Primary Care Physicians’ Perceptions of the Effects of Being Overweight on All-cause Mortality

To the Editor:

Given current evidence, clinical guidelines¹ consider obesity (body mass index [BMI] ≥ 30), but not overweight (BMI 25–29.9), to be a risk factor for all-cause mortality. Two prominent meta-analyses in general populations^2,3 have estimated associations of different BMI categories with all-cause mortality; these meta-analyses reported modest associations in opposite directions for the overweight category. Another meta-analysis suggested that in older adults, being overweight is associated with modestly reduced mortality.⁴ Although these meta-analyses had important methodologic limitations,⁵ individual studies with stronger designs have also generally found null or modest associations. Given that primary care physicians (PCPs) frequently advise patients on BMI, we assessed PCPs’ perceptions of the direction and strength of the overweight-mortality association.

Using Centiment, we recruited 192 PCPs in the United States who spent at least 20% of their time in direct outpatient care. Participants read two clinical vignettes in randomized order, both describing a 60-year-old female patient with age-typical clinical characteristics (eAppendix; https://links.lww.com/EDE/C7). The vignettes differed only regarding the patient’s BMI, which was either 28 (overweight but not obese) or 23 (normal weight); we informed participants that this was the only difference. After each vignette, participants estimated the patient’s risk of dying from any cause in the next 20 years (i.e., by age 80). As an anchor, we informed participants that the average 20-year mortality risk for 60-year-old women in the U.S. is 29%. Each participant’s “vignette-elicited” risk ratio (RR) for the overweight-mortality association was the ratio of their estimate for the overweight vignette versus the normal weight vignette.

We then directly asked participants whether being overweight (not obese) versus normal weight decreases, increases, or does not affect a patient’s risk of dying in the next 20 years, independent of all other factors. Participants who responded “increases risk” or “decreases risk” then numerically estimated by how many times being overweight increases or decreases risk, respectively. We provided concrete examples (e.g., 1.05 means 5% increased risk). These estimates represent participants’ directly estimated RR. For participants who responded “does not affect risk,” we assigned RR = 1. To reduce demand characteristics, participants were not informed at the beginning of the study that they would be providing both vignette-elicited and direct estimates. The Stanford University IRB approved this research, and we conducted analyses in R (version 4.2.0).

eTable 1; https://links.lww.com/EDE/C7 shows participant characteristics, and the Figure shows the distribution of participants’ direct estimates and vignette-elicited estimates. A large majority of participants (90%; n = 172) reported that being overweight increases mortality risk, 9% (n = 18) reported that being overweight does not affect risk, and 1% (n = 2) reported that being overweight decreases risk. When participants directly estimated effect sizes, the median RR was 1.59 (95% confidence interval: [1.50, 2.00]), indicating a perceived 59% increased mortality risk associated with being overweight. Participants’ median vignette-elicited RR was 1.25 (95% confidence interval: [1.20, 1.31]), indicating a perceived 25% increased mortality risk. We report exact confidence intervals based on the binomial distribution. Substantial minorities of participants estimated that being overweight more than doubles mortality risk (31% of direct estimates; 11% of vignette-elicited estimates). In post hoc analyses, we estimated associations of participants’ demographic and professional characteristics with their estimates (eAppendix 2; https://links.lww.com/EDE/C7).

FIGURE.: Distribution of each participant’s direct estimate versus their vignette-elicited estimate. Rug plots along each axis depict marginal distributions. Reference lines indicate the estimates from each meta-analysis. Both axes are presented on a log scale. Twelve points falling outside the depicted axis ranges are omitted; a plot with expanded axis ranges that shows all data points appears in eAppendix 2 (eFigure 1); https://links.lww.com/EDE/C7.

The aforementioned meta-analyses in general populations estimated associations between overweight and mortality of hazard ratio = 0.94 (Flegal et al’s meta-analysis²) and HR = 1.11 (Global BMI Mortality Collaboration’s meta-analysis³). Participants’ median estimates of RR = 1.59 (directly estimated) and RR = 1.25 (vignette-elicited) exceeded each of these meta-analytic estimates. (In both meta-analyses, the outcome was rare, so the HRs are approximately equal to RRs.) Almost all participants’ estimates exceeded the first meta-analytic estimate² (99% by direct estimation and 90% by vignette elicitation), and a majority also exceeded the second meta-analytic estimate³ (70% and 71%). Effects of being overweight may differ across study designs and populations; however, participants’ median estimates exceeded even the estimated 90th quantiles of effect sizes in both meta-analyses (HR = 1.09 and 1.21, respectively).⁵ These quantiles refer to the population effects rather than the point estimates (which include statistical error); we estimated the distribution of population effects using nonparametric meta-analytic methods.⁶

In summary, a large majority of PCPs (90%) perceive that being overweight increases all-cause mortality risk, contrasting with findings of clinical guidelines’ authors.¹ PCPs typically estimate a strength of association that substantially exceeds current empirical estimates.^2–4 Limitations of our study include potential demand characteristics and exclusive focus on mortality outcomes. Future research could examine how PCPs’ perceptions affect decision-making and treatment for overweight patients, a timely issue given newly approved, effective weight-loss drugs.⁷