The J‐shaped relationship between body mass index and mortality in patients with COVID‐19: A dose‐response meta‐analysis

Abstract

The coronavirus disease 2019 (COVID-19) pandemic has caused a considerable number of deaths. Identifying individuals at higher risk of critical illness and death is critical for planning prevention strategies, such as assigning vaccination priority. Several studies have linked obesity to more severe illness and higher mortality in COVID-19 patients. However, the relationship between underweight and COVID-19 mortality remains inconclusive; previous dose-response meta-analyses did not include the underweight population in their evidence synthesis. We conducted a systematic review and doseresponse meta-analysis to investigate the relationship between body mass index (BMI) and mortality in both obese and underweight patients with COVID-19. We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines. The protocol was registered in the International Platform of Registered Systematic Review and Meta-analysis Protocols (registration number: INPLASY2020120090). We searched the PubMed, Embase, Cochrane Library, Scopus and Web of Science databases from inception until February 11, 2021 using the keywords “COVID-19”, “body mass index , “obesity”, “overweight”, “underweight” and “mortality.” Details of the search strategies and article selection process are shown in the Supplementary Materials. We included studies if they: (i) reported mortality risk for patients with COVID-19; (ii) divided patients into at least three different BMI categories and reported the relative risk (RR) of mortality for each category; and (iii) reported adjusted estimates (adjustment for age and sex at minimum). We only included studies that reported at least three BMI categories and the numbers of patients and deaths for each BMI category to investigate a potential nonlinear trend in dose-response meta-analysis. Both clinical trials and observational studies that provided sufficient data were eligible. Review articles, case reports, editorials, letters and conference abstracts were excluded. Studies that reported only crude estimates without adjusting for confounders were excluded. The primary outcome was mortality. Three reviewers (H.K.H., K.B. and D.P.H.) independently assessed the relevant articles to identify eligible studies, three reviewers (H.K.H., K.B. and D.P.H.) independently extracted the data, and two reviewers (K.B. and D.P.H.) assessed the quality of the studies using the Newcastle-Ottawa Scale. Discrepancies were resolved via discussion among the study team. We first conducted a meta-analysis for the difference in the risk of mortality between the highest and the lowest category of BMI using a DerSimonian and Laird random-effects model (the high vs. low meta-analysis). We then conducted the random-effects doseresponse meta-analysis to estimate the linear and nonlinear trends in the association between BMI and mortality. The linear trend was estimated by using the generalized least squares model described by Greenland and Longnecker. We used the two-stage approach to estimating the nonlinear trend by first fitting a restricted cubic splines model with knots at the 10th, 50th and 90th percentiles for each study and then undertaking a multivariate meta-analysis for the model variables. The Wald test was used to test for nonlinearity by comparing the model fit between the linear and nonlinear models. When the BMI level was presented as a range, the dose was assigned using the midpoint of the upper and lower boundaries; for the open-ended highest and lowest BMI categories, the width between the boundaries was assumed to be equal to that of the adjacent category. RRs for mortality with 95% confidence intervals (CIs) were used to report the outcome. For the dose-response meta-analysis, a sensitivity analysis was conducted by pooling only studies specifically evaluating underweight patients (BMI < 18.5 kg/m). We assessed heterogeneity among studies with I statistics. The heterogeneity was considered low, moderate and high for I < 50%, 50% to 75%, and > 75%, respectively. Potential publication bias was assessed using funnel plots, Egger s test and Begg s test. A leave-one-out sensitivity analysis was performed to evaluate the influence of each study on the overall pooled estimate. All statistical tests were two-sided, with the significance level set at 5%. Statistical analyses were conducted using Stata version 15.1 (StataCorp, College Station, Texas) and R software version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria). Institutional ethical approval was not required because this was a meta-analysis of primary published studies only. Of the 7443 potential studies screened, 4455 duplicate studies, 2393 irrelevant studies, and 567 studies without usable data on this topic were excluded, yielding 28 studies comprising 112 682 patients for the analysis (Figure S1). The characteristics of the included studies are summarized in Table 1. The mean ages of the patients ranged from 51 to 71 years, the proportion of female participants ranged from 9% to 67%, and the sample sizes ranged from 191 to 25 952. The majority of the included studies were conducted in the United States and Europe and were retrospective cohort studies. Among them, 13 studies evaluated underweight patients specifically. All the included studies had an Received: 10 February 2021 Revised: 15 March 2021