The primary objective of this study was to investigate whether the combined use of WC and BMI improves the identification of high costs individuals. The present analysis complements a previous study based on the same dataset which found that a combination of WC and BMI strongly predicted all-cause mortality, in both men and women 
. Our hypotheses were that: for all levels of BMI, increased WC implies added health care costs (Hypothesis 1) and for a given WC increased BMI implies reduced health care costs (Hypothesis 2).
The results based on the analysis where BMI and WC are treated as continuous variables show that for a given BMI inclusion of WC improves the identification of high costs individuals, reflecting that a more abdominal fat distribution for a given level of BMI gives higher health care costs. The complementary analysis in which BMI and WC are treated as categorical variables suggests the added value of WC is mainly found amongst individual with BMI<30 kg/m2
, which corresponds to the findings on mortality 
. demonstrates that future means costs are driven by differences in consumption of health care costs at base-line as well as a difference in the rate of increase in costs over the 7 year observation period.
The previous findings that BMI and WC have opposite effects on total mortality when conditioned on each other 
, that fat mass and lean body mass may have opposite effects on health 
and that WC captures the deleterious effects of fat mass on mortality 
have increased the interest in combining these two measurements, thereby potentially achieving an increased accuracy in the identification of individuals at greatest risk. As in many previous analyses, we find a strong positive correlation between WC and BMI (0.87 for men, 0.86 for women). Despite this high correlation between BMI and WC, the results show that the categorization of individuals into risk groups according to their obesity status differs between the two obesity measures. We observe that WC is a more sensitive measure for capturing the high cost individuals. Individuals with a normal BMI but increased or substantially increased WC incur higher mean costs than those with a normal BMI and a normal WC. Interestingly, this group incurs significantly higher mean future health care costs than do individuals who are overweight and obese but have a normal WC.
These results illustrate our general finding that WC is a better predictor of future health care costs. This finding is in agreement with a small study (n
424) which found that total health care costs were better correlated with WC than with BMI 
. The findings are also in agreement with recent epidemiological studies which indicate that WC is a stronger marker of health risk than BMI is 
. BMI has been criticized for misclassifying muscular subjects as being overweight when, in fact, they are lean 
and it has been shown that the abdominal fat mass can vary dramatically within a narrow range of BMI 
. The fact that WC is a good indicator of the location of the excess adipose tissue, and that visceral fat seems to be highly related to health risk 
most likely explains this study's observations.
One may, however, argue that the use of WHO's categorization may be suboptimal in this context since the WC and BMI cut-off points were not designed to be used in combination. Rather, the WC cut points were derived by use of BMI 
. However, as discussed above the categorization of individuals into risk groups according to their obesity status differs between the two obesity measures. Arder et al. 
have in a recent study shown that the optimal WC thresholds increased across BMI categories when predicting future coronary events, and Bigaard et al. 
have found the same for total mortality. As also recently discussed by an expert panel 
these findings indicate that there is a need to investigate whether it is necessary to develop special thresholds when evaluating BMI and WC in combination. Furthermore, it should be noted that despite the widely used recommendation of the applied action levels of WC these are still under debate.
Our results show that BMI coupled with WC did not predict obesity related health risk better than WC did alone. This finding is in agreement with the findings in a previous study which found that BMI coupled with WC did not predict obesity-related health risk (measured by the odds ratios for different metabolic variables, e.g. blood pressure and cholesterol) better than WC did alone when these two anthropometric measures were examined on a continuous scale 
. However, when WC was dichotomized into normal and high-risk categories, BMI remained a significant predictor of health risk. The authors suggest that the reason for this observation lies in the fact that when WC is treated as a categorical variable whilst BMI is a continuous variable, BMI may capture some of the variation in WC within a WC category. In addition, in a recent article Han et al. conclude that due to the large correlation between BMI and WC, a combination of the two measures adds relatively little to the risk prediction 
Recently a panel composed of members with expertise in obesity concluded that the current WC cut points are useless in a clinical practice when BMI is already applied, and that more useful WC cut points are recommended to be found 
. According to our findings WC rather than BMI should be used.
In this study we have chosen to focus on all types of health care consumption and not only health care consumption related to an a priori specified disease associated with obesity such as diabetes and cardio vascular diseases. A specific focus on one disease provides a fragmented description of the disease profile of obese individuals. Instead, we chose the non-discriminatory approach and focused on all types of health care consumption in order to provide a fuller description of the association between obesity status and need for health care services. This strategy introduces a large degree of variation in health care costs that may not be directly related to obesity status. Consequently, the statistical models in produce R2 values in the range 0.04 and 0.08. However, despite this broad perspective and without focus on predefined specific obesity related diseases we are still in position to show that WC is a better predictor than BMI.
Zweifel and colleagues have previously proposed the so-called ‘red herrings’ hypothesis that proximity to death is a more important predictor of health-care costs than is age 
. Our results support the red herrings hypothesis, as individuals who died in the observation period incurred higher health care costs in the year prior to death. To the extent that persons with high BMI and/or high WC were more at risk of dying, they would on average incur higher future costs. Amongst men, we found a higher mortality amongst the obese (when defined in terms of WC as well as BMI) hence proximity to death may be one of the drivers underlying the observed higher rate of increased health care costs amongst obese men. The differences in mean annual costs may in this case overestimate potential cost savings associated with reduction in obesity.
There are both strengths and limitations to this study. A major strength is the large population sample (n
31,840), which ensures sufficient power in the analyses. In addition, individual data on health care consumption and associated cost were extracted from valid registers and the anthropometric measurements were measured by trained staff. Potential sources of bias and confounding factors including illness-related weight losses were sought eliminated with the exclusion of subjects with BMI under 18.5 kg/m2
or with a history of cancer. Except for the cost data, the analyses were based on cross-sectional data, implying only one single measurement of WC and BMI, leaving information's about the WC and BMI status in the follow-up period missing. This implies that the study results are based on associations of inter-individual WC and BMI differences, and not intra-individual WC and BMI changes on health care costs. Also, there is a risk of selection bias, since only one-third (35%) of the invited individuals participated in this study, and it is likely that in general it is the healthier fraction who chose to participate in the study.
Our results show that combined use of WC and BMI increases the prediction of high cost individuals, for a given BMI, reflecting that a more abdominal fat distribution for a given BMI gives higher health care costs. However, inclusion of BMI information for a given WC only increases the prediction of future health cost amongst women with BMI<30 kg/m2 and WC <88 cm.