Study population

The National Health Insurance System (NHIS) in South Korea is compulsory, and all citizens in the Republic of Korea are required to participate. As described in detail previously,[15] the NHIS in Republic of Korea oversees the national health examination programs.

The NHIS released NHIS-National Health Screening Cohort (NHIS-HEALS), comprised of 514,795 randomly sampled individuals from population that received NHIS health screening examinations between 2002 and 2003, as basically recommended by biennial intervals (annually for manual workers), and were followed up until 2013. The cohort represents approximately 10% of the source population who underwent NHIS examination between 2002 and 2003, and has been followed up to either the time of the participant’s disqualification of health services due to death or emigration or the end of the study period in 2013. The database contains information regarding the health screening questionnaire and laboratory tests, health insurance type, medical bill details, medical treatment, disease histories, and prescription records.

Fig 1 displays the selection criteria for the study population. From the overall cohort of 514,795 subjects who received health examinations between 2002 through 2003, we excluded those with pre-existing histories or treatment history of CV disease, cancer, or cerebrovascular disease. We then selected healthy men and women aged between 40 and 79 years at baseline health examination and excluded those who had died within 3 years following the first examination. We finally selected 379,851 subjects who underwent the health examination more than three times for measurement of BMI variability; patients with missing values for baseline BMI were omitted. In total, 379,535 individuals with more than three repeat measurements for BMI and CV risk factors formed the analytic sample. This study was based on data derived from the NHIS, and informed consent was not specifically obtained from each participant. Data were fully anonymized and de-identified for all analyses. This study was approved by the Institutional Review Board of Yonsei University, Severance Hospital, Seoul, South Korea (IRB no.4-2016-0421).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow chart of the study population.

NHIS-HEALS, National Health Insurance Service-National Health Screening Cohort.

https://doi.org/10.1371/journal.pone.0189180.g001

Clinical variables

BMI was defined as subject weight in kilograms (kg) divided by the square of height in meters (m²) and was measured at each health examination. Based on baseline BMI, subjects were classified as underweight (BMI less than 18.5 kg/m²), normal weight (BMI 18.6–22.9 kg/m²), overweight (BMI 23.0–24.9 kg/m²), or obese (BMI greater than 25.0 kg/m²).[16] Systolic and diastolic blood pressures were also measured at each visit. Serum samples for fasting glucose and total cholesterol were obtained after overnight fasting at each examination visit. A detailed history of smoking status and alcohol consumption were obtained via questionnaire. The latter two measures were categorized for the purpose of this study according to current, former, or never for smoking status and drinker or non-drinker for alcohol intake.

Medical history of participants was identified using a combination of the following: clinical and pharmacy code of the 10^th revision of the International Classification of Diseases (ICD-10), lists of prescribed medicine, and previous medical histories. To identify hypertension and diabetic mellitus status, laboratory data were utilized in addition to medical records and health examination survey results based on the criteria of systolic blood pressure (SBP) ≥140 mm Hg or diastolic blood pressure (DBP) ≥90 mm Hg for hypertension and fasting serum glucose ≥126 mg/dL for diabetic mellitus. The laboratory and survey questionnaire data of general and life-transition health examinations for all cohort members were merged.

BMI change categories

Based on BMI changes from baseline to midpoint follow-up and from midpoint follow-up to final health check, individuals were stratified into five groups as follows:[7] (1) Stable: less than 4% change in BMI from baseline to midpoint to last health checkup; (2) Sustained gain: increased BMI (≥4%) since baseline without BMI decrease (≥4%) during follow-up. This included BMI stability or gain from baseline to midpoint follow-up and further BMI gain from midpoint follow-up to last visit, or BMI gain from baseline to midpoint follow-up followed by stable BMI; (3) Sustained loss: decrease in BMI (≥4%) since baseline without increased BMI (≥4%). This included BMI stability or loss from baseline to midpoint follow-up and further BMI loss from midpoint follow-up to last visit, or BMI loss from baseline to midpoint follow-up followed by stable BMI; (4) Loss-gain: BMI loss from baseline to midpoint follow-up, followed by BMI gain from midpoint follow-up to last visit; and (5) Gain-loss: BMI gain from baseline to midpoint follow-up, followed by BMI loss from midpoint follow-up to last visit. Groups 4 and 5 were included in the category of BMI fluctuation, with no assumptions as to whether such BMI changes were intentional or unintentional.

Mean values and standard deviations (SDs) of BMI at three time points (baseline, midpoint, and last follow-up) were calculated. Direction and magnitude of change in subject’s BMI was determined by regression slope of BMI values (BMI slope). Slopes of SBP, DBP, and total cholesterol were calculated by each regression equation at three time points over time.

Study endpoint

The study outcomes were death from all causes as well as death from CV and non-CV causes. Mortality status of each participant was ascertained together with specific cause of death. The latter events were classified according to ICD-10 codes that were obtained from the South Korean National Statistical Office.

Statistical methods

Variables are reported as percentages or as means ± SDs for normally distributed variables and as median (with 25–75% ranges) for non-normally distributed variables, as appropriate. A multivariate Cox proportional hazards regression model was employed to assess the relationship between BMI changes and mortality outcomes by estimating hazard ratios (HR) with 95% confidential intervals (95% CI). To assess the relationships between BMI measures and CV risk factors, we used a multivariate linear regression analysis in which each CV risk factor was treated as the dependent variable, and the independent variables were the baseline value of each CV risk factor of age, smoking status, alcohol consumption, hypertension, and diabetic mellitus. All analyses were conducted using SAS, Version 9.4 (SAS Institute Inc., Cary, NC, USA), with a P value <0.05 considered significant in all analyses.

BMI change and survival

The potential detrimental effects of BMI loss have been substantially reported.[17–19] weight loss has long been associated with increased mortality and increased risk of hip fracture, irrespective of current weight or intention to lose weight. In the current study, BMI loss was predominantly associated with an increased risk of non-CV mortality, which is somewhat concordant with previous studies.

To date, however, the harmful or beneficial impact of BMI gain on mortality in middle-aged adults has received little attention, particularly beyond Western societies. There are conflicting results on this topic. Some have reported that weight gain up to midlife is considered to impair quality of health in later life,[20] and childhood obesity is associated with premature mortality.[21,22] In contrast, the present study indicated that a mild weight gain, reflected by BMI gain, might actually reduce the rate of all-cause mortality, which is somewhat consistent with the few previous studies showing better outcomes with BMI gain.[7,12]^,[23]

A novel finding of the current study was that BMI gains beyond midlife might actually associated to a more favorable prognosis, despite the finding that the baseline prevalence of obesity and overweight in the weight gain group was reasonably high at 25% and 22%, respectively. Notably, the favorable outcome on all-cause mortality was not solely attributable to a decline in non-CV mortality. Instead, the beneficial impact of a sustained BMI gain on all-cause mortality appeared to reflect reduction in both CV and non-CV mortality in the sustained BMI gain group, which contradicts our expectation that CV mortality risk would most likely increase on the background of BMI gain. Clearly, this warrants further investigation. Interestingly, BMI loss was associated with unfavorable outcomes, even after adjusting for baseline or mean BMI. The overall prevalence of obesity or overweight at baseline was higher in the BMI loss group, in comparison to BMI gain group or stable BMI group. Therefore, in contrast to our perception that BMI loss would be helpful for those with overweight or obesity, impact of BMI loss on clinical outcomes needs to be further investigated in healthy middle-aged persons, even when they are overweight or obese.

Metabolically healthy obesity

One plausible explanation for improved outcomes associated with BMI gains might be linked to the term ‘metabolically healthy obesity,’ which was generated to describe the absence of components of metabolic complications in the presence of increased BMI.[24] The study population demonstrated a low prevalence of chronic conditions such as hypertension and diabetic mellitus and presented with a nearly normal range of laboratory tests and blood pressure at baseline examination. Intriguingly, improved clinical outcomes were observed in the sustained BMI gain group, even though BMI gains were associated with increased levels of CV risk factors including blood pressure and total cholesterol in this group. To this end, we speculate that any BMI gains observed in the metabolically healthy individuals might have conferred a protective effect against the risk of CV and non-CV mortality. This finding is somewhat thought provoking considering that BMI gain itself typically induces an increased presence of CV risk factors such as blood pressure or cholesterol, as found in the current study, according to the linear relationships between BMI and other CV factors, as well as an increase in systolic blood pressure and total cholesterol during follow-up in BMI gain group. This might, in part, be associated with the notion that incidence of coronary heart disease is often not increased in metabolically healthy obese persons [25], and high BMI may survive longer when compared to their counterparts with lower BMI, although high BMI is associated with higher prevalence of diabetic mellitus and CV disease. Also, heart failure [26] and stroke [27], as well as coronary heart disease, have demonstrated this obesity paradox regarding survival, and these factors might have influenced the results of this study in light of the lower rates of all-cause and CV mortality observed in the sustained BMI gain group.

Change in body composition after midlife

Another possible explanation for the favorable impact of BMI gain on prognosis might be the change in body composition that is often observed during the latter stages of the aging process. Body weight gradually increases throughout adulthood until approximately 60 years of age, after which time the mean body weight appears to decrease.[28] Most notably, lean body mass, predominantly muscle, is lost at an accelerated rate, with and without intentional weight loss, as individuals age.[29] Stable weight individuals who maintain adult weight through their 6^th and 7^th decades are not necessarily protected from age-related muscle loss;[30] this has been reported through a substantial shift in body composition marked by progression of sarcopenia across a 5-year period in healthy, ambulatory, stable weight elderly subjects.[31] Newman et al. also reported that lean body mass decreased in stable weight patients with increasing age, but increased among those who gained weight, although it should be noted that fat mass gain was still considerably higher in that study.[32]

Lean body mass could therefore be a feasible explanation for why BMI loss carries a higher mortality risk, whereas BMI gain might be associated with a better clinical prognosis. Considering a report that mortality was highest in coronary heart disease patients with both low body fat and low lean mass and lowest among those with high fat and high lean mass,[19] maintaining lean body mass through subtle BMI gains during mid-to-late life might prove critical for better health in older individuals, even though mechanism for explanation needs further investigation regarding theories stating that BMI gain might provide a safer storage location for environmental toxins.[33] If proven in forthcoming studies, preserving the concept of “healthy age-related mild weight gains” to maintain lean body mass in order to reduce adverse outcomes should be advocated following midlife. This might also result in the most favorable clinical outcomes in BMI gain individuals, who might not lose lean body mass during the aging process.

Limitations

An inherent limitation of this study includes a potential selection bias. The NHIS recommends, but does not obligate all health insurance subscribers to undergo biennial health examinations. Hence, this cohort probably consisted of healthier individuals or those who are more concerned about their health. However, the NHIS-HEALS data reflect a representative sample of Koreans who were prospectively followed during the study period. Additionally, the participation rate of the general health screening program was high: up to 74.8% in 2014 according to NHIS.[34] We cannot discount the possibility that the current findings might have differed if we used information on other more robust measures of obesity including body composition, visceral fat mass,[35] and waist circumference. Amounts of drinking and smoking and menopause status in women influence BMI changes and mortality. Further studies would be helpful to evaluate the interactions among these factors. In this study, we used BMI rather than weight to assess body weight changes, since our dataset lacks information regarding body weight and height, and only includes BMI value calculated from weight and height. Since BMI could be affected by both weight and height, we cannot exclude the possibility that changes in height and weight could affect BMI changes, although height change might be relatively smaller than weight change. It would be helpful to look at the prevalence of metabolic syndrome among the groups. However, we do not have information regarding waist circumference, HDL-cholesterol and triglyceride. Further studies regarding association between moderate BMI gain and changes in prevalence of metabolic syndrome is warranted. BMI change groups were defined by percent changes, which could be affected by the initial body weight of the individual. However, as there is no universal definition of BMI change groups, we followed criteria of a previous study.[7] The lack of reliable information related to intentional and unintentional weight loss or gain further limits interpretation of the present study findings. BMI change, as well as events and changes of CV risk factors, were all assessed during the same follow-up period, which can be biased. If we had performed analysis for BMI change in a certain period and then performed clinical follow-up thereafter, different formulations might have led to different data analytic results. However, as our main focus was to investigate the way BMI changes as people go toward event, we decided to perform the concurrent analysis of BMI changes and events. Although we analyzed BMI at three separate time points over the course of the study follow-up period, BMI variability between examinations was not assessed. Future studies should provide additional insight by integrating a prospective and possibly time-dependent assessment of the relationship between BMI-associated weight gain and risk of poor outcome in middle-aged adults.