The European Prospective Investigation Into Cancer and Nutrition (EPIC) is large diet and cancer cohort [32]. The EPIC-Norfolk study participants, aged between 39 and 79 years, were recruited from 35 General Practitioners’ surgeries in the Norfolk area of East Anglia, from 1993 to 1997 [33]. General practice age sex registers act as a population sampling frame as practically all of the UK the population are registered with a general practice through the National Health Service. Many of the characteristics of this cohort at baseline were comparable to the UK national population, including age, sex and anthropometry measurements but this cohort had a lower proportion of current smokers [34].

Ethical approval for the study was given by the Norwich District Health Authority Ethics Committee and all participants gave written, informed consent.

Trained nurses measured participants’ weight and height, with participants wearing light clothing and without shoes. Weight was measured to the nearest 100 g using digital scales (Salter, UK). Height was measured to the nearest 0.1 cm using a free-standing stadiometer. A D-loop non-stretch fibreglass tape was used to measure WC, which was measured at the smallest circumference between the ribs and iliac crest to the nearest 0.1 cm while the participant was standing erect, with the abdomen relaxed, the arms at the side, the feet together and at the end of a normal expiration, without the tape compressing the skin. The measurement was taken at the level of the umbilicus where there was no natural waistline. Body mass index (BMI) is defined as the body mass (weight) divided by the square of the height, and is expressed in kg/m2, and is commonly used to categorise individuals as underweight (< 18.5 kg/m2), normal weight (≥18.5 to < 25 kg/m2), overweight (≥25 to 30 kg/m2), or obese (≥ 30 kg/m2) [4].

Absolute change in WC (cm) was calculated as WC measured at 2HE minus WC at 1HE. Absolute weight change (kg) was calculated as weight measured at 2HE minus weight measured at 1HE. Participants were assigned to one of five WC change categories: > 5 cm loss, > 2.5 to ≤5 cm loss, ≤ 2.5 cm loss or gain (‘maintenance’, including zero change; considered the reference category), > 2.5 to ≤5 cm gain, > 5 cm gain. Annual WC change was calculated from the absolute difference in WC, divided by the participants’ time lapse between the health examinations (cm/y).

In order to study the possible association between changes in WC and changes in weight, participants were assigned to one of nine categories, based on a loss of > 2.5 cm or 2.5 kg, within 2.5 cm or 2.5 kg loss or gain (maintenance), or a gain of > 2.5 cm or 2.5 kg respectively. The nine categories were as follows: WTM (weight maintenance) and WCM (WC maintenance) (reference category); WTL (weight loss) and WCL (WC loss); WTL and WCM; WTL and WCG (WC gain); WTM and WCL; WTM and WCG; WTG (weight gain) and WCL; WTG and WCM; and WTG and WCG.

Measurements of WC, weight and height measurements at both time-points were essential for participant inclusion (N = 15,010). Participants were excluded from analyses if they were categorised as underweight (BMI < 18.5 kg/m²) or who self-reported CVD or cancer at either examination (N = 2095); those with missing data on adjustment variables (smoking, social class, educational level, physical activity, and menopausal status in women (N = 529) were also excluded. This left 12,337 participants for analyses (Fig. 1).

Data from two self-administered Health and Lifestyle Questionnaires (HLQ1 and HLQ2), at 1HE and 2HE respectively, were used to create categories of a number of variables. Current smokers were defined as those currently smoking cigarettes, former smoker as being a smoker previously and non-smokers were those who never smoked (derived from HLQ1 and HLQ2). Self-reported physical activity (derived from HLQ1) was classified into four levels, using both occupational and leisure activity: inactive, moderately inactive, moderately active and active [35, 36]. Social class at 1HE was defined according to the Registrar General’s occupation-based classification scheme and categorised into the following six groups: professional, managerial and technical, non-manual skilled, manual skilled, partly skilled and unskilled [37]; for our analyses, these categories were combined to create 2 categories, non-manual (professional, managerial and technical, non-manual skilled) and manual (manual skilled, partly skilled and unskilled). Educational status at 1HE was categorised into four groups: degree or equivalent, A level or equivalent, O level or equivalent and less than O level or no qualification, corresponding to levels within the International Standard Classification of Education 1997 [38]. In our analyses, those with an educational level of O level and above were combined into one category. Menopausal status was categorised as premenopausal, early perimenopausal (< 1 year), late perimenopausal (1–5 years) or postmenopausal (> 5 years). Included adjustment variables were measured with HLQ1, with the exceptions of smoking status and menopausal status, which were measured with HLQ2. Participants were asked about their medical histories at both time-points. The diagnosis of chronic diseases such as heart attack, stroke and cancer were recorded as present when participants answered “yes” to question “Has a doctor ever told you that you have any of the following conditions?”

All fatal events occurring between 1998 and 31st March 2016 were captured by the Office of National Statistics, United Kingdom. Death certificates were coded by nosologists according to the International Classification of Diseases (ICD). An underlying cause of death from cancer or CVD was defined by using the following ICD codes: cancer death (ICD9, 140–208 or ICD10 C00-C97), or CVD death (ICD9 400–438 or ICD10 I10-I79).

Over a 3.5 year period, men had a mean WC and weight increase of 0.83 cm (SD 5.19) and 1.29 kg (SD 3.61) respectively; similar mean increases of 0.81 cm (SD 5.60) in WC and 1.38 kg (SD 4.14) in weight were found in women. The baseline characteristics of those who attended both 1HE and 2HE, before and after exclusions were applied were similar (Additional file 1: Table S1). However, the prevalence of self-reported CVD and the percentages of current smokers and physically inactive participants were higher in those who attended the 1HE only, compared to those who also attended the 2HE. Also, the percentage of deaths that occurred was lower in both men and women, after exclusion criteria were applied, with the highest percentage found in those who only attended the 1HE.

Table 1 describes the characteristics of men and women, by WC change category. WCM was observed in 40% of men and 38% of women. Participants with the greatest WCL were, on average, older and had the largest WC, weight and BMI at 1HE. Men with the greatest WCG were more likely to have a smaller WC at 1HE, whereas women with the greatest WCG had a similar WC at 1HE, compared to those in the WCM category. Current smokers at 2HE were more likely to have a WCL whereas former smokers, a WCG. Male non-manual workers were more likely to have a WCL > 5 cm whereas male manual workers were more likely to have a WCG > 5 cm. Participants with no qualifications were more likely to have a WCG > 5 cm. A greater percentage of deaths occurred in those with a WCL > 5 cm.

At the end of a median (IQR) follow-up time of 16.4 (15.7, 17.2) years, 1551 deaths in men and 1315 deaths in women were recorded, 440 and 382 deaths from CVD, respectively. Total and CVD mortality HRs for men and women by WC change category are shown in Table 2. The addition of change in weight to the model strengthened the positive WCG-mortality associations (model 3), especially in men, where the findings for CVD mortality were stronger than for all-cause mortality, in those with a WCG > 5 cm (HR 1.84 and 1.51 respectively). Women with a WCG > 5 cm had a significantly higher hazard of all-cause mortality of 1.25 (CI 1.06–1.46) compared to WCM. No significant associations were found regarding changes in WC and CVD mortality in women. No significant associations were found for WCL and total or CVD mortality in the multi-variate adjusted models, in either men or women.

HRs for all-cause mortality in men and women, adjusted as per model 3, grouped by confounder categories are shown in Tables 3 and 4 respectively. Participants with a WCG > 5 cm had a higher HR than those in the WCM category, irrespective of their WC categorisation at 1HE, although stronger associations were more evident in men. Men who had a WCG > 5 cm had significantly higher HRs, irrespective of whether they were < 65 years (HR 1.78 CI 1.40–2.26) or > 65 years (HR 1.38 (CI 1.13–1.68) of age at 1HE (Table 3) although a significant association was only found among women > 65 years (HR 1.35 (CI 1.10–1.67) (Table 4). Higher hazards were observed for participants with a WCG > 5 cm, irrespective of smoking status, with the exception of women who were former smokers, where a WCG between 2.5 and 5 cm had a significantly lower HR of 0.62 (CI 0.45–0.86) (Table 4). Irrespective of BMI, physical activity, educational level or social class categorisation at HE1, WCG > 5 cm was associated with a higher HR in both men and women (Tables 3 and 4 respectively). Almost 71% of women were classified as postmenopausal and a higher hazard was found in those with a WCG > 5 cm (Table 4). Even after excluding participants who died within 3 or 5 years of the 2HE, participants with a WCG > 5 cm had significantly higher HRs for all-cause mortality, than those in the WCM category.

As significant findings were evident for CVD mortality and men, but not women, in relation to changes in WC, we further explored these in subgroups, stratified by potential confounding characteristics, adjusted as per model 3 (Additional file 1: Table S2). Generally, a higher hazard was evident for all values of the confounding variables in men with a WCG; findings relating to WCL were inconsistent. In men who had a WCG > 5 cm, the strongest associations were found in those who were younger, current smokers, overweight, physically active, had no qualifications, had a manual job and a normal WC. Even after excluding participants who died within 3 or 5 years of the 2HE, men with a WCG had significantly higher HRs for CVD mortality, than those in the WCM category.

Weight change was strongly positively correlated with a change in WC in both men (r = 0.60) and women (r = 0.57) (P < 0.001). We also explored the association of category changes in WC and total mortality in combination with category changes in weight; the interaction between WC change categories and weight change categories was not found to be significant (p = 0.29 for men and 0.82 for women). No significant interaction was found between WC change categories and weight change categories with regard to CVD mortality (p = 0.71 for men and 0.52 for women).

Figure 2a illustrates the associations between category changes in WC and weight and total mortality, adjusted for age, BMI, baseline WC, physical activity, smoking, educational level, social class and change in weight, and menopausal status in women. Similar trends in both men and women were generally evident. In men, significantly higher HRs were observed for all WTL categories, irrespective of whether men were categorised as WCL, WCM or WCG, as well as for WCG and WTM (HR 1.37 (CI 1.17–1.61)), compared to those in the WTM and WCM category. In women, significantly higher HRs were observed for total mortality for 2 categories – WTL and WCL (HR 1.40 (CI 1.14–1.70)) and WTL and WCM (HR 1.50 (CI 1.16–1.95)), compared to the reference category of WTM and WCM.

Figure 2b illustrates the similar associations with CVD mortality, adjusted for age, BMI, baseline WC, physical activity, smoking, educational level, social class and change in weight, and menopausal status in women. In men, significantly higher HRs were found in the WTL and WCL (HR 1.81 (CI 1.32–2.49)), WTL and WCG (HR 2.22 (CI 1.03–4.82)), and WTM and WCG (HR 1.38 (CI 1.02–1.88)) categories, compared to the reference category of WTM and WCM. A significant association for CVD mortality in women was only found in the WTL and WCM category (HR 1.81 (CI 1.15–2.85)). In all analyses, the category with the lowest HR was WTG and WCL.

In this study of 12,337 middle-aged and elderly men and women, a WCG of more than 5 cm over approximately 4 years was associated with higher total mortality in both men and women and higher CVD mortality in men over the next 16 years of follow-up. Particularly among men, the associations were also evident after stratification for age, smoking, BMI, physical activity, educational level, social class and after the exclusion of deaths within the first 5 years of follow-up. Associations for WCL and mortality were inconsistent.

The major strengths of this study include its large population of free-living, middle-aged and elderly men and women, its prospective design, and the long follow-up time. Additionally, WC, weight and height were objectively measured, not self-reported, and information on a large number of factors associated with WC and weight was available. To minimize the possibility of reverse causality, that is, participants who were sub-clinically ill were more likely to lose weight and WC which would have resulted in a higher mortality risk; participants were excluded with self-reported cancer or CVD at both time-points, as were those who had a BMI < 18.5 kg/m². Those who died within the first 3 and 5 years were also excluded in our subgroup analyses. Nevertheless, we cannot completely rule out the possibility of some effect of reverse causality and/or residual confounding.

The main limitations of our study include self-reported disease history, healthy volunteer bias and attrition. The prevalence of self-reported CVD was lower in those who also attended the 2HE, as was the percentage of current smokers and physically inactive participants, which may be indicative of healthier individuals being more likely to return for a follow up assessment (Additional file 1: Table 1). Additionally, the percentage of deaths that occurred was lower in both men and women, after exclusion criteria were applied; the percentage of deaths was highest in those who only attended the 1HE. It is possible that some individuals in our study had undiagnosed/sub-clinical conditions such as digestive disorders, or various psychosocial diseases, such as depression, that they did not report, resulting in weight and potentially WC loss and subsequent death, and possibly an overestimation of the association [39]. The more frail participants may have been excluded and/or have not returned for the 2HC and therefore be under-represented in our analyses. However, it is also conceivable that some of the participants in the weight and WC loss categories were pre-frail at 2HE. A further limitation is the inability to account for all behaviour changes during follow-up, including physical activity, which may have had effects on the observed associations.

The highest HRs for both total and CVD mortality in men, which were statistically significant, were observed in the WTL and WCG category, although this was a small category, consisting of only 35 participants and 19 events. In women, the highest HR was found in the WTL and WCM category, for both total and CVD mortality. The category with the lowest HR in both men and women, for both total and CVD mortality was the weight gain and WC loss category, although this was not statistically significant. The findings that WTL, in combination with changes in WC, was associated with the highest HRs, agrees with our previous study, which found that weight loss of more than 2.5 kg over approximately 4 years is associated with a higher mortality over 15 years of follow-up [31] in this middle-aged and elderly population.

Our findings that WCG is associated with higher all-cause mortality are in agreement with a number of previous publications. A recent pooled sample of 2492 Danish and Swedish women found that WCG over 6 years was associated with increased total and CVD mortality [23]. We did not find significant associations with WCG and CVD mortality in women; statistical power was more limited due to the lower number of CVD deaths in women in our study. In the study of Danish and Swedish women, associations were particularly strong for women with normal weight and for ever-smokers. In our analyses, we found that WCG > 5 cm had a similar HR for all BMI categories and only never smokers had a statistically higher HR. The differences in findings may be due to the characteristics of the 2 cohorts; our cohort of women was older, had a higher mean BMI and a lower percentage of ever-smokers. A prospective study of 26,625 healthy middle-aged men and women has also shown that increases in WC over a 5.3 year period were positively associated with mortality in men and women over 6.7 years of follow-up; the HR was 1.09 (1.02, 1,16) per 5 cm for both men and women combined, after adjusting for baseline BMI, WC and change in BMI [24].

However, results from the Melbourne Collaborative Cohort Study of 21,298 men and women aged 40–69 years at baseline found that WCL over approximately 12 years was associated with an increased risk of all-cause mortality (HR: 1.26; 95% CI: 1.09–1.47) over 7.7 years of follow-up, especially in older adults, compared to those who had minimal changes, but did not find an association with WCG [25]. It is possible that the shorter time period of less than 4 years between our WC measurements may partially explain our lack of associations with WCL and mortality. However, in a study of older adults aged 70–77 years, WCL over a 3 year period (≥3.1 cm) was significantly associated with an all-cause mortality risk over 20 years of follow-up of 1.52 (95% CI: 1.01–2.31), which was comparable to their weight loss finding, in agreement with our previous results [31]; no significant associations were found for WCG and all-cause mortality, or changes in WC and CVD mortality [40]. No significant associations were found with self-reported changes in WC over a 12 years and concurrent mortality, after 9 years of follow-up, in a large, prospective study among 23,254 Swedish women [26], although the cohort was relatively young (median age of 52 years at follow-up), with a relatively low number of deaths (N = 570) during follow-up, of which 79 were attributable to CVD. A smaller study of 1138 older adults, predominantly female, from the Bambuí (Brazil) Cohort Study of Aging [27] did not find any associations with WC change and mortality; the relatively short time interval of 3 years between measurements, in addition to the follow-up period of 8 years may have contributed to the lack of findings. No associations were found for WC change over 3.6 years and mortality in a study of 1805 Iranian men, aged ⩾30 years [28]; the limited number of deaths (N = 88) during the relatively short-term follow-up of 6.6 years resulted in limited power to assess the effect of changes in WC for mortality events.

Weight, particularly in older populations, is an indicator encompassing not just fat but also bone and muscle mass and these different components may relate to health in different directions; thus, weight loss in later life may be an indicator of increasing frailty. Body fat distribution changes with age, with a reduction in subcutaneous fat and an increase in central adiposity [41]. The accumulation of body fat in the abdominal region has been shown to be associated with a number of adverse health outcomes such as diabetes, metabolic syndrome, CVD and all-cause mortality, independently of BMI [12, 42‐44]. Adipose tissue has an important role to play in numerous metabolic and endocrine functions, including the expression and secretion of inflammatory cytokines, such as leptin, adiponectin, and interleukin 6 (IL-6), which are important regulatory factors of energy intake and inflammatory responses [45]. In a study of 20 abdominally obese, older women, hyperinsulinemia was positively associated with adipose IL-6 gene expression, but negatively associated with adipose adiponectin expression [46]. A 20-week weight loss program in obese, older women decreased leptin production in both gluteal and abdominal adipose tissue, but only increased adiponectin production from abdominal adipose tissue [47]. These results highlight the importance that regional adipose tissue hormone/cytokine production may play in mechanisms of metabolic complications associated with abdominal obesity and the importance of improving body fat distribution, through appropriate exercise and diet programs [48].

Obesity is a complex condition, for which there are numerous risk factors, many of which are modifiable [49, 50]. Attention needs to be focussed on various levels of policy strategies for its prevention [51, 52]. Aa a number of studies have shown that WC is increasing at a faster rate than BMI or body weight [53‐55], current recommendations, based on observational cohort studies, including findings from this publication, should advocate prevention of WC gain into adulthood, and weight and WC stability from midlife onwards.