Introduction
Weight loss has been reported to result in several health benefits, such as significant improvements in CVD risk factors (blood pressure, lipid profiles, glucose tolerance)(Reference Lee and Paffenbarger1, Reference Van Gaal, Wauters and De Leeuw2). It may therefore be reasonable to expect that weight loss would lead to decreased mortality in the long term. Indeed, this seems to be the case in obese individuals with serious medical complications(Reference Goldstein3, Reference Williamson, Thompson and Thun4) or when substantial weight loss has followed surgical procedures(Reference Sjöström, Narbro and Sjöström5). However, the long-term effects of more moderate degrees of weight loss for those who are not severely obese and do not have co-morbidities are unclear. Many prospective studies show conflicting results, while some recent studies indicate either excess(Reference Diaz, Mainous and Everett6–Reference Sauvaget, Ramadas and Thomas9) or unchanged mortality(Reference Wannamethee, Shaper and Lennon10) following weight loss. Reviews of the data suggest that inconsistent results might be due to failure to control for known confounding factors (for example, underlying disease, intention to lose weight)(Reference Eilat-Adar, Goldbourt and Resnick11, Reference Nilsson12) while also noting that many of the existing studies were not specifically designed to test the hypothesis that weight loss increases or decreases relative risk (RR) of all-cause mortality(Reference Sørensen13, Reference Yang, Fontaine and Wang14). Methodological problems have also been identified, for example, the method by which the weight loss was achieved has usually not been reported (although dietary energy restriction is likely to have been a major factor), while weight changes before and after the recording periods have usually not been determined(Reference Nilsson12–Reference Simonsen, Hundrup and Obel16).
In light of the current obesity epidemic and the resulting focus on encouraging those with BMI above 25 kg/m2 to lose weight by changing their diet and lifestyle(17), it is important to establish whether the long-term effects of weight loss benefit life expectancy. The current advice from the UK Department of Health's Obesity Care Programme is for those who are overweight or obese to reduce energy intake and increase physical activity as a method of lifestyle modification(18). Further treatment and advice may need to consider a broad spectrum of evidence so as not to rule out potential investigations that identify subgroups of patients, or certain conditions, where weight loss may be detrimental to health and increase mortality(Reference Astrup19).
The aim of the present study was to examine the available evidence of the impact of weight loss, as a lifestyle intervention, on the RR of all-cause mortality and to quantify this using meta-analysis. Data were pooled in a number of different ways in order to examine the influence of a number of possible confounders. Meta-analysis was used to provide a more objective appraisal of the evidence, integrating data from multiple prospective cohort studies to increase the power and precision of estimates of effect and reducing the likelihood of false negative results(Reference Egger and Smith20, Reference Egger, Smith and Phillips21).
Methods
Search strategy
A literature search was carried out independently by two investigators to identify prospective cohort studies that evaluated the effect of weight loss as a lifestyle intervention on mortality risk. A web search was undertaken on PubMed/Medline and ScienceDirect databases. Articles published between 1987 and 2008 and in the English language were included. Search terms included ‘weight, BMI, loss, change, mortality, intentional, unintentional, relative risk, prospective and cohort’. Identified citations and abstracts were obtained from journals, libraries or authors. A hand-search of the bibliographies of retrieved papers and linked articles was also carried out.
Data selection
Inclusion criteria were prospective studies in English of adults (men and/or women) with data on body weight and weight loss over more than 1 year. Studies needed to present RR of mortality and associated 95 % CI for the group that lost weight relative to a comparable reference group who lost minimal or no weight. Drug treatment studies and studies that measured weight loss following bariatric surgery were excluded, as the aim was to assess the effect of lifestyle interventions. Twenty-six publications were identified that met the inclusion criteria. Data on RR of mortality and 95 % confidence limits were extracted for all subgroups presented by the authors (for example, men and women, intentional v. unintentional weight loss, obese v. overweight).
Data analysis
Meta-analysis was performed using Comprehensive Meta-analysis software (CMA version 2; Biostat Inc., Englewood, NJ, USA). Moderator variables such as baseline BMI (normal, overweight, obese), reason for weight loss (intentional, unintentional), baseline health status (healthy, unhealthy), method used to estimate weight loss (measured weight loss, reported weight loss) and physical activity adjustment (adjusted data, unadjusted data) were used to classify subgroups for separate analysis. For the subgroup analysis based on baseline BMI the ranges used in papers generally corresponded to those recommended by WHO(22). Analysis was carried out using adjusted data because papers gave insufficient data on CI for unadjusted data. Although multivariable adjustment of the data varied from study to study, all adjusted for smoking. Results are shown in the form of schematic plots (Forest plots), which illustrate the size and direction of effect for each study and the weighted effect of all studies combined, with 95 % (lower and upper) CI. Meta-analysis uses a weighted average of the results, in which the larger and more precise studies have more influence than the smaller ones. Results are shown for the random effects model, which assumes the underlying effect may vary for each population. This is the most appropriate model where heterogeneity is present(Reference Egger and Smith20, Reference Egger, Smith and Phillips21). Statistical significance of the overall pooled effect was based on P < 0·05.
Results
Study characteristics
Table 1 shows a summary of the characteristics of the study populations and subgroups. Sample sizes ranged from 34 to 5008 subjects and the majority of the data was collected from white populations of US and UK origin. All of the studies were designed to investigate RR of mortality and weight change. The stage of life during which weight change occurred varied between adulthood, middle age and old age and the follow-up period ranged from 2 to 20 years.
Prospective studies of weight loss and mortality*
NHANES, National Health and Nutrition Examination Survey; NIDDM, non-insulin-dependent diabetes mellitus; NHIS, National Health Interview Survey; DOM, Diagnostisch Onderzoek Mammacarcinoom (Diagnostic Investigation into Breast Cancer); IQR, interquartile range.
* 1 lb = 0·4536 kg.
Quantitative data synthesis
Owing to the acknowledged importance of whether weight loss is intended or not, results are presented for (a) intentional, (b) unintentional and (c) weight loss not specified. For the main category of interest, i.e. intentional weight loss, sub-analyses are given for healthy v. unhealthy subjects. These have then been further analysed to examine the influence of moderators and confounders.
Intentional weight loss
Figure 1 shows the RR of all-cause mortality in relation to intentional weight loss. Overall, there was no significant effect (RR 1·01 (95 % CI 0·93, 1·09); P = 0·89). However, among healthy subjects, RR was increased 11 % by weight loss (RR 1·11 (95 % CI 1·00, 1·22); P = 0·05), whereas it was reduced in unhealthy subjects by a similar amount (RR 0·87 (95 % CI 0·77, 0·99); P = 0·028).
Mortality risk for intentional weight loss according to health status. RR, relative risk.
Unintentional weight loss
Unintentional weight loss was associated with higher mortality (RR 1·22 (95 % CI 1·09, 1·37); P = 0·001) (Fig. 2), as has been shown in other studies. Unintentional weight loss is usually considered an indicator of pre-existing or silent disease and this group was not considered further.
Mortality risk for unintentional weight loss according to health status. RR, relative risk.
Unknown or unspecified cause of weight loss
Where the cause of weight loss was unspecified, there was also excess mortality (RR 1·39 (95 % CI 1·29, 1·51); P < 0·001) (Fig. 3). Most of these studies were on ‘healthy’ subjects, but the subgroup who were unhealthy had even higher mortality associated with weight loss (RR 1·75 (95 % CI 1·24, 2·46); P = 0·001). Studies where weight loss intention was not explored may suffer from the same problem of confounding by illness as those in which weight loss was unintentional. The remaining analyses were all performed using studies of intentional weight loss only.
Mortality risk for weight loss (intention unknown) according to health status. RR, relative risk.
Subgroup analyses of intentional weight loss
Relative weight at baseline
Weight loss appeared to benefit obese weight losers who were also classified as unhealthy at baseline (RR 0·84 (95 % CI 0·73, 0·97); P = 0·018) but had no benefit for healthy obese (RR 1·02). Overall, there was no change in risk for the obese group (RR 0·94 (95 % CI 0·86, 1·04); P = 0·002) (Fig. 4). For intentional weight losers whose baseline BMI was within the normal to overweight range, or for mixed-weight populations, the RR of mortality was increased (RR 1·09 (95 % CI 1·02, 1·17); P = 0·008) (Fig. 5).
Mortality risk for intentional weight loss among obese adults. RR, relative risk.
Mortality risk for intentional weight loss among overweight or mixed populations. RR, relative risk.
Method of assessing weight loss
The majority of study groups with data on intentional weight loss (fifteen out of eighteen studies) relied on reported measurements of weight or weight loss. Among these, RR associated with weight loss was near unity. However, the three study groups with actual measurement had a net RR of 1·28 (95 % CI 1·07, 1·53) (Fig. 6).
Mortality risk for intentional weight loss according to weight loss assessment method. RR, relative risk.
Physical activity adjustment
Adjustment for physical activity was made in most studies (fourteen out of eighteen studies) but there was essentially no difference in the RR according to whether the models had adjusted for activity or not (RR 0·98 v. 1·01 where adjusted for physical activity) (Fig. 7).
Mortality risk for intentional weight loss according to adjustment for physical activity. RR, relative risk.
Discussion
Main findings
Meta-analysis was used to explore the effect of weight loss on mortality using sensitivity and subgroup analysis to explore some of the likely causes of heterogeneity, especially intentionality, health and baseline BMI. Whereas weight loss for unknown or unspecified reasons was clearly associated with excess mortality, intentional weight loss resulted in virtually no change in mortality overall. Importantly, we found opposing effects among healthy and unhealthy adults and between the obese and those with more moderate degrees of overweight or from the general population. The excess risk of weight loss in healthy adults was estimated to be of the order of 11 %. This was counterbalanced by a benefit of about 13 % among unhealthy adults (i.e. those with diabetes or obesity-related health conditions).
Other studies
The literature is equivocal on the risks and benefits of weight loss(Reference Poobalan, Aucott and Smith15, Reference Simonsen, Hundrup and Obel16). Many prospective studies and reviews appear to show an increased mortality associated with weight loss(Reference Nilsson12), which runs counter to conventional wisdom relating to the adverse effects of obesity and the beneficial changes in risk factors associated with weight loss(Reference Heitmann, Svendsen and Martinussen23). It has been argued that methodological weaknesses explain much of this paradox, including failure to adjust for known confounders(Reference Stampfer24). In particular, it has been claimed that intentionality of weight loss is key(Reference Simonsen, Hundrup and Obel16) but many studies fail to distinguish between intentional and unintentional weight loss, the latter being a cardinal sign of ill health and a predictor of increased mortality in old age(Reference Ryan, Bryant and Eleazer25, Reference Shahar, Shahar and Kahar26).
Some clinical trials have demonstrated beneficial effects of weight loss with regard to morbidity in individuals suffering from either diabetes, obesity-related health conditions, cancer or other diseases(Reference Goldstein3). There are also an increasing number of favourable reports from bariatric surgery, such as the ‘Swedish obese subjects’ (SOS) study which has shown that substantial long-term weight reduction appreciably improves the cardiovascular risk profile of morbidly obese subjects, ultimately resulting in a decrease in overall mortality(Reference Nilsson12). Although such data may be encouraging, their success cannot necessarily be extrapolated to the public health setting where the weight losses normally achieved by diet are modest and difficult to sustain, and the subjects generally less severely obese and with few co-morbidities. Another study, due to report in 2015, will provide valuable additional data. This is the Look AHEAD (Action For Health in Diabetes) clinical trial, which is assessing the long-term effects (up to 11·5 years) of an intensive weight-loss programme delivered over 4 years in overweight and obese individuals with type 2 diabetes.
Interpretation of present analysis
In the present review and meta-analysis, intentional weight loss modestly reduced the risk of all-cause mortality only among the subgroup of unhealthy adults (by approximately 13 %), especially among those who were also obese (by approximately 16 %). All these studies relied on reported estimates of body weight. Self-reporting of body weight may be cheap and easily carried out; however, it is affected by a number of biases. Actual measurement of body weight using appropriate devices is recommended for complete accuracy and reliability of the data(Reference John, Hanke and Grothues27).
Our finding of a marginally increased risk of death among overweight but otherwise healthy adults who lost weight intentionally, if true, has important public health implications. This observation is consistent with recent findings using National Health and Nutrition Examination Survey (NHANES) data that showed that the ideal weight for longevity was the overweight category, or BMI 23–30 kg/m2(Reference Fontaine, Redden and Wang28–Reference Flegal, Graubard and Williamson30).
Why should intentional weight loss have opposing effects in different groups of individuals? One possibility is that obese individuals with risk factors may show a benefit because they are more motivated to make a series of changes such as reducing fat intake or increasing exercise level, and these may lower RR of mortality by benefiting overall health status(Reference Gregg, Gerzoff and Thompson31). Unhealthy individuals are also more likely to be recipients of health care and medical interventions. It is more difficult to explain why intentional weight loss should have an adverse effect among healthy but overweight individuals. More data on method of weight loss, persistence of weight loss and body composition would be helpful in this regard. Weight loss via energy restriction may do little to alter the relative distribution of body fat and may result in decreased lean body mass. A reanalysis of the Framingham Heart Study and the Tecumseh Community Study suggests that weight loss as a result of a reduction in body fat may reduce all-cause mortality while weight loss as a result of a reduction in lean body mass may increase it(Reference Allison, Zannolli and Faith32). Given the significance of fat distribution and the lean body mass:fat ratio in health prognosis(Reference Berentzen and Sørensen33), it is imperative that future studies attempt to measure more than just weight or BMI. Furthermore, studies must adequately disentangle the influence of physical activity and/or fitness, which may influence both body weight and the morbidity and mortality outcomes under study. Most studies did not include assessment of physical activity and those that did used questionnaires rather than physical fitness, which is a stronger predictor of mortality(Reference Warburton, Nicol and Bredin34, Reference Myers, Kaykha and George35). The focus of new research may most usefully be directed to examining survival among those population groups that might be expected to benefit most from weight loss. These include those with diabetes, those with obesity-related conditions (such as hypertension) and certain ethnic groups.
Limitations
The present study inevitably has some limitations. The literature search was carried out using only two databases, but was complemented by thorough checking of cross-references and inclusion of new reviews published in 2008. Limitations of the evidence base include the fact that none of the studies provided information on the method of weight loss, which is relevant because it is not clear if weight loss through energy restriction or increased energy expenditure differentially influences long-term outcomes. Second, weight loss was usually assessed retrospectively and subjectively, often at two time points some distance removed from the ultimate outcome, i.e. death. It is thus difficult to be sure that the weight loss estimate does not represent a transitory phase and that it is representative of a reasonable period of adult life. Third, the studies differed in the statistical treatment of covariates or confounders in adjusted models (for example, some excluded smokers, others adjusted for smoking). These problems are common to all attempts to review and pool data from different studies, and the present results are consistent with other recent reviews that have not used meta-analysis(Reference Simonsen, Hundrup and Obel16, Reference Fontaine and Allison36). Furthermore, using a meta-analysis stratified by intentionality, health and baseline BMI, we were able to quantify effect sizes in different groups. The robustness of intentionality measures has been questioned(Reference Simonsen, Hundrup and Obel16) because it depends on the question asked and may change during the course of the follow-up(Reference Coffey, Gadbury and Fontaine37). The study by Sørensen et al. (Reference Sørensen, Rissanen and Korkeila8) was unusual in assessing intentionality prospectively and also reported the largest effect size (RR 1·87)(Reference Sørensen, Rissanen and Korkeila8). However, as it was of high quality (as judged by Simonsen et al. (Reference Simonsen, Hundrup and Obel16)), we did not consider its exclusion justified in the main analysis. Instead, sensitivity analysis showed that the effect of excluding this paper would be to reduce the RR from 1·11 to 1·09. On balance we think it unlikely that our estimates of higher risk are inflated, since most sources of misclassification and measurement error would tend to result in underestimation of effect (for example, self-reported body weight).
Conclusion
Recently a great emphasis has been placed on weight loss by lifestyle change for everyone who is, even slightly, overweight. However, a review of the available literature, complemented by meta-analysis, suggests that at-risk individuals may benefit, but for healthy overweight individuals intentional weight loss does not decrease mortality and may even increase it. Appropriately designed intervention studies in subgroups differing by age, sex and ethnic group, as well as by risk status, are urgently needed. Until more reliable data are available to demonstrate consistent improvements in survival, the question remains as to whether the correction of obesity per se should have such emphasis as a clinical and public health target.
Acknowledgements
We wish to thank Orlaith McDaid for her contribution to the preparation of this paper.
The present review was funded by The World Sugar Research Organisation. The findings and conclusions are those of the authors and do not necessarily represent the views of the funding organisation.
The authors have no conflicts of interest to declare.
