This review shows that the estimation of AM due to excess weight is methodologically inconsistent and remains confined to a limited set of countries (United States, Canada, Mexico, Chile, Germany, Italy, Spain, England and Wales, The Netherlands, China, Iran, Indonesia, Taiwan, New Zealand, and the European Union as a whole). Studies spanned to 14 countries and covered nearly all continents except Africa. However, most studies originated from the U.S., which had five estimates. It should also be highlighted that the estimates covered only 7% of all countries. This underscores a gap in global evidence. Estimates are available since 1985, with the first publication dating from 1999 [16], showing that the characterization of the obesity epidemic and its burden has been of interest for a considerable period. Despite this, given the importance of this quantification, relatively few studies have been performed, as evidenced by the 23 studies included in this review. This is further supported by comparisons with systematic reviews that summarize AM estimates for other risk factors, such as tobacco, where the number of included articles exceeds 50 [37].

No clear pattern was observed in the methodology used to estimate AM to excess weight. This aspect creates challenges in comparing the impact of excess weight on mortality across populations, countries and over the time. Nevertheless, methodological differences in estimating AM are not exclusive to excess weight. Previous reviews have also shown that such differences appear when estimating AM related to tobacco consumption and SHS [37, 38]. In fact, some of the methodological inconsistencies identified in this review, such as the variation on causes of death and the relative risks (RR) used, are consistent with those reported in these reviews.

Overall, the main methodological constraints were that studies focused on applying prevalence data contemporaneous with mortality (20 studies), estimating excess weight AM referring to all-cause mortality (nine studies), and risk estimates derived from meta-analysis (14 studies). In most cases, (ten studies) the RR used were not related to the population under study.

Absence of a Lag Time Considered

When evaluating the methodology of the studies included in this review, a key aspect to consider is the failure to account for a lag time between the exposure (excess weight) and the outcome (mortality) in the estimation of AM. In most cases, mortality data are contemporaneous with prevalence data, without incorporating a temporal delay between exposure and outcome. Only three studies considered a latency period of 10 to 15 years [21, 22, 25], all of which estimated the impact of excess weight specifically on cancer-related mortality. Conversely, several authors have hypothesized the existence of a specific lag period required for the development of cardiovascular diseases in individuals with obesity [39, 40]. For instance, a 10 years lapse is proposed for hypertension [41]. In this review, nine studies assessed AM related to cardiovascular diseases. In those studies, mortality may have been overestimated because they did not include a latency period. In general, the lack of consideration of lag times should be noted, since mortality could be overestimated in countries where obesity prevalence is increasing.

Differences in Causes of Death Considered

In this review, the most common approach to estimate AM due to excess weight was based on all-cause mortality analysis. All-cause mortality includes deaths from, for example, infectious diseases that are not related to obesity. This approach may overestimate AM, as it includes deaths not causally associated to excess weight [12]. Nevertheless, all-cause mortality may be the only feasible option in countries with low-quality mortality records. It has been estimated that in 2019, only 54 countries worldwide had complete mortality records and high-quality cause of death registration, meaning that only 23% of deaths globally were comprehensively recorded. Furthermore, there were significant differences depending on the region. While in the Americas and Europe more than 90% of deaths were recorded with their causes, in the Eastern Mediterranean and Africa these percentages dropped to 27% and 8%, respectively [42]. In relation to causes of death associated with excess weight, the IARC reported in 2002 the types of cancer associated with overweight and obesity, ruling out those with no association [4]. These results were confirmed in the 2018 IARC Handbook (Volume 16) [43]. Incorporating this evidence is essential for obtaining valid estimates of AM to excess weight. In this review, 14 studies estimated AM to cancer, seven of which made the estimation for obesity-related cancers.

The association between diabetes and obesity is well-established; however, the inclusion of diabetes mellitus as a cause of death in AM estimates was considered in only four studies [26, 30, 32, 33]. The exclusion of diabetes from the AM estimation leads to an underestimation of the mortality burden. It is important to highlight that physicians tend to omit diabetes on death certificates if unaware of a prior diagnosis, potentially leading to underestimation of AM to excess weight. For example, in one study, diabetes was recorded as the underlying cause of death for only 10% of people living with diabetes and was referenced elsewhere on the certificate for 39% [44].

Selection of the Relative Risks and Possible Confounders

In the AM analysis included in this review, the use of RR that correspond to the population under study was limited. Most studies instead relied on meta-analysis combining cohort and case-control data from multiple countries. According to Flegal et al., the utilization of these RR may result in the presence of bias, given that they do not precisely replicate the risks that would be observed in the population under study [45]. Even so, only a few of the included studies explicitly identify this as a limitation [23, 24, 28]. Availability of country-specific mortality risks associated with excess weight remains a challenge, as long-term cohort or large case-control studies are not always available in most countries. This explains the use of RR from meta-analysis in these cases, as it can provide precise estimates if all available evidence is compiled. Indeed, the estimation of AM to tobacco consumption often relies on meta-analysis-derived RR, even when cohort-derived data are available [12, 46]. In the case of applying risks derived from meta-analysis, it would be advisable to consider the populations characteristics and, to meta-analyze the results separately by geographical regions with similar trends in the obesity epidemic.

A further consideration regarding RR, is confounder adjustment, which varies across the scientific literature. A potential confounder is the presence of pre-existing chronic diseases at baseline, which may contribute to unintentional weight loss and increase the risk of mortality [34, 47]. To address this potential bias, some authors have proposed excluding individuals living with such conditions at study inception [48]. However, it has also been acknowledged that, given the complex interplay between weight status, disease and mortality, the possibility of residual confounding cannot be ruled out [18]. Smoking represents a debated confounding factor in the estimation of AM related to excess weight. The considerable independent mortality risk conferred by tobacco consumption complicates the accurate quantification of the mortality burden attributable to excess weight [49]. Among the studies included in this review, eight explicitly reported using RR adjusted for smoking status [14, 16, 18, 28, 33,34,35]. Furthermore, some authors have observed that the risk of death associated with obesity in smokers is lower than in non-smokers [20, 50]. This may be partly explained by the fact that smokers tend to increase their energy expenditure by approximately 10%, resulting in weight loss and a lower BMI [20, 50]. Indeed, a review suggested that, in the U.S., smokers weigh on average 4 to 5 kg less than non-smokers [50]. These differences have important implications for the estimation of the prevalence and risks, impacting the PAF. For instance, Stokes et al. reported a PAF of 31.9% attributable to high BMI among never-smokers in a U.S. cohort. In contrast, the proportion was 11.3% among current smokers. The authors concluded that the lower RR observed in smokers largely explained the substantially reduced PAF in this group. This finding was also attributed to the dominant risk posed by tobacco consumption itself, which independently accounts for a substantial number of deaths [49].

Physical activity is another relevant confounder, as active individuals are likely to have lower BMI and less adiposity [51]. Other confounders not commonly considered may exist, including recent unintentional or long-term intentional weight loss and duration of overweight or obesity [34].

Importance of the BMI Cut-Off Points and Source of Data to Estimate Prevalence

An important methodological issue that affects the comparability of results across studies is the variation in the BMI cut-off points used to estimate the prevalence of overweight and obesity. While some studies report AM estimates separately for overweight and obesity [22], others combine both categories [27, 30, 31], and another’s considers only obesity [20, 33, 35], with further differences in the obesity grades [9, 14, 16, 17, 19, 24]. These inconsistencies complicate the interpretation and comparison of findings across the existing literature. Another relevant aspect is debated principally among the Asian studies [33]. Evidence indicates that the average BMI in South and East Asian countries is generally lower than that observed in North America and Europe [52]. Furthermore, various studies indicate an increased risk of metabolic diseases, such as diabetes, from a BMI of 23 kg/m2, as well as an increased risk of mortality from cardiovascular diseases from 25 kg/m2 in Asian populations [53, 54]. This suggests that Asian populations may have an increased risk of diabetes and cardiovascular diseases at lower BMI levels than non-Asian populations. This increased risk is hypothesized to be related to a higher percentage of body fat at lower BMI values. In addition, differences in BMI across Asian ethnic populations have been documented. For example, Chinese and rural Thai women have values similar to Europeans, whereas this is not observed in other Asian women [52, 55]. These differences suggest that the use of the general cut-off points proposed by the WHO for the population of the Asian region can be inaccurate [53]. Nevertheless, a WHO expert consultation group concluded that the current BMI cut-off point of 25 kg/m² does not provide an adequate basis for assessing risks related to overweight and obesity in many Asian countries, in addition to recognizing the aforementioned differences. Further studies are needed to establish appropriate cut-off points for this population [55].

A critical factor in estimating AM due to excess weight is the source of anthropometric data, as self-reported measures tend to underestimate overweight by 1.8–3.9 and obesity by 0.7–13.4% points [56]. In this review, only half of the studies specified their data source, evenly split between self-reported and measured data. Biases in self-reporting, particularly overreported height and underreported weight among women, lead to underestimated overweight and obesity prevalence and likely an underestimation of the mortality burden attributable to excess weight [56,57,58]. Nevertheless, two issues could be considered: (1) in epidemiology, the use of self-reported weight and height is common, a method both simpler and less expensive than taking objective data. Given these limitations, countries commonly rely on self-reported nationally representative data on weight and height to estimate the prevalence of overweight and obesity. Although measured data would be ideal, this is not commonly available. Self-reported data therefore represents a necessary and appropriate alternative in the absence of measured data. (2) When self-reported data are the only available option, the associated error and bias may be systematic and consistent over time. This consistency may help preserve the validity of temporal comparisons. Undoubtedly, this aspect is relevant when estimating AM to excess weight, since one of the studies included compared the results obtained from self-reported and measured data from European Union countries [34]. This study concluded that self-reported data underestimated the true prevalence of obesity, as it was about 6% lower than when using measured data. Estimates based on self-reported data provide a conservative estimate of the impact of excess weight on population mortality. Nevertheless, it should be noted that the study specified that the measured data was not representative of the EU countries.

Limitations and Strengths

In this review, the bibliographic search was limited to biomedical databases, which may have excluded relevant data not indexed in these sources. However, multiple databases were consulted, resulting in the identification of a substantial number of studies, thereby minimizing the likelihood of missing pertinent information. In addition, the reference lists of the included studies were screened to identify potentially relevant studies not captured by the initial search. Furthermore, no language restrictions were applied to the search strategy, and studies published in English, Spanish and Portuguese were also considered for inclusion in this review.

The quality of the studies was evaluated using STREAMS-P, a validated tool specifically designed to assess AM studies. Although this tool does not provide a global quality score for each study, this should not be considered a limitation. On the contrary, in the context of scrutinizing the methodology of AM studies, this represents a strength. It allows for the identification and detailed assessment of each domain by addressing the key components relevant to estimating AM.

A major strength is that a thorough peer review of the studies was conducted independently, with the involvement of researchers who have expertise in the field of AM to different risk factors. This approach also facilitated the methodological evaluation of the studies.

Conclusion

The estimation of attributable mortality constitutes a highly pertinent measure for decision-making and raising awareness regarding prevalent risk factors such as overweight and obesity. When it comes to overweight and obesity, estimations remain restricted to a limited number of countries. This review highlights important methodological considerations that should be taken into account and may serve as a starting point for the scientific community and decision-makers to reach a consensus on these issues.

In our view, the selection of causes of death should, whenever possible, be those with evidence of association with excess weight, such as obesity-related cancers, cardiovascular diseases, and metabolic disorders, while avoiding the use of an exclusively all-cause mortality approach when feasible. This last approach might overestimate the actual burden of disease. Lag times should be considered as mortality could be overestimated, especially in regions where obesity prevalence is increasing. The relative risks used should be derived from the population under study. The WHO BMI cut-off points should be used to permit comparability, including the full categorization provided by this organization (overweight, obesity, and its grades). Finally, measured data should be favoured over self-reported data. Likewise, by standardizing these factors, future studies can minimize bias, enhance comparability, and provide more accurate assessments of the mortality burden attributable to excess weight across diverse populations. Nevertheless, these represent ideal conditions for the estimation of attributable mortality. When appropriately acknowledged, the absence of such data should not be considered a barrier to estimating attributable mortality.

Given that attribution studies are not commonly performed and only a small proportion of countries currently have available estimates, this lack represents an important gap in knowledge. Increasing the number of estimates available worldwide, ideally comparable, would improve knowledge of the burden of overweight and obesity.