See discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/273140297 A dose-response meta-analysis of the impact of Body Mass Index on stroke and all-cause mortality in stroke patients: A paradox within a paradox ARTICLE in OBESITY REVIEWS · APRIL 2015 Impact Factor: 7.86 · DOI: 10.1111/obr.12272 DOWNLOADS VIEWS 44 84 5 AUTHORS, INCLUDING: Minoo Bagheri John Speakman Tehran University of Medical Sciences University of Aberdeen 6 PUBLICATIONS 4 CITATIONS 502 PUBLICATIONS 13,378 CITATIONS SEE PROFILE SEE PROFILE Sakineh Shab Bidar Kurosh Djafarian Tehran University of Medical Sciences Tehran University of Medical Sciences 24 PUBLICATIONS 128 CITATIONS 39 PUBLICATIONS 127 CITATIONS SEE PROFILE SEE PROFILE Available from: John Speakman Retrieved on: 23 June 2015 obesity reviews doi: 10.1111/obr.12272 Review A dose-response meta-analysis of the impact of body mass index on stroke and all-cause mortality in stroke patients: a paradox within a paradox M. Bagheri1, J. R. Speakman2,3, S. Shabbidar1, F. Kazemi1 and K. Djafarian4 1Department of Community Nutrition, School Summary of Nutritional Sciences and Dietetic, Tehran The obesity paradox is often attributed to fat acting as a buffer to protect University of Medical Sciences, Tehran, Iran; individuals in fragile metabolic states. If this was the case, one would predict that 2Institute of Biological and Environmental the reverse epidemiology would be apparent across all causes of mortality Sciences, University of Aberdeen, Aberdeen, including that of the particular disease state. We performed a dose-response Scotland, UK; 3Institute of Genetics and meta-analysis to assess the impact of body mass index (BMI) on all-cause and Developmental Biology, Chinese Academy of stroke-specific mortality among stroke patients. Data from relevant studies were Sciences, Beijing, China; 4Department of identified by systematically searching PubMed, OVID and Scopus databases and Clinical Nutrition, School of Nutritional were analysed using a random-effects dose-response model. Eight cohort studies Sciences and Dietetic and School of Public on all-cause mortality (with 20,807 deaths of 95,651 stroke patients) and nine Health, Tehran University of Medical Sciences, studies of mortality exclusively because of stroke (with 8,087 deaths of 28,6270 Tehran, Iran patients) were evaluated in the meta-analysis. Non-linear associations of BMI with all-cause mortality (P < 0.0001) and mortality by stroke (P = 0.05) were Received 26 January 2015; accepted 2 observed. Among overweight and obese stroke patients, the risk of all-cause February 2015 mortality increased, while the risk of mortality by stroke declined, with an increase in BMI. Increasing BMI had opposite effects on all-cause mortality and Address for correspondence: Dr K Djafarian, stroke-specific mortality in stroke patients. Further investigations are needed to Department of Clinical Nutrition, School of examine how mortality by stroke is influenced by a more accurate indicator of Nutritional Sciences and Dietetic, Tehran obesity than BMI. University of Medical Sciences, No 44, Hojjat-dost Alley, Naderi St., Keshavarz Blvd, Keywords: Body mass index, dose-response meta-analysis, obesity, paradox, Tehran 1416-643931, Iran. stroke. E-mail: [email protected] obesity reviews (2015) increasing BMI above the threshold level (7–9). Following Introduction the onset of some disease states, however, it has been Among otherwise healthy subjects, increasing body mass observed that it is the more obese subjects with greater BMI index (BMI), generally reflecting an increased level of who survive the longest (10). This effect, variously called obesity (1) above a threshold value of around BMI = 23 to ‘reverse epidemiology’ or the ‘obesity paradox’ has been 25, is associated with increasing risk of various non- demonstrated to occur in stroke patients (11–13), as well as communicable diseases such as type 2 diabetes (2), sufferers from chronic kidney disease (14,15), CVD (16) hypertension and cardiovascular disease (CVD) (3), non- and specifically heart failure (HF) patients (17), general alcoholic fatty liver disease (4) and some forms of cancer surgery patients in intensive care (18), patients with surgi- (5,6). Consequently, large epidemiological surveys reveal cally induced peritonitis (19) and patients following bone that there is an increasing risk of all-cause mortality with fractures (20). © 2015 World Obesity 1 2 BMI effects on stroke and overall death M. Bagheri et al. obesity reviews The widespread nature of this effect across a variety of were extracted from each study independently by two seemingly unrelated disorders has led to speculation that it reviewers. The extracted data included first author’s last may be an artefact of biased subject selection and hence name, publication year, country, study design, follow-up contributes to biased estimates of obesity-related mortality length, sex, mean age, number of patients and number of (21,22). Yet another interpretation, however, is that in the cases, adjusted RR/HR with the corresponding 95% CI, fragile metabolic state following various traumatic events, case ascertainment, stroke subtype and variable adjusted elevated body fat levels may act as a buffer protecting for in the multivariable analysis. Disagreements in data the individual and therefore facilitating recovery extraction were settled by consensus with a third (11,12,23,24). If this latter interpretation is correct then it reviewer. would be predicted that the mortality paradox would be observed with respect to all causes of mortality among Statistical analysis patients with specific disease states and not just in the mortality because of each particular disease. Yet, this dis- We calculated HRs related to five-unit increases in BMI for tinction has been rarely addressed. To address this topic, each study by transforming supplied category-specific risk we performed a dose-response meta-analysis of studies of estimates with the use of generalized least squares for trend stroke patients separating the effects of BMI on all-cause estimation described by Orsini et al. (25). and stroke-specific mortality. When the lowest group was not the referent, HRs and CIs were estimated as relating to the referent for which data were required (26). In studies with stratified analyses by Methods sex, the combined estimates of HRs and CIs were calcu- lated using a fixed-effects model. Search strategy and data extraction For the dose-response meta-analysis (25,27), the mid- PubMed, OVID and Scopus databases were searched to point of each BMI category was utilized if BMI mean or identify the papers written in English published up to 7 median for the category was not provided in the study. The July 2014. Further studies were identified by hand- open-ended categories were assumed to have the same searching the reference lists of the relevant reviews and amplitude as their neighbouring categories. meta-analysis. The following key words were used to A two-stage hierarchical regression model (28) was extract the prospective articles pertinent to the study administered to estimate the non-linear dose-response rela- objectives: ‘obesity’ OR ‘overweight’ OR ‘Body Mass tion across different BMI levels. In this model, the differ- Index’ OR ‘BMI’ OR ‘body weight’ OR ‘adiposity’ OR ence between the midpoints of category-specific and ‘fat mass’ OR ‘body fat’ OR ‘body size’ OR ‘body com- reference-specific quadratic terms for BMI, based on position’ AND ‘death’ OR ‘survival’ OR ‘mortality’ OR studies with non-zero BMI level as reference, was exam- ‘prognosis’ OR ‘paradox’ AND ‘stroke’ OR ‘prestroke’ ined. Then, the dose-response relationship, considering OR ‘poststroke’ AND ‘follow-up’ OR ‘cohort’ OR ‘pro- within- and between-study variances, was estimated using spective studies’ OR ‘retrospective studies’ OR ‘longitudi- spline transformations. Random-effects dose-response nal studies’ OR ‘observational studies’. The criteria used models using logarithms of HRs and CIs, the number of for study selection included a prospective design, a study deaths and the number of participants across BMI catego- population covering adult aged >18 years, assessment of ries were performed assuming linearity in the potential BMI as the exposure of interest, assessment of mortality relationships. by stroke or all-cause mortality in stroke patients as the The Q and I2 statistics were used for exploring outcome of interest, measurement of the relative risk (RR) statistical heterogeneity among the included studies. The or hazard ratio (HR) with corresponding 95% confidence values of 25%, 50% and 75% were regarded as low, intervals (CIs), measurement of the number of patients moderate and high heterogeneity, respectively. Stratified and number of cases in each BMI category. The studies analyses by sex, mean or median years of follow-up, the were excluded if they were reviews, commentaries, study quality and the time of BMI measurement were research notes, letters or conference proceedings and if conducted to evaluate the potential sources of heteroge- they were conducted on pregnant populations. When neity. We also performed a sensitivity analysis in which duplicate records on the same study population were iden- a single study was excluded from the analysis at a time tified, the bigger sample size study was considered for to investigate whether different results were obtained. To inclusion in the review. Authors were contacted by email detect publication bias,