measured by trained stuff according to a standardized procedure with minimum mortality, the linear predictor at optimum BMI (f(age, precision of 1 cm and weight with precision of 1 kg. Ethical approval was obtained by the ethics committee of the province bmimin(age))) was subtracted from the linear predictor (f(age, of Vorarlberg. bmi)) and exponentiated subsequently. The study population consists of 185367 individuals with 7 15 414 ð ; Þ¼ ðÞðÞ; - ðÞ; ðÞ examinations. We excluded examinations of current or former smokers, HR age bmi exp f age bmi f age bmimin age examinations where the individual died within the following 3 years and R (R Foundation for Statistical Computing, Vienna, Austria) version examinations in individuals with prevalent cancer (identified via linkage to 2.11.0 and the R package mfp version 1.4.9 were used for analyses. the Vorarlberg cancer registry). Information on diagnosis of cancer has been available since January 1985; mortality follow-up was conducted until December 2009. Missing data on BMI lead to loss of additional 33 RESULTS individuals and 118 examinations. In all, 129 904 individuals with 459 740 examinations were remaining for the analysis. Reasons for exclusion from Characteristics of the study population consisting of 76 590 the study population and corresponding portions are shown in Table 1.