Weight change is an important risk factor for mortality. The association of weight change with mortality is frequently shown as a reverse J-shaped curve.^1,2) Weight loss in overweight or obese people and weight gain in underweight people have beneficial effects on health.¹⁾ However studies have reported conflicting results depending on ethnicity and age groups. While some studies reported that weight loss and weight gain increased mortality,^3,4) others found no associations.⁵⁾

Instead of using baseline body mass index (BMI) as in previous studies,^6,7) some studies used body weight change as a predictor of mortality in Korea.^1,8) Thus, weight status changes throughout a lifetime should be considered.^9,10) However, most studies have focused only on the mortality consequences of weight (or BMI) changes between two time points during the study period.^1,8) Because there is a limit to defining an individual’s subgroup based on short-term weight or BMI changes, this approach obscures heterogeneity in weight changes that occur after the second time point, as well as weight fluctuations.¹¹⁾ In particular, the dynamic measures of BMI may better predict mortality in older adults.^11,12)

Several studies have performed repeated observations of weight at regular intervals over a period longer than 10 years to evaluate the associations between weight changes and mortality in older adults. However, the study using data from the Health and Retirement Study by Zheng et al.¹¹⁾ included a relatively young population with mostly white ethnicity. Many non-western societies have varying distributions of body weight from the United States;¹³⁾ thus, studies of older Asian adults are needed.

Therefore, this study followed up BMI changes over 10 years among community-dwelling Korean adults aged 65 years or older and investigated the effects of these changes on mortality using data from the Korean Longitudinal Study of Aging (KLoSA). A trajectory model was developed to classify different subgroups according to BMI.

This is the first study to identify the heterogeneity of BMI trajectories and demonstrate its association with mortality in the older population in Korea. The participants were classified into obesity, overweight, high-normal, and low-normal groups, all of which showed a slight decrease in BMI over a 10-year period. BMI decreased with age in the LNWG and HNWG but was stable in the OWG and OG. Of the four trajectory groups, the group with higher BMI was younger, had better cognition, and exercised regularly compared with the group with lower BMI. However, the group with a lower BMI had a relatively lower prevalence of HTN and DM and included more male and current smokers. As shown in Model 2 (Table 2), after adjusting for confounding variables, the group with lower BMI had a consistently higher risk of all-cause mortality.

A meta-analysis of prospective studies reported that weight loss was related to an increased risk of all-cause mortality in middle-aged populations.¹⁷⁾ However, as the relationship between body weight and mortality appears to vary by age,¹⁸⁾ it is important to investigate whether the association between weight change and mortality persists in older populations.

Many studies have reported the association between BMI and mortality in Korea.^1,6-8) Several large cohort studies have been conducted based on baseline BMI in patients aged over 65 years receiving regular health checkups.^6,7) These studies reported higher mortality risks in the group with lower baseline BMI. Park et al.⁸⁾ investigated the association of BMI change with mortality in addition to baseline BMI. Their subgroup analysis showed that BMI change better predicted mortality than a single measure. Using the National Health Insurance System health checkup data, Kim et al.¹⁾ reported the association between weight change and subsequent mortality, demonstrating a reverse J-shaped all-cause mortality curve for weight change; furthermore, weight loss was associated with a higher risk of mortality than was weight gain. However, these studies measured BMI just twice (at baseline and later) in a short period and included a relatively young population. In contrast, the current study attempted to classify BMI subgroups by repetitive measurement of BMI through long-term follow-up and by using a trajectory model.

Several studies have explored the relationship between BMI trajectories and mortality. Using data from the Health and Retirement Study for older Americans, Zheng et al.¹¹⁾ defined six linear trajectories (overweight stable, normal weight downward, normal weight upward, overweight obesity, class I obese upward, and class II/III upward). The normal weight downward and class II/III obese upward group had high mortality risk, while the overweight stable group had the lowest mortality. In our study, the weight trajectories of the LNWG and HNWG showed a decreased pattern during the study period and those of the OWG and OG were relatively stable. The OG had the highest survival rate, while the LNWG showed a higher risk of mortality. Although Zheng et al.¹¹⁾ analyzed a younger population (51–61 years), the differences were more likely due to racial differences. Since racial/ethnic differences in obesity status, body composition, and weight management behaviors have been reported, weight change and its association with life span may vary across racial/ethnic groups.⁴⁾

However, our results are similar to those of one Japanese study by Murayama et al. ¹³⁾. They identified four distinct BMI trajectories in a 19-year follow-up of older Japanese adults and reported that adults who were overweight but stable had the highest survival rate. Thus, older East Asians may differ from older Americans in terms of weight distribution and patterns of weight change over time.

Compared with short-term (within 5 years) weight changes reported previously,^6,7) weight changes in the four trajectory groups were small. However, body weight can fluctuate substantially over the adult lifetime; thus, weight changes measured at multiple points over an extended time (approximately 10 years) may provide additional information.¹³⁾ Moreover, while previous studies categorized weight change groups using arbitrary criteria such as 5% gain or loss from baseline body weight,¹⁾ these approaches did not provide precisive individual classifications into the various bodyweight groups.¹³⁾ Our trajectory modeling method can more precisely define weight change groups of real older populations, even if the changes are small.

People with low BMI are more likely to have less muscle and bone mass and increased body fat. Reduced muscle and bone mass increase the likelihood of osteoporosis and fractures and decrease body function, which in severe cases, leads to frailty and disability.^19,20) Low muscle mass in the low BMI population causes deterioration of lung function and makes it difficult to withstand catabolic stress in respiratory crisis situations.²¹⁾ In some studies, underweight people were more susceptible to respiratory disease and showed higher mortality rates.²²⁾ In contrast, high BMI suggests that muscle and body fat, which serve as energy and nutrient reserves, are likely to be sufficient.²³⁾ Underweight people could be undernourished, which reduces immune cell function, increasing these individuals’ susceptibility to infection.²⁴⁾ Underweight people are reportedly susceptible to infectious diseases such as tuberculosis.²⁵⁾ Studies have also shown higher probabilities of complications and mortality among individuals with chronic diseases, including CVD, and among those with lower BMI.^26,27) Chronic disease can result in a catabolic state that, in turn, leads to wasting syndrome and metabolic stress situations with muscle, bone, and body fat loss.²⁷⁾ Accordingly, people who lack sufficient energy and nutrient reserves are intolerable. Moreover, insufficient nutrition prolongs hospitalization and increases mortality in hospitalized patients.²⁸⁾ Thus, high BMI may have a protective effect against various deteriorating conditions, which may improve survival.

Several limitations should be considered when interpreting our outcomes. In general, people overreport their height, and they (especially obese people) tend to underreport their weight, resulting in under-estimated BMI. The present study used subject-reported height and weight to calculate BMI, which could lead to biased results in BMI trajectory patterns. Second, there was the possibility of selection bias. The participants of the KLoSA study were residents enrolled nationwide to form a nationally representative database. Individuals in long-term care facilities were not included in this study. Moreover, 16.5% of participants were excluded from the analysis due to drop-out during the study period. Comparisons of the characteristics of the included and excluded participants are shown in Supplementary Table S4. In general, the excluded participants were older and had more morbidity than the participants included in the analysis. This difference means that more vulnerable participants were more likely to drop out during the study period. The sample admitted to long-term care facilities owing to chronic diseases during follow-up may have been excluded from the main panel data. The participants often have more functional limitations and comorbidity than those who live at home. Therefore, this analysis of the effect of BMI on mortality could have a limitation among people with high levels of frailty. Third, intentional and unintentional weight changes were not distinguished. As unintentional weight loss has a higher mortality risk than intentional weight loss,^29,30) we adjusted for various underlying diseases and functional limitations that could lead to unintentional weight changes. Fourth, in determining the degree of obesity, BMI is not an optimal indicator. Since BMI is calculated based on total weight, it is not possible to directly measure the weight of body fat and muscle composition. Moreover, this calculation does not reflect the distribution of body weight. It is more advantageous to measure waist circumference when a person had central adiposity with high levels of fat in the abdomen;³¹⁾ however, the present study did not adjust for waist circumference. Finally, only all-cause mortality was assessed. The associations with cause-specific mortality, such as death from cardiovascular or other causes, should also be investigated in future studies.

Nevertheless, this study has some important strengths. Most previous studies adopted methods of classifying BMI change into predetermined intervals arbitrarily, as defined by researchers. In contrast, this study used a trajectory model to express BMI change more objectively as a result, we identified four homogenous trajectory groups. Moreover, the relationship between BMI subgroup and mortality could be more accurately demonstrated through continuous and repeated BMI measurement over 10 years of follow-up. This study of data of the older population extracted from the KLOSA database has the advantage of a relatively large sample size and a prospective design. Physical assessment was performed using validated measures, and variables related to lifestyle and health were considered. Since this study included healthy community-dwelling residents in Korea, the findings can be generalized to the Korean population. Moreover, we performed the analyses after adjusting for multiple sociodemographic and behavioral variables (age, sex, marital status, educational level, co-morbidity, smoking, drinking, and regular exercise).

In conclusion, the results of our study showed dynamic associations between BMI and mortality among older adults in Korea. We identified a slight downward trajectory of BMI in the LNWG and HNWG groups and a stable trajectory in the OWG and OG groups. These findings are contrary to those in western populations, in which weight gain in the obese group and weight loss in the normal weight group were observed, but the findings are similar to those in Japan, where weight loss was observed in all subgroups except for the OWG. The OG group had the highest survival rate, while the LNWG group had a higher risk of mortality. This research suggests that baseline obesity status and degree of weight loss during the follow-up period contributed to mortality in later life. Further studies are needed to determine the mechanism of these results in more detail.