Body mass index (BMI) is an important predictor of mortality risk.^1–6 However, increasing evidence suggests that this association is not as strong in older populations as it is in younger groups.^4,7–10 Moreover, in contrast to the general population, older populations tend to exhibit a reverse J-shaped or L-shaped association between BMI and all-cause mortality^11–13, which has also been found in Asian counties.^4,14–16 More specifically, these studies report that underweight is an important predictor of mortality, while overweight has the lowest mortality risk in old age. This pattern of association has been referred to as the “obesity paradox.”¹⁷

Although many previous studies have used BMI measured only at one point in time to predict subsequent mortality, they overlooked the possibility that weight changes occurring over time might also affect one’s risk of dying. To address this gap, several studies have analyzed mortality risk in relation to weight changes between two points in time. Particularly, it has been reported that, in older adults, weight changes between two points in time predict mortality better than a single weight assessment.¹⁸ More recently, Cheng et al.¹⁹ have published a meta-analysis of the association between weight change and all-cause mortality in old age, focusing on 17 articles. They report that both weight loss and weight gain show associations with higher mortality, with pooled relative risks (95% confidence interval) of 1.67 (1.51–1.85) for weight loss and 1.21 (1.09–1.33) for weight gain. Such findings highlight a potential adverse effect of weight change on all-cause mortality in later life.

In addition, Cheng et al.¹⁹ reported no potential difference in the deleterious effects of weight loss and gain on mortality across geographic regions (Western countries vs Asian countries), as well as by age, sex, or follow-up duration. They demonstrated that, to date, there have been just two studies from Asian nations exploring the linkage between weight change during a certain period of time and all-cause mortality in old age: one from Japan and the other from Hong Kong. Nanri et al.²⁰ reported that, among Japanese adults aged 60–69 years, weight loss of more than 2.5 kg and weight gain of more than 5 kg over a five-year period were both associated with a greater risk of mortality than weight change within 2.5 kg, even after adjusting for potential covariates. Ho et al.²¹ illustrated that, after controlling for potential covariates, weight loss of 2 kg or more during a two-year observation period was associated with higher mortality rates among Hong Kong Chinese people aged 70 years and over compared to no weight change.

Research on weight change and mortality has the potential to yield important insights with regard to clinical and public health approaches to body weight management. However, prior research has commonly categorized trajectories of change in body weight using assignment rules that are based on a priori, subjective criteria (e.g., 5 kg gain or loss from baseline body weight), measured at only two points in time over a relatively short interval, e.g., 2 to 5 years. As body weight can fluctuate substantially over the adult life course, weight changes measured at multiple time points over an extended period can provide additional important information. Only a small number of studies have attempted to assess underlying body weight growth curves, i.e., trajectories of body weight, involving both initial levels of body weight and their rates of change over time) in old age.^22,23 This approach assumes population homogeneity (i.e., a presence of a single group with an average trajectory) with respect to body weight change over time.

However, it is more likely that systematic population heterogeneity exists in the pattern of body weight change over time, perhaps reflected in qualitatively distinct subgroups of the population with regard to body weight trajectories. Therefore, the assumption of population homogeneity might be too simplistic. By identifying and modeling heterogeneity in the BMI trajectory within a given population, it is then possible to assess whether specific trajectory subgroups carry differential mortality risks, ultimately helping to guide the development of individualized targeted interventions.

Group-based trajectory models represent a statistical approach that can be useful for understanding this type of population heterogeneity.^24–28 This technique assumes the presence of and identifies latent groups of individuals who share a particular developmental trajectory of some attribute, thereby allowing a better understanding of the pattern of change in the variable of interest. The technique has been used extensively in criminology and behavioral research and less in medicine and public health research. Therefore, the number of studies utilizing this approach and examining the association between patterns of BMI trajectories and mortality among older adults might be quite small. However, even though the current research is limited, a review of existing published articles should lead to a better understanding of the most important future directions for this topic. The purpose of this study, therefore, was to review studies that utilized group-based trajectory models to investigate the association between the long-term trajectory of BMI and mortality risk among older adults.

PubMed search for BMI trajectories and mortality

We used the PubMed database to conduct a systematic search of peer-reviewed studies on the relationship between BMI trajectory and mortality, published before January 31 2017. The following keywords were used in the search: [“body mass index” OR “body weight”], [“trajectory” OR “pattern”], [“mortality”], and [“elderly” OR “older”]. Keywords were combined in the searches. When the searches were completed, we first reviewed the title, keywords, and abstracts. If this initial review suggested that the study was relevant, we then reviewed the full text of the article for final selection.

The search included original articles published only in English. Additional inclusion criteria were the following: (1) the study used BMI or body weight as a main predictor; (2) the study set mortality as a main outcome (and included any type of mortality); (3) the study included non-institutionalized individuals in the sample; and (4) the study focused on BMI trajectories during middle and old age.

Main results

These search strategies identified three articles suitable for review.^29–31Table 1 shows the characteristics of the three reviewed articles, including information on the study sample (study country, dataset, sample characteristics, and sample size), approach to identifying BMI trajectories (observation points, measurement method of height and weight, and statistical model), details of the survival analysis (outcome, follow-up, statistical model, and covariates), and main results.

Two of the reviewed articles used data from the U.S. Health and Retirement Study (HRS)^29,30, and the other, from Japan, used the National Survey of Japanese Elderly.³¹ The sample in the HRS was community-dwelling individuals aged 51–61 years at baseline, and both HRS-based studies used the nine-wave panel data from 1992 to 2008 (16 years). The Japanese sample consisted of community-dwelling individuals aged 60 years and older at baseline, and the study used the seven-wave panel data from 1987 to 2006 (19 years). Each of the three studies employed different analytic models to identify distinct BMI trajectories, respectively: a semiparametric group-based trajectory analysis, a growth mixture model, and a group-based mixture model. However, it is known that these models produce comparable trajectories³² and are collectively called “group-based trajectory models.”^24–28 With these approaches, it is recommended that the best-fitting model be chosen on the basis of the Bayesian Information Criterion (BIC) scores and an examination of 95% confidence intervals.²⁷ Typically, investigators estimate models with some trajectories (e.g., 1, 2, 3, …, n), select the best fitting model by comparing the BIC associated with various solutions and the average posterior probabilities of group membership, and evaluate whether successive models identify additional distinct groups as indicated by non-overlapping confidence intervals.

Zheng et al.²⁹ reported six distinct trajectories: “class II/III obese upward” (3.4%), “class I obese upward” (11.7%), “overweight obesity” (22.8%), “overweight stable” (29.5%), “normal weight upward” (24.2%), and “normal weight downward” (8.4%). Roughly 15% of the sample was within the obese range, 52% within the overweight range, and 33% within the normal range. Survival analysis indicated that mortality was more likely in individuals with the “class II/III obese upward” trajectory and “normal weight downward” trajectory were more likely to die compared to those with the “overweight stable” trajectory. They also conducted an attributable risk analysis and found that the mortality risks associated with the “class I obese upward” and “class II/III obese upward” trajectories were 3.0% and 4.2%, respectively.

Zajacova et al.³⁰ stratified data by sex and identified three trajectories in both sexes. The majority exhibited the “overweight stable” trajectory, 92.9% of men and 89.5% of women. Additional trajectories included “obese gaining” (2.8% for men and 6.2% for women) and “obese losing” (4.3% for men and 4.4% for women) trajectories. These researchers found that, among both men and women, people in the “obese gaining” and “obese losing” trajectory groups were about 1.5 times and 3–4 times more likely to die relative to those in the “overweight stable” group, respectively. These risk estimates were calculated without adjustment for potential covariates.

Finally, Murayama et al.³¹ analyzed Japanese data and identified four distinct trajectories: “low-normal weight, decreasing” (23.8%), “mid-normal weight, decreasing” (44.6%), “high-normal weight, decreasing” (26.5%), and “overweight stable” (5.2%). Approximately 95% of the Japanese sample were within the normal weight range, and they had decreasing BMI trajectories. Researchers observed that, relative to those with the “mid-normal weight, decreasing” trajectory, those with the “overweight, stable” trajectory and those with the “high-normal weight, decreasing” trajectory had lower mortality risks, and those with the “low-normal, decreasing” trajectory had higher mortality risks.

Comparison across studies

We reviewed articles with regard to the relationship between BMI trajectory and mortality among older people and identified three published studies. A prior meta-analysis concluded that both weight loss and gain were mortality risks in old age.¹⁹ However, although weight change during a short interval measured at two points in time might be associated with increased mortality risk in subsequent follow-up periods, it is important to note that body weight can fluctuate substantially over a given period of time. Therefore, group-based trajectory models, which examine heterogeneity of long-term weight change over time using panel data with multiple BMI data points, can facilitate our understanding of distinct patterns of BMI change and the associated risks of mortality. Moreover, as trajectory models can identify individuals with distinct BMI trajectories, this type of analysis offers potential for helping investigators and practitioners in targeting different interventions specific to each trajectory group.

Of the three identified articles, two were conducted in the U.S. and the other in Japan. Even though the two U.S. studies were based on the same data, each identified different patterns of BMI trajectory. Zheng et al.²⁹ reported that the six-trajectory model provided the best fit for the data compared with other models, based on the BIC index. They also noted that the three-trajectory model identified by Zajacova et al.³⁰ risked concealing heterogeneity in BMI trajectories because this model collapsed most respondents (approximately 90% in both sexes) with underweight, normal weight, overweight, and obesity into the “overweight stable” trajectory group. These different findings from the same data suggest that more research is needed regarding strategies on the best-fitting, group-based trajectory model.

In addition, BMI trajectory patterns were quite different for older Americans and older Japanese. When comparing the U.S. research by Zheng et al.²⁹ with the Japanese research by Murayama et al.³¹, we found that trajectories within the overweight and obese ranges accounted for 70% of the U.S. sample, and all of these trajectories were increasing, whereas only a little over 5% of older Japanese followed the “stable overweight” trajectory. Furthermore, BMI changes in the older Japanese were primarily within the normal-weight range (i.e., BMI: 18.5–25.0), and the three trajectories identified within this range were decreasing over time. In particular, approximately 24% of the sample followed the trajectory of “low-normal weight, decreasing,” which started with a BMI of 18.7 and decreased to 17.8 over 19 years. By contrast, in the U.S. sample, only 8% had a downward trajectory within the normal weight range. Therefore, older Japanese appear to differ from older Americans not only in terms of the weight distribution at a single point in time, but also in the long-term patterns of weight changes over time.

In terms of survival analysis, findings from the three studies were consistent. Both among older Americans and older Japanese, those who were overweight but stable had the highest survival rate. These findings are also consistent with a prior meta-analysis focusing on BMI (measured at a single point) and all-cause mortality in older adults, which reported that overweight status had the lowest risk of mortality.³³ Similar associations have also been found in several Asian studies.^{14–16,34–36} In addition, prior research has found that stable weight status is associated with a lower mortality rate compared to weight change among older adults both in Western and Asian countries.¹⁹ The findings of our review of studies with longer follow-up periods are consistent with these prior findings. Taken together, these findings indicate that older adults with stable overweight trajectories over time have the lowest mortality rates, regardless of whether they reside in Western or non-Western societies.

However, our review suggests that priorities for weight control in older adults might need to differ between Western and non-Western countries. Based on the findings from Zheng and his colleagues²⁹, it appears that body weight control interventions are most needed for older Americans with the “class I obese upward” and “class II/III obese upward” trajectories because approximately 7.2% of deaths were found in these trajectory groups, according to attributable risk analysis. In contrast, it appears that weight control measures for older adults in Japan should focus on those with the “low-normal weight, decreasing” trajectory because this subgroup had the highest mortality rate.³¹ Researchers contended that two factors were related to this result: first, thin people tend to have less energy and nutritional reserves, which results in less resistance to infection;^14,37 second, decreasing body weight is often accompanied by an accumulation of muscle or bone loss^38,39, resulting in a reduced ability to maintain energy balance and the sharing of many immunological and neuroendocrine features of disease-associated wasting syndromes.⁴⁰

This review identified three papers focusing on the association between BMI trajectories and mortality in old age. Although the distribution of long-term BMI trajectories was different between older Americans and older Japanese, the finding that those with an overweight stable trajectory had the lowest mortality rate was consistently observed. This research provides insights concerning the complex and dynamic relationship between BMI and mortality in later life that is not captured in research focusing on the mortality risks associated with BMI measured at one point in time or weight change measured at just two-points in time. Evidence from other nations is warranted in order to validate the findings from the U.S. and Japan. Such work can also contribute to development of clinical and public health strategies for body weight management aimed at improving the health and survival of older adults in each country.