The obesity pandemic continues expanding unabated. Currently, 66% of adults in the United States are overweight or obese, and this figure is expected to rise to 75% by 2015 (1). Although obesity is a well- recognized risk factor for mortality from cardiovascular disease and cancer (2,3,4), it is uncertain whether it could influence the acute risk of death in critically ill patients. While some studies conducted in intensive care units (ICUs) indicate increased risk of death for overweight or obese patients (5,6,7,8), others have reported inverse (9,10,11,12,13,14,15,16,17) or null (18,19,20,21) associations. It is relevant to examine to what extent obesity is a risk factor for mortality in critically ill patients, given that BMI has been proposed as a criterion to routinely predict survival of patients at the time of admission to the ICU (5,12,22). We conducted a systematic review and meta-analysis of published research papers that addressed the association between overweight or obesity and mortality in critically ill adults. The aims of the review were to investigate whether overweight and obesity are independent predictors of mortality, duration of stay, or multiple organ dysfunction (MOD) in patients admitted to the ICU. The systematic review included observational studies that examined the associations between BMI as a categorical variable and mortality among critically ill adult patients (age >18 years). Secondary endpoints included the duration of stay in the ICU and the incidence of MOD. BMI (kg/m2) exposure categories were defined according to the World Health Organization's (WHO) proposed cutoff points (23) as follows: underweight, <18.5; normal, 18.5–24.9; overweight, 25–29.9; obese, 30–39.9; and severely obese, ≥40. We searched for articles published in English and indexed in MEDLINE from 1966 to June 2007 and EMBASE from 1980 to June 2007 with the use of the following strategy: ("critical illness" or "critical care," or "critically ill patients" or "critically ill*") and ("body mass index" or "BMI," or "obesity" or "obese," or "morbid obesity" or "morbidly obese," or "overweight") and ("mortality" or "ICU mortality" or "intensive care unit mortality," or "length of stay" or "duration of stay," or "MOD" or "multiple organ failure"). Both authors independently assessed the titles and abstracts and selected potentially eligible studies for full-text review. We designed a data capture form that included the study population (patients with medical conditions, surgical conditions, trauma, acute lung injury, or combinations of patients), the BMI exposure and reference categories used, and the outcomes reported. Data collection was carried out independently by the authors. After an additional review of the studies, we resolved discrepancies by consensus. All the studies that fulfilled inclusion criteria assessed the exposure (BMI) on admission to the ICU. We obtained an initial summary measure of association between BMI and mortality for all the studies, by estimating the odds ratio (OR) and 95% confidence intervals for each category of BMI >25, compared to normal BMI. We used both fixed and random effects models to obtain these estimates. We evaluated whether the study population and the categorization of BMI were significant sources of heterogeneity across studies. Only the use of different BMI categories significantly explained heterogeneity, hence we conducted subgroup analyses according to categories of BMI. From each study, we extracted the risk of mortality adjusted for potential confounders when this was available. The majority of studies adjusted the estimates for disease severity on admission to the ICU, by using the Acute Physiology and Chronic Health Evaluation score. We assessed the presence of publication bias with the use of the Begg-Mazumdar test (24). Analyses were carried out with the use of CMA (Biostat, Englewood, NJ) and Stata version 9.0. Our search strategy yielded an initial list of 260 publications (Figure 1). Thirty-seven of these publications met the inclusion criteria after the title and abstract review and were selected for a full-text review. Twenty-three studies were selected for data extraction after further exclusions. Nine studies reported mortality by the end of hospital stay, while another eleven studies included the mortality endpoint by the end of ICU stay. Three studies reported mortality after 1 month from ICU admission. : Studies selection process. The study populations varied markedly (Table 1). Three studies included patients with medical conditions (respiratory or heart failure, infection/sepsis, and neurological, metabolic, or kidney disorders); another two studies were conducted among surgical patients (vascular, cardiothoracic, or general surgery, transplantation, neurosurgery, and other assorted surgical procedures), one study analyzed medical and surgical patients separately, six included a combination of medical and surgical patients, three were done in patients with acute lung injury, and eight studies included trauma patients. Twelve of the twenty-three studies selected for data extraction examined the risk of mortality in relation to BMI by using 18.5–24.9 as the reference category. Tremblay et al. (10) and Ray et al. (13) used slightly different lower cutoff points but were also included in the calculation of pooled estimates. By pooling the studies according to BMI categories (Table 2), we found a decreasing trend in the risk of mortality when BMI was between 25 and 39.9 compared to BMI between 18.5 and 24.9. The pooled OR estimate for the 25–29.9 category compared to 18.5–24.9 was 0.91 (P = 0.01) (Figure 2a). The estimate for the 30–39.9 BMI category was 0.82 (P = 0.03), but the interpretation is hampered by significant heterogeneity of the studies in this comparison (Table 2). There was no apparent association between severe obesity (BMI ≥ 40) and mortality (OR = 0.94; P = 0.26) (Figure 2b). We found no evidence of publication bias for the main comparisons of mortality risk by BMI categories, namely, 25–29.9, 30–39.9, or ≥40 compared to 18.5–24.9. : Pooled odds ratio (OR) for mortality in (a) overweight (BMI: 25–29.9) and (b) severely obese (BMI > 40) critically ill patients compared to normal BMI patients (18.5–24.9). The OR for overweight patients from a random effects model was 0.91 (95% confidence interval (CI) = 0.84, 0.98; P = 0.01). The OR for severely obese patients was 0.94 (95% CI = 0.82, 1.07; P = 0.26). Eleven studies reported the average length of stay in the ICU for the BMI categories considered (Table 3). Five studies (5,9,14,20,25) found significantly longer durations of stay in obese or severely obese patients, compared to those of normal BMI, while six studies did not show significant differences (7,13,16,17,18,26). The pooled average durations of stay in the ICU were significantly longer for underweight, overweight, and severely obese patients compared to those of normal BMI (Table 4). Obese patients (BMI 30–39.9) also appeared to have longer lengths of stay, but the estimates were not statistically significant. There was evidence of heterogeneity across studies in this BMI category. Six studies reported associations between BMI and MOD (Table 5), including Brown et al.'s re-analyses (27) of Neville et al.'s report (7). Four studies found significantly increased risk of MOD associated with overweight or obesity (12,14,21,27). We found an apparently inverse association between overweight or obesity and mortality in critically ill adults, while severe obesity was not significantly related to this outcome. By contrast, overweight and severely obese patients appeared to have longer stays in the ICU and increased risk of MOD, compared to patients of normal weight. The interpretation of these associations needs to be cautious, since the observational studies reviewed may suffer from common methodological limitations. Reverse causation could explain apparent benefits of overweight and obesity in critically ill patients, since preexisting conditions that led to the most serious diseases could have caused weight loss prior to ICU admission. Studies conducted in patients who suffered traumatic injuries offer an opportunity to assess whether reverse causality is likely to explain the inverse association between overweight or obesity and mortality, since these patients may not have had serious preexisting conditions. In these studies, obesity was either a risk factor for mortality (7,8,25,27) or was not significantly related to this outcome (14,19,21,26), suggesting that reverse causality cannot be ruled out. The timing when body weight was assessed also deserves to be considered in the interpretation of results, given that increased body weight measured at the time of ICU admission could represent fluid retention from prior in-hospital treatment, rather than usual weight. The apparently "protective" impact of obesity on mortality would seem contradictory to its positive relation with ICU stay, an indicator of ICU complications. It could be argued that leaner patients die shortly after ICU admission and, therefore, have both shorter stays and greater mortality than obese patients. On the other hand, longer duration of stay in obese patients, independent of mortality, could be related to their greater dependence on mechanical ventilation due to increased airway resistance, abnormal chest elasticity, and inefficiency of the respiratory muscles (5). A contradiction with the increased incidence of MOD is more difficult to reconcile. Some authors argued that obese patients receive superior care at the ICU (26), as they may be considered "high risk," and, therefore, they may have reduced mortality even if their incidence of complications is greater than that of normal weight patients. Obesity has been found to be protective against mortality in patients suffering from chronic, debilitating conditions, including congestive heart failure, end-stage renal disease, advanced malignancies, and HIV/AIDS—a phenomenon called the "obesity paradox" (28,29). One explanation for this paradox is that high BMI may confer survival benefits by providing nutritional reserves in these patients. It is conceivable that a similar mechanism may be at play in some critically ill patients, who are suddenly exposed to intense inflammatory and metabolic stress. Alternatively, specific hormonal mechanisms could play a role in the relation between obesity and mortality. Obese patients have higher levels of leptin, and Bornstein and colleagues reported a positive association between leptin concentrations and survival of septic patients (30), suggesting that leptin could play a role in the adaptive response to critical illness. Under the "obesity paradox" hypothesis, the fact that obesity is not as "protective" in trauma patients compared to those admitted to the ICU for medical or surgical reasons could be explained if the former are younger than the latter, as nutritional reserves may offer greater benefits to elderly patients than to younger ones. Some authors have proposed the inclusion of BMI in the algorithms that are routinely used on admission to the ICU to predict survival (5,12,22). This suggestion has been made in light of the studies that found positive associations between obesity and mortality. The results of our review indicate that there is no sufficiently robust evidence at this point to support the inclusion of BMI as a predictor of mortality for any critically ill patient in such algorithms. Future studies assessing the relation between obesity and mortality in critically ill adults should incorporate measures of the "usual" nutritional status prior to the occurrence of diseases that may have led to ICU admission. Measurements of body composition or adiposity other than BMI are not routinely obtained in patients admitted to the ICU; some of these measurements could also contribute to elucidate the association between obesity and mortality in future investigations. The authors declared no conflict of interest.