Study population
This longitudinal cohort study included individuals who participated in the Taipei Elderly Health Examination Program between 2006–2011, which is free for adults aged 65 years or older and is supported by the Taipei City Government.
Figure 1 illustrates the selection process for the study population. A total of 100,273 individuals participated in the Taipei Elderly Health Examination Program between 2006–2011. Among them, 60,742 participants who underwent repeat examinations were screened for inclusion in the study. Participants with incomplete covariate data (n = 5,391) were excluded from the study. Ultimately, 55,351 individuals were included in the study.
All participants in this study were followed-up until death or December 31, 2012, whichever occurred first. The death of the study participants was confirmed using the death certificate database of Taiwan [9].
Data collection and statement of ethics
While participating in the Taipei Elderly Health Examination Program, individuals were instructed to complete a self-administered questionnaire that was used to collect their demographics, medical history, and lifestyle behaviors, such as smoking status, alcohol consumption, and dietary and exercise habits. During the health examinations, body weight, height, and blood pressure were measured using standard procedures. Overnight fasting blood samples were collected for laboratory analyses, including complete blood count, liver function tests, blood glucose, and serum creatinine levels.
This study was approved by the Institutional Review Board of Taipei City Hospital (IRB No.: TCHIRB-11205001-E-F), and all data used in the analysis were de-identified and anonymized. The study was conducted in accordance with the tenets of Declaration of Helsinki.
Outcome variable
The outcome variables were all-cause and CVD-related mortality, which were confirmed using the death certificate database in Taiwan [9]. In Taiwan, when a patient dies, the law requires the death certificate of the patient to be issued and registered by the physician in charge according to the International Classification of Diseases (ICD) 9 or 10. Trained medical registrars review and code all death certificates at the central office of the National Death Certification Registry. Therefore, cause-of-death coding in Taiwan has been considered very accurate [9].
CVD-related mortality in study participants was defined as death from coronary heart disease (ICD-9 codes 410 − 414 and 420 − 429; ICD-10 codes I20-I25), stroke (ICD-9 codes 430 − 438; ICD-10 codes I60-I69), and other circulatory diseases (ICD-9 codes 390–392, 393–398, 401–405, and 440; ICD-10 codes I10-I15, I01-I02.0, I05-I09, I27, I30-I52, I70, and I71), as described previously [10].
Exposure variable
The main exposure variable was the change in BMI. Trained nurses measured the BMI of the study participants during each health screening. It was calculated by dividing the weight in kilograms by the square of the height in meters. The BMI change was expressed as a percentage, which was calculated as the change in BMI between the first and last health screening divided by the first health screening BMI. The participants were divided into five groups based on the percentage of BMI change: >10% decrease in BMI; 5–10% decrease in BMI; stable BMI with < 5% change; 5–10% increase in BMI; and > 10% increase in BMI.
Covariates
Covariates entered in the core analyses included sociodemographic characteristics (e.g., age, sex, BMI, level of education, marital status, smoking status, and alcohol consumption), fruit and vegetable consumption, and exercise. BMI in study participants at the first health screening was categorized as underweight (< 18.5 kg/m2), normal (18.5–23.9 kg/m2), overweight (24–26.9 kg/m2), and obese (≥ 27 kg/m2) [11]. Education level was classified as 0–6 years, 7–12 years, or > 12 years. Marital status was categorized as married, unmarried, widowed, or divorced.
Smoking status was assessed by asking, “In the past six months, how often did you smoke?” The answers were “no smoking,” “occasional smoking,” or “smoke every day.” The question regarding alcohol consumption was “In the past six months, how often did you consume alcohol?” Responses included “no drinking, occasional, or frequent drinking.” The question used to record dietary information was “Do you eat at least three servings of vegetables and two servings of fruit every day?” The replies were “yes” or “no.” The information on exercise habits was gathered by the question “In the past six months, how often did you exercise for more than twenty minutes in a week?” The possible answers were “No exercise,” “1–2 times in a week,” or “more than 3–5 times in a week.”
Serum samples for biochemical assays were collected after overnight fasting for 8 h. Laboratory data such as liver function tests, fasting blood glucose, and serum creatinine levels were obtained in the hospital using a standardized procedure with high accuracy.
Statistical analyses
The baseline characteristics of BMI changes in the groups were compared using the chi-square test for categorical variables and one-way analysis of variance (ANOVA) for continuous variables, as appropriate.
Unadjusted all-cause and CVD-related mortality per 1000 person-years were calculated for individuals with different BMI changes. The cumulative incidence probabilities of all-cause and CVD-related mortality were plotted using Kaplan–Meier curves and compared among the five BMI groups using the log-rank test.
A multivariable Cox proportional hazards model was used to determine the association between changes in BMI (reference group: stable BMI with < 5% change) and the risk of all-cause mortality after adjusting for sociodemographic characteristics, lifestyle behaviors, and comorbidities. The Fine-Gray sub-distribution hazard model was used to determine the association between changes in BMI and the risk of CVD mortality, with death from non-CVD causes as the competing risk [12].
To examine the robustness of our primary findings, sensitivity analyses were conducted after stratifying the participants based on their BMI at the first health screening. All data management and analyses were performed using the SAS software (version 9.4; SAS Institute, Cary, NC, USA).