Factors associated with body mass index changes among older adults: a ten-year follow-up Fatores associados a alterações no índice de massa corporal entre idosos: um seguimento de dez anos Factores asociados con índice de masa corporal entre adultos mayores: un seguimiento de 10 años

To examine changes in body mass index (BMI) among older Brazilian adults and associated factors. Longitudinal, population-based study, conducted in São Paulo, Brazil. Adults aged 60 years or over (n = 1,796) from the first wave of data collection from the Health, Well-Being, and Aging Study (SABE Project) conducted from 2000 to 2010. Repeated mixed-effects linear regression was used to analyze longitudinal changes in BMI and to examine whether sociodemographic characteristics, health conditions, and social behaviors were associated with these changes. Mean BMI decreased after 70 years. Men had lower BMI than women (β = -1.86, 95%CI: -2.35; -1.37). Older adults who consumed alcohol (β = 0.30, 95%CI: 0.06; 0.54), had more than one chronic disease (β = 0.19, 95%CI: 0.26; 0.72) and who did not perform physical activity (β = 0.56, 95%CI: 0.38; 0.74) had higher BMI. Subjects who smoked (β = -0.40, 95%CI: -0.76; -0.04) and who reported having eaten less food in recent months (β = -0.48, 95%CI: -0.71; -0.24) had lower BMI. In older Brazilians, several sociodemographic characteristics, health conditions, and behaviors predict BMI. Increasing prevalence of chronic diseases and growing sedentary behaviors in Brazil may have detrimental effects on BMI at older ages. Aged; Aging; Body Mass Index Correspondence T. A. Araujo Rua Tapuios 1305, Uberlândia, MG 38408-416, Brasil. tanniaraujo@hotmail.com.br 1 Faculdade de Saúde Pública, Universidade de São Paulo, São Paulo, Brasil. 2 Centro Universitário de Patos de Minas, Pato de Minas, Brasil. 3 Universidade Estadual de Campinas, Campinas, Brasil. 4 University of Illinois at Urbana-Champaign, Urbana, U.S.A.


Introduction
Due to the burden of chronic diseases associated with increased body mass index (BMI) and obesity in the population 1 , measures to reduce weight, such as reducing the intake of unhealthy foods and encouraging physical exercise 2 , have been supported worldwide. However, for older adults, the relationship of BMI (and its changes) to health status is more complex 3 . Generally, body weight tends to decline gradually at older ages. Weight loss in old age is mainly a result of reduced ability to preserve homeostatic mechanisms that control hunger and satiety 4 . Besides, several physiological changes occurring in old age contribute to changes in body composition among older adults 5 , such as reduced height due to vertebral compression and changes in intervertebral discs, increase in total body fat percentage, decrease in muscle mass, fat redistribution, central fat, bone changes due to lower calcification levels and decrease of body water content 5 .
According to data from the Protective and Risk Factors for Chronic Diseases by Telephone Survey (Vigitel), a system that provides annual estimates on the prevalence of overweight and obesity in the adult population of all Brazilian capitals, the prevalence of obesity increased from 11.6% to 17.4% from 2006 to 2012, representing an average increase of 0.9% per year. This increase was lower among individuals aged 65 and older (0.7% a year), but still significant 6 . Between the 1970s and 2013, the median BMI increased in all segments of the Brazilian population, mainly among older adults, women, and people with lower schooling levels 7,8 .
Understanding which variables are associated with body composition -such as BMI -can help predict health outcomes, such as cardiovascular disease risks and frailty 9,10 . BMI is also widely used in clinical practice and research and has a high correlation with weight 11 . While much is understood about factors associated with BMI at older ages in developed countries, less is known about these associations in middle-income countries 12 . Moreover, most studies are based on cross-sectional data 7,8,13,14,15 , with few exceptions 16,17 . Given the rapid nutritional transitions ongoing in these countries, obtaining information about older adults' health is critical 18 , mainly due to the consequences for disability and mortality 19 . We address this gap in the literature by examining the factors associated with body weight, based on BMI, among Brazilian older adults using longitudinal data.

Participants
This study comprises a sample of participants from the Health, Well-Being, and Aging Study (SABE Project), a cohort study that began in 2000. The participants were selected using a multiple-stage probabilistic sample that is representative of community-dwelling individuals aged 60 years or older dwelling in the city of São Paulo, Brazil 20 . In 2006, a second wave of the study was conducted, in which 1,115 of the participants from the first wave were re-interviewed. A third wave was conducted in 2010/2011 with 748 people who had participated in both the first and second waves of the study. In this study, we included only participants with complete data on weight and height, at least in the baseline. Figure 1 shows the details of the sample used in all waves.
Trained interviewers collected data in household interviews using a structured questionnaire that addressed socioeconomic variables, general health, living conditions, and anthropometric measures -details on the methodology in Andrade et al. 20 .

Measures
The dependent variable was BMI (kg/m 2 ), calculated by dividing body weight (in kilograms) by the square of height (in meters). A trained interviewer measured body weight using a calibrated scale and height using a stadiometer fixed to a wall, all individuals were barefoot, wearing light clothing.
Sociodemographic characteristics were obtained as follows: sex, age (in years), race (white, mixed black, other races -Asian, indigenous), schooling level (illiterate, 1-3 years, ≥ 4 years) and marital   status (married, not married -divorced, separated or widowed/widower). Age and marital status were the only sociodemographic characteristics included as time-varying.
Self-reported health conditions were as follows: self-rated health (good, not good); number of chronic diseases diagnosed by physician (categorized as up to one disease and two or more diseases): hypertension, diabetes, cancer, chronic pulmonary disease, stroke, cardiovascular disease, and osteoarthritis. All health conditions were included as time-varying.
Behavioral aspects were as follows: physical activity (yes, no), whether the participant reported eating less (yes, no), alcohol consumption (yes, no), and current smoking (yes, no). The level of weekly physical activity was assessed using the Brazilian version of the International Physical Activity Questionnaire. Individuals who reported practicing physical activity less than three times a week were classified as physically inactive. Those who answered positively to the question "In the past three months, have you decrease your food intake?" were categorized as eating less. Alcohol intake was assessed by asking participants whether they were non-drinkers, drank once a week, drank two to six days a week, or drank every day. Given the low prevalence of any alcohol consumption, the participants Cad. Saúde Pública 2021; 37(12):e00081320 were divide into non-drinkers and drinkers (those who drank any alcohol). All behavioral variables were included as time-varying.
Furthermore, to examine potential confounding effects of death commonly seen in longitudinal studies, a mortality indicator variable was included to indicate whether the participant died during the study period.

Statistical analyses
Descriptive statistics (means, standard deviations, and percentages) were used to provide estimates at baseline. The Lowess curve 21 was used to evaluate the BMI trends across age. The same methodology was used to examine whether BMI changes differed across sociodemographic, health, and behavioral characteristics over time. Results (not shown) indicate differences in BMI at baseline and linear trends in BMI across age. Marital and multimorbidity status also showed differences in BMI changes.
A repeated mixed-effects linear regression was used to analyze longitudinal changes in BMI and to examine which factors were associated with BMI changes over time 22 . Repeated mixed-effects regressions handle nested data inherent to repeated observations within individuals and allow the comparison of an unequal number of observations across individuals. BMI was treated as a continuous variable.
We examined whether BMI as a function of age followed a linear growth -and we included random effects -that allows individuals to vary in the initial level BMI. Also, random slopes were included to evaluate the variability in BMI growth rate. Unconditional model (not shown) included only age as a covariate indicated that the model followed a linear growth with random intercept and random slope. Then, covariates were included, and finally, interactions with age were tested. The final model includes all covariates and statistically significant interactions. Model diagnosis showed that the assumption of normally distributed residuals was met. Also, predicted values were close to the observed ones. Homoscedasticity of the errors seems to be met.
Thus, regression coefficients, confidence intervals, and p-values of this final model are presented. However, to facilitate the interpretation of regression results for the interaction effects, the linear predictions obtained with the "margins" command and the "marginsplot" command are used to graph the BMI levels across age. All data analyses were conducted using the statistical software Stata (https:// www.stata.com).

Ethics considerations
The datasets analyzed in the current study were used under license and are not publicly available due to the policies of the SABE Project. The SABE Project and this study were approved by the Research Ethics Committee of the Faculty of Public Health at the University of São Paulo, control numbers 315 (2000), 83 (2006) and 2,044 (2010), and by the Brazilian National Research Ethics Committee. All participants signed an informed consent form to participate in the study. Table 1 shows the characteristics of the participants at baseline. The mean BMI was 26.3kg/m 2 and the mean age was 72.9 years. Most of the sample was female (59.4%). Table 2 shows the association of BMI, sociodemographic characteristics, self-reported health conditions, behavioral aspects, and mortality. Model 1 includes all the covariates, whereas model 2 adds the two statistically significant interaction terms to model 1.

Discussion
A stable BMI is an indicator of good health in old age, and it is a sign that the body can maintain homeostasis 23 . By contrast, changes suggest systemic deterioration, including both the decrease and the increase in BMI 24 . In the SABE Project data, BMI tended to decrease with age, but not equally across groups. The decrease rate was faster for those who had two or more chronic conditions and those who were unmarried.
Cad. Saúde Pública 2021; 37(12):e00081320 Table 2 The association between body mass index (BMI) and sociodemographic characteristics, self-reported health conditions, behavioral aspects and mortality.   Increased BMI is a known risk factor for developing chronic diseases such as diabetes and hypertension 11,25 . This could explain the general tendency towards higher BMI among those with multimorbidity. Data from this work has shown that from the age of 85 years onwards, individuals with two or more chronic diseases tend to have a lower BMI -for those with none or one disease. Diseases such as cancer and lung diseases, due to catabolic or inflammatory stress 26 , can increase the energetic expenditure, leading to lower body weight and BMI. This process may become more evident at advanced ages or with disease progression.
In this work, participants aged over 70 years old had a lower BMI when they were not married. Some who are married may have better assistance on daily tasks, such as shopping and cooking, affecting healthy weight maintenance. In the United States, older adults who lived alone without Cad. Saúde Pública 2021; 37(12):e00081320 family or friends nearby had the lowest consumption of fruits and vegetables 27 . In Japan, individuals who had difficulty to shop, which is more common among older individuals and people with culinary inabilities, showed worse food choices 28 .
Although the relative risks of a high BMI become less pronounced as a person ages 12 , the progressive increase in BMI and obesity among older adults, especially among women, has alarmed experts 3 . The effects of menopause 29 , and behavioral factors such as an improved ability to prepare food 30 , may be related to the higher BMI among women. This study results are similar to those of Asp et al. 31 , who reported lower body weight among older men.
Current studies indicate that obesogenic environments that stimulate deleterious habits have also contributed to increased BMI even among older individuals 32 . We showed that physical inactivity and alcohol consumption were related to higher BMI. According to Gomes 33 , major physical factors in weight gain in old age are modifiable, such as inactive lifestyles. Evidence generated from experimental studies also indicates that regular participation in physical activities promotes the preservation of skeletal muscle mass, strength, and physical function with advancing age 34 .
In contrast to the results found in this study, other studies 35 have shown that older people who consume alcohol have lower BMI than those who do not. Because alcohol affects inhibition of brain regions, such as the hypothalamus, large amounts of alcohol may reduce food intake 36 , while limited amounts of alcohol may increase appetite 37 . Also, because alcohol effects on blood can be longlasting in older people, even small amounts of alcohol can exacerbate chronic diseases occurrence and development 38 .
Reports of decreased food intake and smoking, on the other hand, were associated with lower levels of BMI. In addition to conditions directly related to food consumption, such as reduced odor and taste, delayed gastric emptying, pathological conditions (cancer, lung diseases, dementia), and psychosocial factors (loneliness and mourning) also contribute to weight loss in old age 39 . A proinflammatory effect and reductions in taste and food intake 40 may also be related to weight loss among smokers.
Debate continues about the best weight, or ideal weight, in old age. For example, it is postulated that the weight that maximizes survival would increase with age 3 . However, it is also possible that people who are susceptible to the adverse effects of obesity die early as weight does not increase survival, such as those who survive to old age may be resistant to the detrimental effects of obesity 3 . Our results demonstrated a negative association between mortality and BMI. Other studies also found that not only weight loss 41 but also overall weight changes -weight gain and weight fluctuations -were associated with higher mortality 12,42 .
Obesity and related health problems are becoming growing problems in low-and middle-income countries 43 , and addressing these problems in contexts of poverty, social inequalities, and inadequate access to health care 18 is a challenge. Still, there is little information on factors associated with BMI among older adults in these countries. Our study addresses this gap. Another strength of our study is the use of measured weight and height rather than self-reported measures. The limitations of this study include the fact that height was considered constant during the study period. Also, the classification of respondents with chronic diseases relied upon self-report, although the wording of the question specified physician diagnosis.
Obese older adults are a rising demographic and an important focus for further research. The convergence of aging and increases in BMI will carry significant financial and societal burdens, especially in developing countries. Public agencies and governments must be prepared for this new reality.

Conflicts of interest
The authors declare no conflict of interest.