Sample and procedures

We analysed data from the Health, Well-being, and Aging Study (Saúde, Bem-estar e Envelhecimento – SABE, as it is known in Portuguese), a population-based longitudinal observational study conducted in São Paulo, Brazil, focusing on individuals aged 60 years and older. Details regarding the study design and methodology have been previously described [18, 19].

Briefly, SABE employs a longitudinal, representative design of the municipality of São Paulo, incorporating follow-up assessments and probabilistic refreshment samples at each wave. Baseline data collection (Wave 1) occurred in 2000 with 2,143 respondents. In Wave 2 (2006), 1,115 participants were reinterviewed, and 298 new individuals aged 60–64 were added. Wave 3 (2010) assessed 990 returning participants and 355 new recruits. By Wave 4 (2015), three hundred and eighty-two participants from the 2006 cohort and 456 from the 2010 cohort were reinterviewed. The present study utilized data from three waves spanning a nine-year period: 2006 (baseline), 2010, and 2015. These corresponded to follow-up intervals of approximately four years (2006–2010) and five years (2010–2015). Since frailty assessments were not performed in 2000, we considered Wave 2 (2006) the study baseline. At this wave, 1,413 older adults had complete cognition data, including both reinterviewed participants from 2000 and new recruits in 2006.

The study was approved by the University of São Paulo Research Ethics Committee (COEP 67/99), and the National Council of Ethics in Research (CONEP 315/99). Participation was voluntary, with signed informed consent obtained from all participants in each wave.

Instruments and measuresCognitive function

Cognitive function was assessed using the abbreviated version of the Mini-Mental State Examination (MMSE), containing 13 items and a maximum score of 19 points [20]. Compared with the original 30-point MMSE, this abbreviated version retained core domains of global cognition, including orientation, memory, attention, language, and visuoconstructive ability, while reducing the number and complexity of tasks to facilitate administration in populations with low educational attainment and in large epidemiological studies. This version was developed within the World Health Organization Age-Associated Dementias project with the aim of minimizing the influence of schooling. Based on the original validation study, scores of 12 or fewer were considered indicative of cognitive deficit, whereas scores of 13 or higher indicated no cognitive deficit, with reported sensitivity of 93.8% and specificity of 93.9% [20]. This abbreviated MMSE was used as a brief global cognitive screening instrument in the present study. The MMSE was administered in 2006 (baseline), 2010, and 2015.

Predictor variables and covariates

Frailty was evaluated according to the frailty phenotype proposed by Fried et al. [21], consisting of five components, which are:

Unintentional weight loss: the question “In the last three months, have you lost weight without dieting?” was used. Older persons who reported having lost more than 3 kg scored on this criterion; Self-reported fatigue: was assessed using two items derived from the Centre for Epidemiologic Studies Depression Scale (CES-D): (a) “In the last week, how often have you felt that everything you did required a lot of effort?” and (b) “In the last week, how often have you felt that you could not get going?”. Participants who answered “2” or “3” to at least one of the two questions were classified as positive for this component; Weakness: was assessed based on handgrip strength measured with a dynamometer. Individuals were considered to have weakness if their maximum grip strength fell within the lowest 20%, stratified by sex and body mass index; Slow walking speed: this was obtained by the three-meter walk test, part of the Short Physical Performance Battery Assessment Lower Extremity Function. The older adults who scored in this component are in the highest quintile of distribution (longest time taken to cover the distance), stratified by sex and median height value; Low physical activity level: measured using the International Physical Activity Questionnaire (IPAQ). Participants self-reported information on: walking, moderate activities (light cycling, swimming, dancing, light aerobic exercises, playing recreational volleyball, carrying light weights, working on household, backyard, or garden tasks—sweeping, vacuuming, tending the garden), and vigorous activities (running, aerobic exercises, playing soccer, fast cycling, using a treadmill, playing basketball, working on heavy household, backyard, or garden tasks, and carrying heavy weights). Firstly, the duration (in minutes) for each activity was computed, with values exceeding 180 min truncated. A metabolic equivalent (MET) classification was then assigned: 3.3 METs for walking, 4.0 METs for moderate activities, and 8.0 METs for vigorous activities. The weekly total METs were calculated by summing the MET values for each activity type, multiplied by the duration (in minutes), the number of days the activity was performed. Calorie expenditure was then determined by multiplying the total METs by the participant’s weight (in kg) divided by 60. Weekly calorie expenditure was stratified into sex-specific quintiles, and individuals in the lowest 20%—≤390.5 kcal/week for men and ≤478.15 kcal/week for women—were classified as frail according to the low physical activity criterion [22].

As recommended, participants meeting three or more components were classified as frail; non-frail, who showed none of the five phenotype components; and pre-frail, those who presented one or two phenotype components.

Sociodemographic characteristics

All sociodemographic characteristics were assessed at baseline (2006). The categorization was standardized as follows: sex (female and male), age range (60–69, 70–79 and 80 years and over).

Medical conditions

Multimorbidity: Participants responded to dichotomous items (yes or no) indicating whether a physician had diagnosed any of the following noncommunicable chronic diseases in the past 12 months: systemic arterial hypertension, diabetes mellitus, heart disease, cerebrovascular disease, osteoporosis, chronic respiratory disease, joint disease. Multimorbidity was defined as the coexistence of two or more of these conditions.

Psychosocial characteristics

Depressive symptoms were assessed separately using the 15-item Geriatric Depression Scale (GDS-15). Scores of six or higher may suggest the presence of major depression.

Death reports

Deaths that occurred between the years 2006 and 2015 were verified based on information provided by relatives or neighbours during home visits for each subsequent data collection. These data were then cross-referenced with death records from the Mortality Information Improvement Program, the State Data Analysis System, and the Mortality Information System.

Statistical analysis

Baseline characteristics of participants were described using means and standard deviations, frequencies and proportions, median and interval interquartile where appropriate. The normality of data distribution was verified using the Kolmogorov–Smirnov test, which showed that most ordinal variables were non-normally distributed. Consequently, Mann–Whitney test was employed to compare medians between the two groups (with cognitive deficit and without cognitive deficit). Pearson’s chi-square test, with a 5% significance level, was used to assess the associations between cognitive status and nominal variables. Cognitive transitions were categorized into four patterns: (1) worsening group—participants who transitioned from no cognitive deficit to cognitive deficit, from no cognitive deficit to death, or from cognitive deficit to death; (2) stable group—participants who maintained the same cognitive status across waves, either without cognitive deficit or with persistent cognitive deficit; (3) improving group—participants who transitioned from cognitive deficit to no cognitive deficit; and (4) fluctuating group—participants who exhibited alternating cognitive status across waves, such as transitions from no cognitive deficit to cognitive deficit and back, or vice versa. To identify factors independently associated with transitions in cognitive status between waves, we used Markov stationary models fitted via multinomial logistic regression. This regression treated the cognitive status at time t (i.e., 2010 wave or 2015 wave) as a multinomial dependent variable with three categories: no cognitive deficits (reference), cognitive deficits, and death. Cognitive status at time t-1 (i.e., 2006 or 2010 wave), along with covariates including age, sex, frailty, morbidity, and depressive symptoms, were included as independent variables. The initial model included all candidate covariates listed in Table 1. Interaction terms were incorporated into the model to assess whether the effect of covariates varied by baseline cognitive status. A likelihood ratio test was used to assess the significance of the interaction terms, by comparing model fit with and without the interaction terms. Odds Ratio (OR) and the 95% confidence interval (CI) are reported for the “cognitive deficits” and "death" categories, relative to the "no cognitive deficits" category. All analyses were conducted using Stata version 18.0.

Table 1 Baseline characteristics of the study sample in 2006 (N = 1413)