We found a nonlinear association between the baseline WC and mortality in the NHANES dataset of US diabetic adults. Even after considering confounders, the relationship of WC and death was U-shaped in women, with a lowest death risk in the central obesity range (WC 107.0 cm for all-cause death and 118.1 cm for CVD death, respectively), and J-shaped in men, with a minimum death risk in the normal waist range (WC 89.5 cm for all-cause death and 87.6 cm for CVD death, respectively).
As the different sexes have different WCs, women and men were assessed separately herein. At first glance (presented in Table 2), there was no significant association between WC and mortality using the linear regression model, which was comparable to a prior investigation including the NHANES general population [20]. However, based on the two-piecewise linear regression approach, all-cause or CVD risk markedly increased with elevated baseline WC when it exceeded the turning points, consistent with previous investigations performed in subjects without DM [17, 18, 23–27]. Among a sample of DM, WC has been demonstrated to be positively associated to all-cause mortality (HR = 1.40, 95% CI 1.14–1.72) [6]. Likewise, WC independently and positively captured the possibility of all-cause death in DM patients from the Accord Trial [28]. However, the Fremantle Diabetes Study established no association between WC and mortality in either sex [29]. Furthermore, even though a similar U-shaped pattern was also detected between all-cause mortality and WC among the subgroup of subjects with type 2 DM from the UK Biobank population (n = 23,842), mortality was not obviously altered by WC when it exceeded the turning point in either sex [13]. The different associations across studies involved in DM might result from differences in sample size, ethnicity, health status, baseline levels of WC, and duration of DM. In general, the impact of visceral adipose tissue may explain the association between greater WC and increased mortality [18].
In the present study, interestingly, lower levels of baseline WC (< thresholds) remarkably altered the association between WC and death risk. In subjects with a baseline WC lower than the cut-offs, the risks of all-cause death were reduced by 2% and 9% among women and among men, respectively, with each 1 cm increment in the baseline WC even with full adjustment for confounders, including BMI. This finding is partly consistent with a prior investigation performed by Cho et al. among a health check-up population in South Korea, which illustrated that the death risk decreased with increasing WC when it was lower than 85 cm among men and lower than 80 cm among women [14]. Additionally, other studies also found a remarkably negative association between all-cause death and WC [19, 30–32]. However, such an inverse relationship of WC with mortality existed mainly in older individuals, and all of these populations did not suffer from DM [19, 30–32]. Among the Type 2 DM subgroup in the UK Biobank, a U-shaped trend similar to ours was detected between WC and all-cause death. Unfortunately, the two-piecewise linear regression was not conducted in their study, and it was unclear whether there was a remarkably negative relationship of the baseline WC with all-cause mortality before the turning points [13]. Likewise, the Fremantle Diabetes Study did not report such a negative relationship between WC and mortality in either sex, which might partly derive from the relatively small sample size included. Therefore, this study first reported a significantly negative association between WC and total and CVD mortality risk among patients with DM before the thresholds. In fact, in observational epidemiological research, this inverse relationship denoted the obesity paradox, which was initially hypothesized in response to the unexpected observation that individuals who were overweight or obese could obtain a longer life expectancy than those with normal weight [13]. Certainly, the obesity paradox was earlier reported in other investigations that commonly utilized BMI as the obesity measure [19, 30–32]. Among patients with DM, a meta-analysis including 414,587 subjects regarding the prospective investigations also reported a remarkable reverse relationship of all-cause death with BMI (< 31kg/m2 for men, as well as < 28kg/m2 for women, respectively) [10]. However, the WC-related obesity paradox is rarely reported in DM. A series of possible reasons for the negative relationships of BMI or WC with death has been discussed in other investigations, consisting of survival or selection bias, malnutrition-inflammation complex syndrome, and toxic material storage [19, 30–33]. The above explanation can also accommodate the present study considering that diabetes is also a chronic wasting disease [19]. Selection bias was regarded as the primary problem in the inverse association, and a lower WC may represent the existence of other severe diseases in individuals with DM and then lead to death [33]. Herein, we attempted to mitigate selection bias by subgroup analyses and found that the reverse relationship of WC with mortality tended to disappear among patients without previous CVD or cancer in both sexes.
Another possible reason for the reverse association of WC with death risk could be a result of survival bias. This work supported the speculation since the average age in the lowest tertile of WC of this study was older than that in the middle and highest tertiles, particularly in women (data not shown). Additionally, adipose tissue may release numerous hormones with anti-inflammatory properties and may act as a relatively safe storage place for hazardous substances, partly at least explaining the negative relationship between WC and death risk [34].
It is essential to concentrate on the various correlations of WC with mortality risk stratified by a variety of ages, considering that both women's and men's body compositions, in addition to adiposity characteristics, continue to alter with age. In this study, J- or U-shaped correlations between WC and all-cause mortality risk were likewise discovered in both the aged ≥ 60 y and < 60 y groups. However, in patients aged < 60 y, as opposed to those aged ≥ 60 y, a stronger positive association would be observed between all-cause death and WC once WC was beyond the threshold. Similar to our findings, other studies found a stronger positive association between total death and WC among the younger age group [14, 31]. This phenomenon might be due to that in younger individuals with DM, a larger WC is essentially a risk factor for cardiometabolic disorders, while fat mass serves as part of the nutritional reserve function among older individuals [14].
It seems to be necessary to consider the effect of BMI when investigating the association between mortality and WC [14, 18]. Our findings from subgroup analyses according to BMI categories exhibited that the curve shapes varied across the BMI groups, with U-shaped, reverse U-shaped, and even linear presentations. Particularly in men, our findings showed that the negative relationship between death and WC was only obvious among the normal BMI subgroup, which was similar to a Korean study [18] but different from a European study [27]. We speculated that the increased death risk associated with lower WC only in the normal BMI category may partly be explained by the loss of beneficial fat, such as muscle mass.
It was also found that smoking might be a modifiable factor in the association between mortality and BMI [13, 27]. Particularly among current smokers, being underweight was more strongly related to an elevated death risk, which was also partly responsible for the obesity paradox [13, 27]. In contrast, when smokers' BMI was adjusted, WC became more strongly and positively associated with mortality [27]. The potential reason was that smokers often had a more metabolically unfavourable adipose distribution and were more likely to be obese in the abdominal region than non-smokers [27]. Nevertheless, in our subgroup analyses, this positive correlation between WC and death appeared to be more pronounced among never-smokers of both sexes. Although the reasons are unclear, differences in population chosen may explain variations between the current investigation and earlier studies.
The present study has several strengths. First, to assess if there is a dose-response association between the baseline WC and mortality, we treated WC as a categorical variable, in addition to a continuous variable to explore its nonlinear association with death, which enabled us to find cohort-distinct WC values linked to the lowest mortality risk in US adults with DM from the NHANES. Second, we broadly explored the association in the whole cohort and in the subgroups where the obesity paradox has been documented earlier (i.e., smoking, elderly, previous CVD, or cancer).
Limitations
First, the baseline WC cannot be considered to be causally related to death, considering the observational character of this investigation. As a result, even though we tried to decrease the possibility of confounding by controlling for a series of variables, the influence of unmeasured confounders cannot be fully ruled out. Since obesity is also a risk factor for certain comorbidities, collider stratification bias could also amplify confounding impacts. Second, there were differences between subgroups with respect to the shape of the curves, which might deserve further research. Third, WC and other confounders might have been changed by the symptoms of the disease and/or its therapies but they were only recorded at baseline. Lastly, because NHANES was used to enrol all subjects with DM, conclusions should be extended to other DM cohorts.