Of the five anthropometric measures studied, WHR and WHtR were most strongly associated with mortality, after adjusting for confounding factors. This was true both regarding overall mortality and death from CVD specifically. In accordance with other studies, our results show that BMI is a poorer predictor of death than the other measures 
. These results underscore the advantage of assessing body configuration rather than body weight when estimating mortality risk. Furthermore, when controlling for waist circumference, increasing hip circumference appears to be protective in both genders. In our study, obesity emerged as a more important risk factor among young people, in comparison to older. This is in coherence with earlier studies 
In all parts of our analysis, BMI showed weaker associations with both all cause mortality and CVD mortality, when compared to waist circumference, WHR and WHtR. Furthermore, BMI was the only among these four measures which failed to show a statistically significant association with all cause mortality. BMI also contributed less additional information to the prediction models studied ( and ), and offered poorer fitting models (). Hence, BMI seems to be a poorer indicator of disease risk than the other measures studied, being superior only to hip circumference. When adjusting for WHR, BMI seemed even less predictive, while adjusting for BMI had no effect on WHR mortality associations. This emphasises the superiority of the alternative measures over BMI as indicators of CVD risk.
Waist circumference proved to be a statistically significant risk factor in all analyses, but still showed weaker associations with mortality than both WHR and WHtR. In particular, it emerged as a strong risk factor when adjusting for hip circumference. This underlines the significance of considering body configuration rather than the abdominal girth alone.
Hip circumference showed a weak positive association with mortality. However, when adjusting for waist circumference, it proved to be inversely related to CVD mortality in both genders. This finding is in accordance with previous research 
Both in the presence () and absence of age stratification ( and ), WHR turned out to be a stronger risk factor than WHtR among women, whilst the two measures had similar predictive power among men. This gender difference favoured the use of WHtR among younger men. In any case, our results show that WHR and WHtR are superior to the other measures in relation to prediction of mortality.
Based on the IDIs ( and ), WHR and WHtR offered the greatest improvement to our prediction models, followed closely by waist circumference for women. The improvement was in the range of 2–5‰. In comparison, smoking and systolic blood pressure produced IDIs in the range of 5–6‰ for men and 1.7–2.5‰ for women, using the same models. Some discrepancies was noted between the IDIs and the NRIs (e.g. a negative NRI 2 for WHtR among men, ). This can be explained by the choice of cut-points in combination with low precision of the NRI estimates. The variation in NRIs highlights the importance of careful selection of cut-points, depending on the purpose. Identification of optimal cut-points depends on chosen background factors as well as considerations related to clinical relevance. Our results indicate that the best discrimination is obtained by use of waist circumference, WHR or WHtR.
The main strength of our investigation lies in the prospective and comprehensive nature of the HUNT 2 study, its good participation rates, and it being fairly representative for the entire Norwegian nation. The fact that the HUNT population is ethnically homogenous may also be considered a strength in this context, since ethnic differences (genetic and epigenetic factors) may influence the predictive properties of anthropometric measures 
The HUNT 2 database lacks comprehensive information on the participants' dietary habits and cancer history. However, the exclusion of participants with BMI <18.5 kg/m2 and the sensitivity analysis which excludes the first three years of follow-up minimise the potential for confounding by cancer. Our sensitivity analysis indicates that the impact of other potential confounders is minimal.
Our study adds further knowledge to the evidence that BMI is not the most appropriate measure of obesity in everyday clinical practice. WHR is as easy to calculate as BMI and is presently better documented than WHtR. It therefore appears reasonable to recommend WHR as the primary measure of body composition and obesity, at least when it comes to assessing risk of CVD. There is, however, need for further clarification before determining whether WHtR should be considered an even better alternative than WHR. Single (waist circumference in isolation) or additional measures (involving weight and/or height) may also be added to nuance estimations of CVD risk when indicated, for instance in relation to clearly obese or under-weight individuals with a favourable WHR. A certain weakness of the approach suggested here is the documented, inter-personal variance in measurement of waist and hip circumferences 
. This problem can be addressed by standardised measurement procedures 
and adequate training 
It is hard to determine how much effort should be put into training healthcare workers to measure WHR or WHtR in a standardised and reproducible manner, as the potential for predictive improvement will depend on the selected cut-off points and also the choice of prediction model. In relation to combined risk algorithms 
, our results indicate that a NRI of up to 4% might be reached for women and 1.5% for men, depending on cut-off points, by replacing BMI with waist circumference, WHR or WHtR. Identification of the most appropriate cut-offs for a given prediction model could eventually be addressed in a future study. Most preventive CVD guidelines 
however do not include markers of obesity in their combined risk algorithms. Authoritative guidelines currently treat body composition/configuration as an isolated risk factor and usually lack clear specifications regarding the numerical impact on disease risk. As long as this approach is recommended for use in clinical practice, we argue for the use of the anthropometric measure with the best predictive properties. From this perspective, it appears rational to replace BMI by WHR or WHtR when evaluating CVD mortality risk.