Overall findings
This analysis highlighted a number of issues pertinent to women’s health; (1) many women gained significant amounts of weight between 20 years and middle age; (2) overweight and obesity at 20 years increased risk of overall and premature mortality even after adjustment for BMI in later adulthood; and (3) weight gain in adulthood after being underweight/healthy weight at 20 years also increased overall and premature mortality.
Our finding that BMI in young adulthood is positively correlated with all-cause mortality and the effect is independent of later BMI mirrors that of other large cohort studies that have adjusted for later BMI [10,11,12].
Our analyses suggest that women who are underweight/healthy weight at 20 years should remain in that category or increase at most to overweight, and our results concur with other large cohorts from America and Australia [10, 18,19,20,21].
The increased all-cause mortality risk seen here amongst the minority of women (1.6%) who moved to a lower BMI category is likely due to reverse causality, whereby a proportion of deaths are likely to be due to medical conditions which also resulted in weight loss preceding death as previously reported [22].
The present analysis indicated that the higher rates of premature mortality amongst women with obesity at 20 years and those experiencing an upward shift in BMI category from 20 years to mid-late adulthood are primarily driven by diabetes and CVD. Other studies have reported similar findings with respect to CVD [11, 19, 23], while other studies have shown a higher mortality rate from obesity-related cancers amongst these groups [24, 25].
Strengths, limitations, and public health implications
The strengths of this study include the large sample size with a broad spread of deprivation scores and data on a number of confounders, minimal missing data and 421,359 women years of complete follow up data. We employed multiple imputation to reduce bias caused by missing data, and sensitivity analyses gave consistent results.
Our study encountered several limitations that warrant discussion. Firstly, there are limitations on the generalisability of the present results to the UK population. The PROCAS study recruited only 4.4% non-white participants when 9.8% of the population in North West England were non-white according to the 2011 census, though a lower proportion within the 46–73 year-old age group will be non-white as the non-white population of England and Wales has a lower average age [26]. Despite the broad spread of deprivation scores, the PROCAS cohort was less deprived than the general Greater Manchester population [27], a common issue in health research [28]. Thirty-eight percent of PROCAS participants had a healthy BMI compared with only 21.9% of women age 55–64 years in the Health Survey for England in 2012, the middle year of recruitment to PROCAS [29]. The lower rate of obesity in PROCAS compared to the Health Survey for England suggests that participants were more health conscious than the overall population and/or self-reported heights and weights were over/underestimated. It is recognised that women with overweight and obesity are more likely to both underestimate their weight, and overestimate their height [30]. However, assessing BMI at 20 years in this cohort by using recalled weight has previously been found to be reliable so this should have had minimal effect on the study outcomes [9]. Self-report is also likely to have resulted in underreporting of alcohol and PA data thus it is likely that fewer women adhered to recommendations than reported here. Women in this study were born before 1970 when obesity at 20 years was uncommon. Only 2.5% of the PROCAS sample had obesity at 20 years compared to 37% of current 16–24 year-old females in England [6]. Thus, analyses including this group had wide confidence intervals. Our findings indicate that women of breast screening age with obesity at 20 years are more likely to die prematurely than those with BMI < 25 kg/m2 at 20 years. However, we also note there is potential selection bias such as having fewer women with obesity at 20 years recruited to PROCAS as they were already unwell, or had died before they were recruited at age 46–73 years. In addition to the limitations on generalisability to the UK population, the present results may not be relevant to lower income countries.
Secondly, there are limitations concerning the robustness of the data collection. We excluded 8 385 women from the analyses due to missing BMI because height and/or weight were not completed on the questionnaire. The higher mortality rate amongst women not included in the analysis due to missing BMI could have been caused by a higher proportion of overweight/obesity amongst those not included, or due to the higher deprivation amongst women not included (which could have caused higher mortality directly, or indirectly via higher overweight/obesity levels). Therefore, the associations between weight gain and maintaining overweight/obesity seen here are likely to be stronger if all cases had been included. The analysis was based on self-reported weight at two time points and does not contain any data for weight changes in the intervening years. We also lack information on weight changes between PROCAS entry and the censor date of June 2020. More frequent objective measurement of weight would help determine whether total years spent with overweight/obesity should be included in mortality risk calculation models in a similar manner to the cumulative damage done by cigarette smoking and the concept of “pack years”. This would also help clarify the importance of age at weight gain/loss as some studies suggest that weight gain in early adulthood is more harmful than in later life [10, 19], whereas weight loss has been shown to be detrimental to mortality if it occurs in later life [19]. The questionnaire only collected weight data at age 20 years therefore no comment can be made on other changes between age 20 years and PROCAS entry such as change in alcohol intake or PA levels. Other confounders such as education level, marital status, dietary pattern, and other health problems could mediate the relationship between weight change and all-cause mortality. We did not collect this data thus could not adjust for these factors. In addition, we lack smoking data. Smoking is a major confounder as it is associated with lower weight, whilst increasing risk of mortality [22]. Twenty percent of women in North West England in 2011 were current smokers [31] so this could have impacted on the results by increasing the mortality rate amongst women in the lower weight categories. Deaths were determined through NBSS which is updated for women on the PROCAS study unless they transfer into another area’s NHS breast screening programme. Data on the proportion of 50–70 year-old women in Manchester and Trafford that move out of area annually suggest that up to 18% of this population could have left during the follow-up period so data on missing deaths could be significant [32, 33]. The cause of death analysis using death certificates was limited to a small number of available death certificates due to cost. This analysis was also limited by incomplete and potentially inaccurate reporting of cause of death. Unfortunately, several of those in our study did not follow UK guidelines [34] leading to inaccuracies in the classifications used here. For example, five cases were listed as metastatic carcinoma with no primary cause listed thus were classed as cancer deaths whereas they may have been cancers generally associated with obesity. Obesity is largely associated with type 2 but not type 1 diabetes. However, the type of diabetes was only specified in 4/5 cases, so we could not separate out weight-related cases. We have not distinguished between pre and post-menopausal BC. Only post-menopausal BC is associated with obesity thus some premenopausal BC cases in our analyses could be incorrectly classed as obesity-related cancers.
Thirdly, there are limitations concerning the analysis techniques. We have used multiple imputation to deal with missing data however this relies on data being missing at random, whereas there may be a bias to missing weight data amongst women with higher weights. We did not attempt to minimise impact of reverse causation unlike other studies which have excluded deaths within the first few years of follow-up (e.g. [24]).
Finally, it is important to note that the PROCAS study is still in follow up and the mortality figures presented here are not final, with 81% of the cohort < 75 years and still alive at the time of last follow-up hence the majority of deaths so far recorded are premature. Future research could repeat the current analysis when more of the cohort are deceased, or perform other analyses such as population attributable fraction which would estimate the burden of excess deaths related to overweight and obesity at age 20, and weight gain during adulthood.
Whilst analyses of the PROCAS cohort has previously shown that being overweight or obese at 20 years reduces risk of later BC [9], the current results together with those from other cohorts indicate higher weight in young adulthood significantly increases overall and premature mortality rates. Hence any potential benefits of obesity in early adulthood must not be overstated. With obesity rates amongst 13–15 year-old girls in the UK now at 18%, this is likely to have a substantial impact on health and healthcare costs in future years [6]. More resources should be channelled into helping children and adolescents avoid weight gain in order to prevent having overweight or obesity by 20 years. Results from this and other cohorts also highlight the benefit of avoiding weight gain during adulthood.
The optimal methods for supporting children, adolescents and adults to avoid weight gain are currently unknown. Future research could design and evaluate interventions with the overall aim of finding effective interventions that could be rolled out. Existing, peer-reviewed frameworks should be used such as guidance from the Medical Research Council on developing complex interventions [35] which includes development or identification of an intervention, assessment of feasibility of the intervention and evaluation design, evaluation of the intervention, and impactful implementation.
