Introduction

Obesity is a major public health problem because of its adverse consequences for long-term health and well-being [1]. Clearly, however, not all obese individuals have the same cardio-metabolic disease risk factor (e.g., blood pressure and fasting glucose levels) profiles. The concept that someone can be obese yet metabolically healthy, most commonly called “metabolically healthy obesity (MHO)”, has been highly controversial and widely debated [2, 3]. In particular, two main types of epidemiological studies have questioned whether or not MHO is truly a benign condition, relative to metabolically healthy non-obese (MHNO). The first has investigated the progression of the MHO phenotype over time, demonstrating that this group tends to develop risk factors and transition to being unhealthy more frequently than their non-obese counterparts [4,5,6,7,8,9,10,11]. Hamer et al. [10] for example, demonstrated that, over 8 years of follow-up in the English Longitudinal Study of Ageing, 45% of MHO participants transitioned to an unhealthy state compared to 17% of MHNO participants [10]. The second type of study has investigated disease prognosis or mortality, demonstrating a ranking of risk according to both weight and health status [12,13,14,12,13,14,13]. While many of these studies present baseline differences in cardio-metabolic disease risk factors between these two groups, which manifest from dichotomizing continuous variables, few discuss the impact of these baseline differences on the reported relationships.

To illustrate the fact that applying binary cutoffs to define weight/health status induces differences in cardio-metabolic disease risk factors between MHO vs MHNO groups (and between MUO vs MUNO groups) we also present 20-year trajectories. For HDL-C and triglycerides, higher average values among MHO than MHNO individuals remained remarkably similar in magnitude across follow-up, thereby suggesting that accounting for baseline differences (as in our mortality analyses) is approximately the same as accounting for cumulative differences over time. The differences for SBP and DBP reduced marginally over time, but those for glucose and HOMA-IR increased (e.g., from ~0.0 mmol/L at baseline to 0.2 mmol/L at 20-years for glucose). These findings are in agreement with previous Whitehall II study analyses showing that, relative to MHNO, the incidence of MHO individuals develo** insulin resistance (incidence ratio 3.78 (95% CI 2.38, 5.99)) or high blood glucose (2.27 (1.43, 3.61)) over 20-years is higher than that for hypertension (1.35 (1.03, 1.77)) [5]. It appears that impairment of the glucose-insulin regulatory system might be the main factor driving transition to an unhealthy state, which would explain why the meta-analysed association of MHO with incident type 2 diabetes (relative risk ~4.0) is stronger than that for cardiovascular disease (relative risk ~1.2) [29, 32].

Importantly, our results do not mean that a person cannot be obese and have no complications. Key principles of normal variation mean that two obese individuals, even with exactly the same BMI, can (and most likely do) have different levels of cardio-metabolic disease risk factors. The idea that some people demonstrate some level of “resilience to obesity” is statistically plausible. And experimental studies in animal models [35, 36], in addition observational studies in humans [37], have started to reveal possible biological mechanisms (e.g., a genetic variant in humans near ISR1 has been shown to be related to both increased percentage body fat and a favourable metabolic profile [38]) beyond the obvious (e.g., high BMI due to high fat-free mass). The problem is that MHO is a crude way of capturing heterogeneity in health among individuals with the same BMI level. For this reason, the concept of MHO may have limited clinically utility. In a meta-analysis of nearly 150,000 participants from 14 cohort studies, Lotta et al., for example, found that binary definitions of metabolic health only had satisfactory sensitivity (0.81 (95% CI 0.76, 0.86)) and low specificity (0.42 (0.35, 0.49) for predicting incident type 2 diabetes in obese individuals [33]. Despite these limitations, a large part of the field has not moved on from asking whether or not people can be obese yet healthy. In particular, we think that more research is needed on (1) the joint distributions of BMI and cardio-metabolic disease risk factors and (2) the life course exposures that might modify the relationship of BMI with incident disease or mortality. Such investigation would help us better understand the proportion and type of people who develop a high BMI without any adverse consequences.

The main strength of the present article is the thorough analysis of longitudinal data collected on a relatively large sample over 20 years of follow-up to address a novel research question. In terms of limitations, (1) the Whitehall II study sample is not representative of the wider UK population, although standard risk factor-cardiovascular disease associations in Whitehall II are comparable to those found in nationally representative studies [39], (2) we only used one definition of MHO, which does not incorporate other measures/indicators of adiposity (e.g., waist circumference), (3) the estimated relationships might be subject to residual confounding, and (4) there were not enough cases to investigate cause-specific mortality. While these types of considerations are important when trying to infer causality from observational data, we believe they are less important for our given research aim to demonstrate why other studies (which are subject to the same limitations) find what they find. The results we present are a demonstration of some of the possible consequences of converting continuous variables to binary concepts, and may be relevant to discussions on other related phenomena, such as the “fat but fit” paradigm [40].

Conclusion

This paper demonstrates how dichotomising continuous variables results in different levels of cardio-metabolic disease risk factors at baseline and over 20 years of follow-up between MHO and MHNO individuals, despite both groups having the same label of “healthy”, and to a lesser extent between MUO and MUNO individuals. The greater disease and mortality risk of MHO compared to MHNO individuals, observed in large-scale epidemiological studies, is likely largely explained by the more deleterious risk factor trajectories (in the MHO group) that result from crude stratification. Future research needs to better quantify heterogeneity in disease and mortality risk among people with the same BMI, and investigate the characteristics and life-course factors that explain why some people develop a disease or die while other people with the same BMI do not.