Socioeconomic,Disease Burden,Physical Functioning,Psychosocial,and Environmental Factors Associated With Mortality Among Older Adults: The Brazilian Longitudinal Study of Ageing (ELSI-Brazil)

Abstract

Objectives:

There is little nationally representative information about factors associated with longevity among older Brazilians.

Methods:

Baseline survey data from the Brazilian longitudinal Study of Aging (ELSI-Brazil) were linked to vital statistics systems. Mortality rates and life expectancy estimates were calculated and compared to official sources. Cox Proportional Hazards models and Population Attributable Fractions (PAFs) identified significant predictors of mortality. Results: Calculated mortality rates and life expectancy estimates were similar to official statistics for most ages with higher risk of death among older ages, as expected. High School completion, being partnered, and female sex were negatively associated with mortality, while being underweight, previous diagnosis of a chronic condition, having any functional limitations, poor self-rated health, low grip strength, and smoking were all associated with higher mortality risk. Discussion: The ELSI-Brazil study has potential to identify factors associated with longevity and to inform programs and policies designed to enhance healthy aging among older adults in Brazil.

Keywords

aging mortality Brazil cohort study

Introduction

There is considerable evidence that biological, economic, environmental, and social factors are associated with mortality among aging cohorts (Liao et al., 2018; Puterman et al., 2020). But the bulk of this evidence has come primarily from higher-income settings. In most low- and middle-income countries, the study of mortality continues to rely on vital statistics that, although essential, do not provide sufficient information to assess underlying sociodemographic and health-related factors linked with the risk of morbidity, disability, or mortality (National Academies of Sciences & Medicine, 2018). The accelerated aging of populations in middle income countries has led to the development of a series of cohort studies of ageing (many of which are patterned after the US Health and Retirement Study) in countries such as Brazil, China, India, Indonesia, Mexico, and South Africa. These studies provide an important opportunity to better understand the mechanisms and consequences of risk factors over the life course and to assess whether the same factors associated with mortality and longevity that have been identified in higher income setting apply to these middle-income country contexts (Lee et al., 2021).

Understanding determinants of the aging process is essential for guiding investments in improving health and the policy environment. This is particularly relevant in Brazil, which has the world’s sixth largest population (over 212 million) with 17.8 million adults (8.5%) aged 65 and over. In 2020, life expectancy was estimated at 76 years, placing it squarely within the median for the Latin America and the Caribbean region. While the socio-economic context of the country continues to fluctuate, its nominal GDP (1.87 trillion) has been ranked within the top 10 globally for several years. Hallmark features of Brazil’s social and economic contexts include its high level of income inequality (Gini 48.9 in 2020, making it the 16th highest in terms of income inequality). Brazil is also known as a leader in innovative social policies, including a universal health system, a large conditional cash transfer program, and a national pension program, among others. These investments have already made changes to population health outcomes (Hone et al., 2017, 2019), although their cumulative impact on older adults has yet to be fully understood.

In recent decades, Brazil has also made improvements in enhancing the completeness and quality of vital statistics and other data systems (Marinho et al., 2019). But such systems still have limitations with respect to age reporting, documentation of education, and other socioeconomic variables on death certificates (Júnior et al., 2021), accurate identification of specific causes of death, and subnational variations in completeness and quality (Marinho et al., 2019; Teixeira et al., 2019; Turra et al., 2016). Even with improvements in the quality of vital statistics, there would still be gaps in understanding the aging process in Brazil because many relevant risk factors are not routinely collected in this data system (e.g., biological, socioeconomic, and social factors).

In this study, we aim to begin to fill these gaps in knowledge by assessing rates and predictors of all-cause mortality derived from the Brazilian Longitudinal Study of Aging (ELSI-Brazil), a nationally representative population-based cohort study of Brazilian adults aged 50 years and older.

Methods

This study uses data from the first wave of the ELSI-Brazil survey. The ELSI-Brazil sampling plan combined stratification of primary sampling units (municipalities), census tracts, and households. The baseline sample comprised 10,000 older adults (9412 participated), residing in 70 municipalities from all five Brazilian geographic regions. Details of the ELSI study (See http://elsi.cpqrr.fiocruz.br) have been described elsewhere (Lima-Costa et al., 2022).

The period under examination extends from the participant’s completion of the ELSI-Brazil baseline interview (May 2015 to October 2016) to completion of the second wave (July 2019 to March 2021). The flow diagram (See Online Appendix Figure 1) shows the construction of the analytic sample. Of the 9412 baseline participants, 6172 (65.6%) were retained for the second wave, 2533 were lost to follow-up and 707 had died (with a date of death informed by proxy). Using a probabilistic linkage method that included the participant’s name, sex, birth date, municipality of residence, and their mother’s name, information on those who died and those lost to follow-up was linked to the Brazilian Mortality Information System (SIM). The linkage algorithm identified that 263 (10.4%) of those lost to follow had died and among the 707 deaths reported by proxy, 609 (89.9%) were confirmed with a specific cause of death identified through this linkage. The number of baseline participants who died therefore totaled 970 (707 + 263).

Selection of variables for descriptive statistics and multivariable models was guided by the domains present in the WHO framework for Healthy Aging (Lloyd-Sherlock et al., 2019; World Health Organisation, 2017). These domains include the following: demographic factors (age, sex, self-reported skin color), socioeconomic status (years of formal schooling; per capita household income, home ownership, and occupational classification). Health status measures included fair/poor self-rated health, measured Body Mass Index (in categories), cognitive health (late memory word recall), and chronic conditions (self-reported prior medical diagnosis of cardiovascular disease—heart attack, angina or congestive heart failure, diabetes, stroke, cancer, depression, and measured hypertension (>140/90). Functional ability includes Activities of Daily Living (ADL) limitations, Instrumental Activities of Daily Living (IADL) limitations, and measured grip strength (average of two highest values) (de Souza Moreira et al., 2022). We also assessed behavioral factors (current or ever smoker vs. never), social support (partnered vs. single), and the physical environment (geographic region, and urban vs. rural residence).

Descriptive statistics comparing those who survived to those who did not were calculated as survey weighted proportions and means. Differences between groups are tested using Chi-Squared tests with a Rao–Scott correction for proportions and t-tests for differences between means. A multinomial logistic model was used to test differences between characteristics of those who were lost to follow-up with respect to those who died and those who were retained.

Demographic analyses include calculation of death rates by age and sex and making comparisons to official statistics as published by WHO in 5 year age groups (World Health Organization, 2022). We additionally calculate life expectancy estimates at each age (e_x) and compare results with official data from the Brazilian Institute of Geography and Statistics (IBGE) available in 1 year age groups (Brazilian Institute of Geography and Statistics, 2022). Life expectancy calculations rely on the assumptions that those who died did so on average half-way between consecutive ages (Preston et al., 2001). The life table is based on standard assumptions that the death rate is constant after age 80 and that there were no person-years lived above age 100 in the population.

To understand the role of different factors on mortality, we employ Cox proportional hazards regression models. Because some variables were missing (up to 5% for some measured variables), we implemented multiple imputation via chained equations to create 20 datasets, then used Stata’s “svy” and “mi” commands to perform the survey-adjusted cox regression model drawing from the imputed datasets (StataCorp., 2021). The use of multiple imputation precludes the use of standard goodness of fit test measures (such as AIC/BIC or likelihood ratio tests), so variables introduced in the nested multivariable models were those that presented statistically significant differences among those who died and those who did not, as identified through descriptive (bivariate) analyses.

Finally, we calculate the Population Attributable Fractions (PAFs) for variables found to be statistically significantly associated with mortality in the Cox proportional hazards models. The PAF combines both the strength of the association of the risk factor with mortality as well as the prevalence of the risk factor in the population to derive an estimate of the impact on mortality if that risk factor were to be eliminated (Mansournia & Altman, 2018).

The ELSI-Brazil study was approved by the Committee on Research Ethics of the Fundação Oswaldo Cruz, Minas Gerais (CAAE: 34649814.3.0000.509).

Results

Table 1 provides descriptive statistics, comparing those who died and those who survived the period. The average age was 62 years and was higher (70) among those who died. There was a greater proportion of males in the group who died, but no difference in terms of skin color (white vs. other). There was a statistically significant difference in the fraction of those who survived and those who died in terms of formal education with those who survived tending to have a higher number of years of formal education. Being partnered and in higher household income tertiles were more common among those who survived than those who did not. There was no difference between groups in terms of home ownership. All indicators of health status, except depression, differed significantly among those who survived and those who did not, with generally higher levels of chronic conditions, poorer self-rated health, ADL, and IADL limitations among those who did not survive. Body mass index also differed between groups, with a larger proportion of those who died being underweight. Grip strength and memory (word recall) significantly differed across the groups as did smoking status. Geographic location whether urban or rural or whether the residence was in the richer south east and south regions of the country versus the north northeast and Midwest regions did not differ between groups.

Table 1.

Characteristics of Those Who Died With Those Who Survived.

	Survived (N = 8442)	Died (N = 970)	Total (N = 9412)	Difference¹
Age (mean)	61.74	70.01	62.5	.0000
Age (mean)	[60.94, 62.53]	[68.69, 71.32]	[61.67, 63.33]
Male	45.35	52.9	46.05	.0002
Male	[42.30, 48.44]	[48.80, 56.96]	[43.11, 49.02]
Female	54.65	47.1	53.95
Female	[51.56, 57.70]	[43.04, 51.20]	[50.98, 56.89]
White skin color	41.32	40.42	41.24	.6209
White skin color	[35.85, 47.02]	[34.79, 46.31]	[35.85, 46.85]
Other skin color	58.68	59.58	58.76
Other skin color	[52.98, 64.15]	[53.69, 65.21]	[53.15, 64.15]
No formal schooling	12.51	21.04	13.3	.0000
No formal schooling	[10.35, 15.06]	[16.25, 26.79]	[10.98, 16.02]
1–4 years of schooling	37.36	46.92	38.24
1–4 years of schooling	[34.96, 39.83]	[42.57, 51.31]	[36.01, 40.52]
5–10 years schooling	25.12	20.01	24.65
5–10 years schooling	[22.69, 27.72]	[16.54, 23.99]	[22.31, 27.14]
11+ years schooling	25	12.04	23.81
11+ years schooling	[22.76, 27.39]	[9.46, 15.19]	[21.71, 26.04]
Single	34.95	51.96	36.53	.0000
Single	[32.08, 37.94]	[48.15, 55.75]	[33.69, 39.47]
Partnered	65.05	48.04	63.47
Partnered	[62.06, 67.92]	[44.25, 51.85]	[60.53, 66.31]
Income tertile 1 (lowest)	33.06	36.02	33.34	.0079
Income tertile 1 (lowest)	[29.30, 37.06]	[30.89, 41.49]	[29.62, 37.28]
Income tertile 2	33.02	36.48	33.34
Income tertile 2	[31.42, 34.66]	[32.55, 40.59]	[31.78, 34.94]
Income tertile 3 (highest)	33.92	27.51	33.32
Income tertile 3 (highest)	[30.44, 37.58]	[23.18, 32.30]	[29.88, 36.95]
Home ownership	75.64	76.14	75.69	.7705
Home ownership	73.44, 77.72	71.67, 80.11	73.42, 77.83
Hypertension(Measured)	41.88	51.11	42.74	.0000
Hypertension(Measured)	[39.64, 44.16]	[46.63, 55.57]	[40.57, 44.94]
Diabetes	14.97	23.7	15.78	.0000
Diabetes	[13.76, 16.26]	[20.54, 27.18]	[14.59, 17.05]
Cardiovascular disease	10.81	19.86	11.64	.0000
Cardiovascular disease	[9.76, 11.94]	[16.82, 23.29]	[10.57, 12.81]
Stroke	4.67	11.85	5.33	.0000
Stroke	[4.10, 5.31]	[9.65, 14.46]	[4.74, 5.99]
Cancer	4.77	10.33	5.29	.0000
Cancer	[4.15, 5.49]	[8.30, 12.79]	[4.63, 6.03]
Depression	18.42	20.11	18.58	.2149
Good/Excellent self-	89.39	78.22	88.36	.0000
Rated health	[88.20, 90.47]	[74.67, 81.39]	[87.09, 89.52]
Fair/Poor self-rated	10.61	21.78	11.64
Health	[9.53, 11.80]	[18.61, 25.33]	[10.48, 12.91]
No ADL limitations	85.93	65.22	84.02	.0000
No ADL limitations	[84.60, 87.17]	[61.04, 69.18]	[82.65, 85.30]
1+ ADL limitations	14.07	34.78	15.98
1+ ADL limitations	[12.83, 15.40]	[30.82, 38.96]	[14.70, 17.35]
No IADL limitations	59.98	37.42	57.92	.0000
No IADL limitations	[57.26, 62.64]	[32.41, 42.72]	[55.25, 60.54]
1–2 IADL limitations	27.19	28.13	27.27
1–2 IADL limitations	[25.28, 29.18]	[24.66, 31.89]	[25.42, 29.21]
3+ IADL limitations	12.83	34.44	14.81
3+ IADL limitations	[11.51, 14.27]	[30.25, 38.89]	[13.41, 16.33]
Normal weight	28.07	34.18	28.58	.0000
Normal weight	[26.69, 29.50]	[30.44, 38.12]	[27.29, 29.92]
Underweight	1.67	6.95	2.11
Underweight	[1.37, 2.02]	[4.55, 10.46]	[1.75, 2.53]
Overweight	40.06	35.36	39.66
Overweight	[38.75, 41.38]	[31.34, 39.60]	[38.34, 41.00]
Obese	30.21	23.51	29.65
Obese	[28.73, 31.73]	[19.91 ,27.55]	[28.27, 31.06]
Grip strength (mean)	26.45	23.06	26.16	.000
Grip strength (mean)	[25.62, 27.28]	[22.23, 23.88]	[25.35, 26.97]
Late memory recall(Mean, out of 10)	2.97	2.11	2.90	.000
Late memory recall(Mean, out of 10)	[2.86, 3.09]	[1.92, 2.29]	[2.79, 3.01]
Ever smoked	53.31	65.03	54.4	.0220
Ever smoked	[51.47, 55.15]	[61.00, 68.85]	[52.54, 56.24]
Urban	84.81	83.55	84.69	.5046
Urban	[79.52, 88.92]	[76.54, 88.77]	[79.39, 88.83]
Rural	15.19	16.45	15.31
Rural	[11.08, 20.48]	[11.23, 23.46]	[11.17, 20.61]
Southeast/South region	63.66	64.52	63.74	.7119
Southeast/South region	[52.04, 73.88]	[52.88, 74.65]	[52.2, 73.89]
North/Northeast/Midwest	36.34	35.48	36.26
North/Northeast/Midwest	[26.12, 47.96]	[25.35, 47.12]	[26.11, 47.8]

Numbers are weighted proportions, unless otherwise specified.

1 Statistical significance obtained from t-tests and Chi-square tests with Rao-Scott adjustment to account for the complex sample design and weighting.

The Online Appendix Tables 1 and 2 focus on characteristics of those lost to follow-up. It shows that, as compared to either those who died or those who were retained, those lost to follow-up were less likely to be in the older age groups, more likely to be single, less likely to own a home, more likely to have competed Middle or High School, and more likely to reside in the South or Southeast regions of the country, than those who were retained. In the fully adjusted multinomial logistic regression model (Online Appendix table 2), there was no statistically significant difference between those retained and those lost to follow up in terms of any health status indicator.

Table 2 compares mortality rates by 5 year age groups and sex. The table shows that for both sexes, the WHO rates for 2015 and 2019 largely fall within the 95% confidence Intervals for the ELSI estimates. Exceptions include the 55–59 age group in which ESLI estimates are higher than WHO rates and the 70–74 and 85+ age groups in which ELSI estimates are lower than WHO rates. A similar pattern is seen when mortality rates are broken down by sex.

Table 2.

Comparison of Death Rates by Age, Sex, Data Source, and Year.

Both sexes	WHO 2015	ELSI	ELSI 95% CI		WHO 2019
50–54 years	5.7	6.9	5.4	8.9	5.3
55–59 years	8.4	12.5	10.0	15.7	7.9
60–64 years	12.3	13.5	10.9	16.9	12.0
65–69 years	18.6	17.9	14.6	22.1	17.8
70–74 years	28.7	19.2	15.3	24.4	27.7
75–79 years	44.7	41.4	34.1	50.7	42.3
80–84 years	74.9	53.4	43.2	66.7	66.5
85+ years	165.5	102.9	83.9	127.0	155.3

Males	WHO 2015	ELSI	ELSI 95% CI		WHO 2019
50–54 years	7.5	8.4	6.0	12.1	6.9
55–59 years	11.0	17.6	13.3	23.6	10.3
60–64 years	15.9	15.4	11.2	21.6	15.6
65–69 years	23.8	22.9	17.3	30.9	23.1
70–74 years	36.3	20.2	14.6	28.6	35.0
75–79 years	55.1	61.1	46.9	81.0	51.8
80–84 years	90.9	52.8	36.3	78.6	80.4
85+ years	191.1	119.3	86.5	166.7	180.4

Females	WHO 2015	ELSI	ELSI 95% CI		WHO 2019
50–54 years	4.0	5.6	3.9	8.3	3.8
55–59 years	6.1	7.9	5.5	11.8	5.8
60–64 years	9.1	11.8	8.8	16.2	8.8
65–69 years	14.2	14.0	10.5	19.0	13.4
70–74 years	22.7	18.4	13.4	26.0	21.8
75–79 years	37.1	28.6	22.0	37.8	35.4
80–84 years	64.6	53.8	41.5	70.6	57.3
85+ years	152.3	92.2	70.8	121.7	142.2

To assess mortality at a finer level of detail, Online Appendix Figure 2 plots single-years of age-specific death rate estimates from ELSI along with their 95% confidence intervals against values from the IBGE at the beginning (2015), middle (2018), and end (2020) of the period. For both men and women and for nearly every age, the IBGE estimates lie within the ELSI confidence intervals. The notable exception is among those aged 80+ where the ELSI estimates were significantly lower than official figures.

When comparing life expectancy at each age (e_x), for most ages, the ELSI estimate lies between the 2015 and 2020 IBGE estimates, except for slightly lower estimates in younger ages (50–55) and older (75–80+ years) as can be seen in Figure 1 and Online Appendix table 3.

Figure 1.

Comparison of life expectancy at age x (e_x) for both sexes combined, by data source.

Table 3 presents results from the Cox proportional hazards models. In the first model, older age showed a positive gradient in its association with mortality, but being female and being partnered versus being single were both protective factors. In model 2, completing High School (11+ years of formal schooling) was negatively associated with mortality, but household income was not statistically significant. In model 3, all chronic conditions (diabetes, hypertension, cardiovascular disease, stroke, and cancer) were positively associated with greater risk of death. In model 4, poor self-rated health, being underweight (compared with normal weight), and being an ever smoker were positively associated with mortality. In model 5, higher grip strength was negatively associated with mortality, while having any ADL limitation and three or more IADL limitations showed a positive association.

Table 3.

Results From Cox Proportional Hazards Models.

Variable	Model 1	Model 2	Model 3	Model 4	Model 5
Age 54 to 59 (v 50–54)	1.82**	1.76*	1.68*	1.7*	1.57
Age 54 to 59 (v 50–54)	1.16, 2.85	1.11, 2.78	1.06, 2.68	1.05, 2.73	.97, 2.57
60–64	1.94***	1.87***	1.71**	1.73**	1.59*
60–64	1.38, 2.73	1.32, 2.64	1.20, 2.44	1.21, 2.47	1.11, 2.27
65–69	2.49***	2.33***	2.02***	2.05***	1.84***
65–69	1.81, 3.43	1.68, 3.22	1.46, 2.80	1.48, 2.84	1.33, 2.57
70–74	2.66***	2.45***	2.07***	2.16***	1.84**
70–74	1.81, 3.93	1.66, 3.60	1.40, 3.07	1.46, 3.18	1.24, 2.72
65–79	5.75***	5.07***	4.02***	4.17***	3.2***
65–79	4.06, 8.13	3.59, 7.16	2.79, 5.78	2.90, 6.01	2.20, 4.66
80+	9.46***	8.38***	6.93***	7.13***	4.53***
80+	6.62, 13.53	5.85, 12.00	4.81, 10.00	4.95, 10.26	3.09, 6.64
Female	.55***	.55***	.55***	.62***	.42***
Female	.47, 0.65	.46, 0.65	.47, 0.66	.52, 0.74	.35, 0.52
Partnered (v single)	.58***	.58***	.56***	.58***	.61***
Partnered (v single)	.49, 0.69	.48, 0.70	.47, 0.68	.48, 0.71	.51, 0.74
1–4 years schooling		1.05	1.02	1.08	1.18
(v none)		.84, 1.30	.82, 1.27	.87, 1.35	.94, 1.50
5–10 years		.94	0.9	1.01	1.19
5–10 years		.71, 1.24	.68, 1.20	.76, 1.34	.88, 1.61
11+ years		.58***	.58***	.68*	.86
11+ years		.43, 0.80	.42, 0.79	.49, 0.93	.60, 1.22
Household income		.98	1.02	1.08	1.08
Tertile 2 (v poorest)		.81, 1.19	.84, 1.24	.89, 1.30	.89, 1.32
Tertile 3 (richest)		.88	0.9	.97	1.08
Tertile 3 (richest)		.72,1.08	.73, 1.10	.80, 1.19	.88, 1.32
Diabetes			1.47***	1.48***	1.38**
Diabetes			1.21, 1.79	1.21, 1.80	1.13, 1.69
Hypertension			1.2*	1.22*	1.27**
(measured)			1.01, 1.43	1.03, 1.45	1.07, 1.51
Cardiovascular disease			1.44**	1.33**	1.27*
Cardiovascular disease			1.15, 1.79	1.07, 1.64	1.03, 1.57
Stroke			1.67**	1.57**	1.15
Stroke			1.29, 2.16	1.21, 2.04	.88, 1.50
Cancer			1.69***	1.64***	1.6***
Cancer			1.35, 2.11	1.31, 2.06	1.27, 2.01
Poor self-rated health				1.75***	1.32*
(v good/excellent)				1.45, 2.10	1.06, 1.64
Underweight (measured)				2.2***	2.08**
(v normal weight)				1.40, 3.46	1.35, 3.21
Overweight				.88	.91
Overweight				.72, 1.07	.74, 1.11
Obese				.86	.88
Obese				.67, 1.12	.68, 1.13
Ever smoker				1.39***	1.41***
Ever smoker				1.17, 1.65	1.19, 1.67
Late memory recall					.95
Late memory recall					.90, 1.01
Grip strength					.97***
Grip strength					.96, 0.98
Any ADL					1.49***
(v none)					1.23, 1.80
1–2 IADL					1.13
(vs. none)					.90, 1.43
3+ IADL					1.53**
3+ IADL					1.14, 2.06
N	9412	9412	9412	9412	9412

* = p < .05; **p < .01; ***p < .001.

PAF represent the change in the probability of death if the factor were to be eliminated.

PAF calculated only for statistically significant predictors of mortality, based on Model five in Table 3.

Figure 2 presents PAF estimates for those variables found to be statistically significant predictors of mortality in the Cox proportional hazards models (Model 5 in Table 3). These measures represent the change in the probability of death if the factor were to be eliminated, thus combining the relative risk of death but also the prevalence of the condition within the population. Two factors (11 or more years of schooling and being partnered vs. being single) both had positive PAFs, meaning that if they were to be eliminated, then mortality would increase in the population. In contrast, the largest population attributable fraction leading to lower mortality would come from, in descending order of magnitude, elimination of underweight, cancer, diabetes, 3+ IADL limitations, fair poor self-rated health, cardiovascular disease, low grip strength, any ADL limitation, hypertension, and smoking.

Figure 2.

Population Attributable Fraction (PAF) estimates and 95% confidence intervals.

Discussion

In this study, we demonstrate that ELSI-Brazil can be used to provide measures of mortality and life expectancy for older Brazilians. These estimates largely coincide with official statistics, with the important exception of underestimating mortality among the oldest age groups (80 and above) and slightly overestimating mortality among the youngest age groups. These differences do not appear to be driven by characteristics of those lost to follow-up and may instead be a function of the difficulty of recruiting older, sicker adults into the survey. Importantly, estimates of life expectancy by age were close to official estimates for both sexes.

Specific factors associated with mortality largely align with the literature. As expected, women tend to have lower mortality risk, with an overall higher life expectancy at every age as compared to men. Higher educational attainment was negatively associated with mortality and showed a PAF of .28, suggesting it is an important determinant of longevity. This is consistent with previous analyses that have quantified the importance of educational attainment for mobility (Nascimento et al., 2018) and changes in life expectancy, especially for Brazilian women (Turra et al., 2016). Interestingly, there was no evidence of a mortality gradient by educational attainment, perhaps because there is a low average level of education especially among older members of the cohort—half the sample had 4 years or less years of formal schooling, as previously reported (Andrade & Lopez-Ortega, 2017). While there has been evidence of a more pronounced SES gradient for self-rated health and ADL limitations among older Brazilians (Lima-Costa et al., 2012), to date, there have not been sufficient data to assess the individual-level relationship between mortality and educational attainment in a nationally representative sample of older adults.

Chronic conditions and other risk factors were also largely consistent with the literature, such as those based on data from the Study of Aging and Health (SABE) (Bernardes et al., 2021). Measured hypertension, poor self-rated health, and previous medical diagnoses of cardiovascular disease, stroke, and cancer were all predictive of greater mortality risk, similar to results found in a study of mortality risk among older adults in four countries (Costa Rica, England, US, Taiwan) (Goldman et al., 2016). A study using another Brazilian aging cohort based in a small town (Bambuí) compares results with those obtained from the English Longitudinal Study of Aging (ELSA) and found that hypertension and diabetes were strongly associated with 6 year mortality, with smoking identified as an important contributing factor (de Oliveira et al., 2016). Given the higher prevalence and greater mortality risk among Brazilians (as compared to British older adults), the authors found that such factors had a greater contribution to population attributable risks in Brazil than in England. Another study compared older adults who participated in a different Brazilian cohort, Siga-Bagé (based in Southern Brazil) and ELSA. Kessler et al (2020) confirmed that mortality risk was generally higher among Brazilian as compared with English older adults and identified smoking as an important modifiable risk factor. Unlike this study, Kessler et al did not find that hypertension or diabetes were statistically significantly related to mortality risk. They did, however, find that obesity was not associated with mortality, but did not assess underweight (Kessler et al., 2020). This study found underweight to be one of the strongest predictors of mortality at older ages. This is perhaps due to its relationship with frailty, which has been found to be associated with increased mortality risk among older Brazilians in the SABE study (Fhon et al., 2018).

This study has several strengths. It relies on nationally representative cohort data with rich information on demographic, health status, financial, and social domains that allowed us to investigate relationships with mortality hitherto unexplored in a national sample. Moreover, comparison of mortality and life expectancy estimates to official sources helps to establish the study results’ generalizability.

Study limitations include the fact that a number of questions are based on self-report and we cannot rule out the risk of recall bias that may have affected respondents’ ability to recall certain facts. Moreover, approximately 2% of responses were supplied via a proxy respondent, usually the respondent’s spouse or caretaker. Our sample is made up of community-dwelling older adults and excludes highly vulnerable and sick individuals including those residing in hospitals or other institutional settings. Sampling weights have been calculated to make estimates from ELSI data nationally representative based on demographic and geographic characteristics, but not in regards to the health status of the participants. The lack of participation of older, much sicker adults in the study, a limitation of many aging cohort studies, may explain why our estimates for mortality at the oldest ages were lower than official estimates.

Finally, the COVID-19 pandemic may have affected our results. Recent work estimated a .9 year reduction of life expectancy at age 65 between 2019 and 2020 due to the pandemic (Castro et al., 2021). Other work indicates that COVID had not only a direct impact on mortality in Brazil but also had an indirect effect through other causes of death (Fernandes et al., 2023). For example, mortality related to childhood (e.g., pregnancy and childbirth and puerperium) and adulthood (e.g., diabetes and hypertension) increased while death from some conditions with high prevalence among older adults such as malignant neoplasm, heart and stroke, and pneumonia actually declined (Fernandes et al., 2023). As noted above, the lack of participation of older and much sicker respondents in ELSI, combined with the indirect effect of COVID in reducing mortality risk from some major causes of death among older adults may have contributed to lower number of deaths in our study and a consequent underestimation of death rates at older ages. In this study, we do not analyze the specific contribution of COVID-19, although there were 42 deaths confirmed due to COVID-19 (ICD-10 B34.2) among those whose records were successfully linked. This represents 4.3% of all deaths. Given that the IBGE mortality and life expectancy estimates were projections based on pre-COVID-19 data and that older age is associated with COVID-19 mortality, our estimates of death at the oldest ages may indeed be undercounted.

In conclusion, this study demonstrated the value of ELSI as it allows for the assessment of mortality risk by factors that are not typically collected in other data systems such as vital statistics. The study adds value due to its being the only existing nationally representative aging cohort in the country. ELSI-Brazil therefore provides an opportunity to understand better the contribution of factors such as educational attainment, the household economic situation, health behaviors, and a range of health conditions—separately and in combination, on mortality, longevity, disability, and inequalities in each of these outcomes for a nationally representative sample of older adults. Future research can further advance the state-of-the-art by incorporating ELSI’s mortality data (including specific causes of death) with measures of physical functioning, cognitive performance, metabolic and other biomarkers, and genomic analyses (Lima-Costa et al., 2022). The study’s ultimate objective is to harness such research for promoting healthy aging among older Brazilians and to inform the overall provision and continuous quality improvement of health and social services essential to reaching this goal.

Supplemental Material

Supplemental Material - Socioeconomic, Disease Burden, Physical Functioning, Psychosocial, and Environmental Factors Associated With Mortality Among Older Adults: The Brazilian Longitudinal Study of Ageing (ELSI-Brazil)

Supplemental Material for Socioeconomic, Disease Burden, Physical Functioning, Psychosocial, and Environmental Factors Associated With Mortality Among Older Adults: The Brazilian Longitudinal Study of Ageing (ELSI-Brazil) James Macinko, Hiram Beltrán-Sánchez, Juliana Vaz de Melo Mambrini, and Maria Fernanda Lima-Costa in Journal of Aging and Health.

Footnotes

Acknowledgements

We thank the team from the Health Surveillance Secretariat (Secretaria de Vigilância em Saúde, in Portuguese) at the Brazilian Ministry of Health for performing linkage of the ELSI-Brazil data with vital statistics data (the National Mortality Information System, SIM).

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The first and second waves of ELSI-Brazil were supported by the Brazilian Ministry of Health (DECIT/SCTIE – Department of Science and Technology from the Secretariat of Science, Technology and Strategic Inputs [Grants: 404965/2012-1 and TED 28/2017]; and COSAPI/DAPES/SAS – Healthcare Coordination of Older Adults, Department of Strategic and Programmatic Actions from the Secretariat of Health Care [Grants: 20836, 22566, 23700, 25560, 25552, and 27510]. The funding sources had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication

Data availability Statement

ELSI-Brazil baseline data are publicly available on the study website: <>.

ORCID iD

James Macinko

Supplemental Material

Supplemental material for this article is available online.

References

Andrade

F. C. D.

Lopez-Ortega

(2017). Educational differences in health among middle-aged and older adults in Brazil and Mexico. Journal of Aging and Health, 29(6), 923–950. https://doi.org/10.1177/0898264317705781

Bernardes

G. M.

Saulo

Santos

J. L. F.

da Cruz Teixeira

de Oliveira Duarte

Andrade

F. B. d.

(2021). Effect of education and multimorbidity on mortality among older adults: Findings from the health, well-being and ageing cohort study (SABE). Public Health, 201(3), 69–74. https://doi.org/10.1016/j.puhe.2021.10.001

Brazilian Institute of Geography and Statistics . (2022). Tábua completa de mortalidade para o Brasil. Retrieved from https://biblioteca.ibge.gov.br/index.php/biblioteca-catalogo?view=detalhes&id=73097. Retrieved June 12, 2022, from IBGE. https://biblioteca.ibge.gov.br/index.php/biblioteca-catalogo?view=detalhes&id=73097

Castro

M. C.

Gurzenda

Turra

C. M.

Kim

Andrasfay

Goldman

(2021). Reduction in life expectancy in Brazil after COVID-19. Nature Medicine, 27(9), 1629–1635. https://doi.org/10.1038/s41591-021-01437-z

de Oliveira

Marmot

M. G.

Demakakos

Vaz de Melo Mambrini

Peixoto

S. V.

Lima-Costa

M. F.

(2016). Mortality risk attributable to smoking, hypertension and diabetes among English and Brazilian older adults (The ELSA and Bambui cohort ageing studies). European Journal of Public Health, 26(5), 831–835. https://doi.org/10.1093/eurpub/ckv225

de Souza Moreira

de Souza Andrade

A. C.

Lustosa Torres

de Souza Braga

de Carvalho Bastone

de Melo Mambrini

J. V.

Lima-Costa

M. F.

(2022). Nationwide handgrip strength values and factors associated with muscle weakness in older adults: Findings from the Brazilian longitudinal study of aging (ELSI-Brazil). Bmc Geriatrics, 22(1), 1005. https://doi.org/10.1186/s12877-022-03721-0

Fernandes

Turra

C. M.

França

G. V. A.

Castro

M. C.

(2023). Mortality by cause of death in Brazil: Effects of the COVID-19 pandemic and contribution to changes in life expectancy at birth. medRxiv, 2023.02.13. 23285842. https://doi.org/10.1101/2023.02.13.23285842

Fhon

J. R. S.

Rodrigues

R. A. P.

Santos

J. L. F.

Diniz

M. A.

Santos

E. B. D.

Almeida

V. C.

Giacomini

S. B. L.

(2018). Factors associated with frailty in older adults: A longitudinal study. Revista de saude publica, 52, 74. https://doi.org/10.11606/S1518-8787.2018052000497

Goldman

Glei

D. A.

Weinstein

(2016). What matters most for predicting survival? A multinational population-based cohort study. PLoS One, 11(7), e0159273. https://doi.org/10.1371/journal.pone.0159273

10.

Hone

Mirelman

A. J.

Rasella

Paes-Sousa

Barreto

M. L.

Rocha

Millett

(2019). Effect of economic recession and impact of health and social protection expenditures on adult mortality: A longitudinal analysis of 5565 Brazilian municipalities. The Lancet. Global Health, 7(11), article e1575–e1583. https://doi.org/10.1016/S2214-109X(19)30409-7

11.

Hone

Rasella

Barreto

Atun

Majeed

Millett

(2017). Large reductions in amenable mortality associated with Brazil's primary Care expansion and strong health governance. Health Affairs (Project Hope), 36(1), 149–158. https://doi.org/10.1377/hlthaff.2016.0966

12.

Kessler

Thume

Scholes

Marmot

Facchini

L. A.

Nunes

B. P.

Machado

K. P.

Soares

M. U.

de Oliveira

(2020). Modifiable risk factors for 9-year mortality in older English and Brazilian adults: The ELSA and SIGa-Bage ageing cohorts. Scientific Reports, 10(1), 4375. https://doi.org/10.1038/s41598-020-61127-7

13.

Lee

Phillips

Wilkens

Gateway to Global Aging Data Team (2021). Gateway to global aging data: Resources for cross-national comparisons of family, social environment, and healthy aging. The Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 76(1), S5–S16. https://doi.org/10.1093/geronb/gbab050

14.

Liao

Muniz-Terrera

Scholes

Hao

Chen

Y. M.

(2018). Lifestyle index for mortality prediction using multiple ageing cohorts in the USA, UK and Europe. Scientific Reports, 8(1), 6644. https://doi.org/10.1038/s41598-018-24778-1

15.

Lima-Costa

M. F.

de Melo Mambrini

J. V.

Bof de Andrade

de Souza

P. R. B.

de Vasconcellos

M. T. L.

Neri

A. L.

Castro-Costa

Macinko

de Oliveira

(2023). Cohort profile: The Brazilian longitudinal study of ageing (ELSI-Brazil). International Journal of Epidemiology, 52(1), e57–e65. https://doi.org/10.1093/ije/dyac132

16.

Lima-Costa

M. F.

De Oliveira

Macinko

Marmot

(2012). Socioeconomic inequalities in health in older adults in Brazil and England. American Journal of Public Health, 102(8), 1535–1541. https://doi.org/10.2105/AJPH.2012.300765

17.

Lloyd-Sherlock

Kalache

Kirkwood

McKee

Prince

(2019). WHO's proposal for a decade of healthy ageing. Lancet, 394(10215), 2152–2153. https://doi.org/10.1016/S0140-6736(19)32522-X

18.

Mansournia

M. A.

Altman

D. G.

(2018). Population attributable fraction. BMJ, 360. https://doi.org/10.1136/bmj.k757

19.

Marinho

M. F.

Franca

E. B.

Teixeira

R. A.

Ishitani

L. H.

Cunha

C. C. D.

Santos

M. R. D.

Frederes

Cortez-Escalante

J. J.

Abreu

D. M. X. d.

Abreu

D. M. X.

(2019). Dados para a saúde: Impacto na melhoria da qualidade da informação sobre causas de óbito no brasil. Revista Brasileira de Epidemiologia, 22(3), e19005. supl 19003 https://doi.org/10.1590/1980-549720190005.supl.3

20.

Nascimento

C. F. d.

Duarte

Y. A. O.

Lebrão

M. L.

Chiavegatto Filho

A. D. P.

(2018). Individual and neighborhood factors associated with functional mobility and falls in elderly residents of são paulo, Brazil: A multilevel analysis. Journal of Aging and Health, 30(1), 118–139. https://doi.org/10.1177/0898264316669229

21.

National Academies of Sciences, Engineering, and Medicine. (2018). Future directions for the demography of aging: Proceedings of a workshop. Washington (DC): National Academies Press (US). 26 June 2018.

22.

Preston

S. H.

Heuveline

Guillot

(2001). Demography: Measuring and modeling population processes. Blackwell Publishers.

23.

Puterman

Weiss

Hives

B. A.

Gemmill

Karasek

Mendes

W. B.

Rehkopf

D. H.

(2020). Predicting mortality from 57 economic, behavioral, social, and psychological factors. Proceedings of the National Academy of Sciences of the United States of America, 117(28), 16273–16282. https://doi.org/10.1073/pnas.1918455117

24.

Silva Júnior

W. P.

Freire

F. H. M. d. A.

da Silva

D. N.

Gonzaga

M. R.

(2021). Evaluation of the death records quality in Brazil: Sociodemographic determinants of incomplete education information. Revista Latinoamericana de Población, 15(29), 211–233. https://doi.org/10.31406/relap2021.v15.i2.n29.8

25.

StataCorp . (2021). Stata multiple imputation reference manual Verion 17. StataCorp LLC.

26.

Teixeira

R. A.

Naghavi

Guimaraes

M. D. C.

Ishitani

L. H.

Franca

E. B.

(2019). Quality of cause-of-death data in Brazil: Garbage codes among registered deaths in 2000 and 2015. Revista Brasileira de Epidemiologia, 22(3), e19002. supl 19003 https://doi.org/10.1590/1980-549720190002.supl.3

27.

Turra

C. M.

Renteria

Guimaraes

(2016). The effect of changes in educational composition on adult female mortality in Brazil. Research on Aging, 38(3), 283–298. https://doi.org/10.1177/0164027515620245

28.

World Health Organization . (2022). Life tables by country: Brazil. Retrieved from https://apps.who.int/gho/data/view.main.60220?lang=en. Retrieved June 14, 2022, from World Health Organization https://apps.who.int/gho/data/view.main.60220?lang=en

29.

World Health Organisation (2017). Global strategy and action plan on ageing and health. World Health Organisation Geneva.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.67 MB