General aspects
An analytical study based on estimates of global burden of disease for Brazil made by GBD 2017 was carried out, coordinated by the Institute for Health Metrics and Evaluation (IHME) in partnership with the Ministry of Health of Brazil [11,12,13,14,15]. In the analysis of mortality, information from the Mortality Information System of the Brazilian Ministry of Health was used, with adjustment for underreporting of deaths and declaration of undefined/nonspecific causes, called garbage codes [13,14,15].
The standardized methodology of analysis adopted by the GBD makes it possible to compare countries, regions, and subnational data, also enabling analyzing trends, provided that the time series data are adjusted and comparable [13,14,15].
Mortality estimate due to diseases that have physical inactivity as risk factor
In the present study, physical inactivity was considered a risk factor for breast cancer, colorectal cancer, ischemic heart disease, diabetes mellitus, and stroke [5]. Other causes of mortality that have physical inactivity as one of the risk factors were not determined in the GBD study methodology. Information on the records and how each of these diseases was collected, estimated, and adjusted are found elsewhere in the literature [5, 13,14,15].
The comparative risk assessment conceptual framework used in GBD study established a causal web of hierarchically organized risks or causes that contribute to health outcomes, which allows for quantification of risks or causes at any level in the framework [5]. In GBD 2017, as in previous iterations of the GBD study, we evaluated a set of behavioral, environmental and occupational, and metabolic risks, where risk-outcome pairs were included based on evidence rules [5]. These risks were organized in four hierarchical levels, where level 1 represents the overarching categories (behavioral, environmental and occupational, and metabolic) nested within level 1 risks; level 2 contains both single risks and risk clusters; level 3 contains the disaggregated single risks from within level 2 risk clusters; and level 4 details risks with the most granular disaggregation. Physical inactivity is in the category of behavioral risk factor and stands out at hierarchical level 2 [5]. In addition, physical inactivity does not have any level of disaggregation.
Physical inactivity prevalence estimate
Surveys of the general adult population performed using random sampling procedures that assessed self-reported physical activity in all life domains (leisure/recreation, work, household, and transport) were included. Studies that evaluated only one of the physical activity domains were not included [5].
In general, for GBD estimates, data are primarily derived from two questionnaires, the Global Physical Activity Questionnaire (GPAQ) and the International Physical Activity Questionnaire (IPAQ). However, other studies that evaluated physical activity intensity, frequency, and duration in all domains were included [5].
In the case of Brazil, surveys considered included the National Health Survey 2013, Surveillance System for Risk Factors and Protection for Chronic Diseases by Telephone Inquiry (VIGITEL), Household Survey on Risk Behaviors and Referred Morbidity of Noncommunicable Diseases 2002–2005, World Health Survey 2003, and the International Study on the Prevalence of Physical Activity. Further details can be found at http://ghdx.healthdata.org/gbd-2016/data-input-sources?locations=135&components=6&risks=125.
To standardize all physical inactivity estimates in Brazil, data from the population aged 25 years or more were considered. Physical activity was considered only when accumulated for at least 10 consecutive minutes or more. Physical activity frequency, duration, and intensity were used to calculate the total metabolic equivalent (MET) minutes in the week. Physical activity level was categorized by total MET-minutes per week. The lower limit (600 MET-min/week) is the recommended minimum amount of physical activity to get any health benefit [5, 16]. More details on these models can be found in the literature [5].
Analysis strategy
The contribution of physical inactivity to mortality from all causes investigated in this study was estimated using a conceptual framework of the comparative risk factor assessment [5]. For this, the CODEm simulation model, which is an analytical tool that tests various statistical models of causes of death and creates a combination of models that provide the best predictive performance, was used to estimate indicators by sex, age, state, year, and cause. The DisMod-MR 2.1 software (World Health Organization©, Geneva, Switzerland), a meta-regression tool, was used to simultaneously derive estimates of prevalence, disability, and mortality due to physical inactivity [5]. Spatiotemporal Gaussian process regression (ST-GPR) has been used for risk factors where the data density is sufficient to estimate a very flexible time trend. The approach is a stochastic modelling technique that is designed to detect signals amidst noisy data. It also serves as a powerful tool for interpolating non-linear trends. Unlike classical linear models that assume that the trend underlying data follows a definitive functional form, GPR assumes that the specific trend of interest follows a Gaussian process. Details of all models can be found in literature [5].
In this study, the counterfactual level of risk exposure used is the risk exposure that is both theoretically possible and minimizes risk in the exposed population that consequently captures the maximum population attributable burden [5]. For each risk evaluated in GBD study, included low physical activity, has been used the best available epidemiological evidence from published and unpublished relative risks by level of exposure and the lowest observed level of exposure from cohorts used to select a single level of risk exposure combined to establish the theoretical minimum-risk exposure level (TMREL) [5]. The TMREL for physical activity is 3000–4500 MET-min per week, which was calculated as the exposure at which minimal deaths across outcomes occurred [5].
We also estimated the population attributable fraction (PAF), which represents the proportion of risk that would be reduced in a given year if the exposure to a risk factor in the past was reduced to an ideal exposure scenario [5]. We used a recently published dose-response meta-analysis of prospective cohort studies to estimate the effect size of the change in physical activity level on breast cancer, colon cancer, diabetes, ischemic heart disease, and ischemic stroke [3].
Summary exposure value (SEV) for physical inactivity was used in this study. SEV is the relative risk-weighted prevalence of exposure, a univariate measure of risk-weighted exposure, taking the value zero when no excess risk for a population exists and the value one when the population is at the highest level of risk. We report SEVs on a scale from 0 to 100% where a decline in SEV indicates reduced exposure to physical inactivity and an increase in SEV indicates increased exposure. More details on SEV are also available elsewhere [5, 13,14,15].
The absolute numbers of deaths and mortality rates (per 100,000 inhabitants, crude, and age-standardized) were also used as metrics [5, 13,14,15]. In addition, the 95% uncertainty intervals were estimated (95% U.I.).
Analyses were performed for the Brazilian population and also stratified by sex and age (25–49 years, 50–69 years, 70+ years). Analyses were presented for each of the Brazilian states plus the Federal District in the years of 1990 and 2017. The GBD study created the Socioeconomic Development Index (SDI) [13,14,15] for all evaluated locations, by calculating per capita income, formal education at 15 years of age, and fertility rate. This index was used to compare the metrics among Brazilian states. For this, the Spearman correlation coefficient was applied.