This study assessed data from a population-based health survey conducted by the Belo Horizonte Urban Health Observatory (OSUBH) of the Federal University of Minas Gerais (UFMG) between 2008 and 2009 in two – Barreiro and Oeste – of the nine health districts of the city. Belo Horizonte (2.513.451 inhabitants) is the capital of Minas Gerais State located in southeast Brazil and the main city of the Belo Horizonte Metropolitan Area (5.873.841 inhabitants) [22]. The estimated population of each district is approximately 250,000.
The study area was divided into strata according to the health vulnerability index (HVI), developed by the Belo Horizonte City Health Department. The HIV is a summary measure that estimates the inequalities in the epidemiological profile of different social groups within the census tracts. It includes the following components: sanitation, housing, education, income, and health [22].
In each HVI stratum, selection was performed using a three-stage sampling methodology: census tract, address (residence), and resident (one adult). In total, 150 tracts were selected. Within each census tract, a simple random sample of household addresses registered in the database of the Municipality of Belo Horizonte was taken. Next, one adult resident (18 years or older) was drawn using a random number table.
At the end of the sampling process, 5.436 households had been selected. After being informed about the objectives of the study, residents who were drawn were invited to participate and sign a consent form. In total, 4.048 adults were interviewed with a refusal rate of 25.0%. For this study, we selected only young adults between the ages of 18 and 29 years, representing 955 participants. All participants answered a face-to-face questionnaire administered by trained interviewers. The questionnaire was composed of six modules: household, sociodemographic factors, health, mobility, social determinants of health, lifestyle and behaviors.
Detailed information about the SBH survey methodology can be found in Camargos et al. [23] and Friche et al. [22, 24].
The study was approved by the Research Ethics Committee of the Federal University of Minas Gerais (UFMG), Brazil, under protocol numbers ETCI 253/006 and ETCI 017/07.
Variables
Response variable
Self-rated health was assessed by the question “In general, how would you rate your health?”, with five responses on a five-category scale: “very good”, “good”, “fair”, “poor”, or “very poor”. Responses were later categorized for analysis as fair/poor/very poor, and very good/good (reference category).
Variable of interest
The explanatory variable of interest was perceived urban violence, defined as a negative emotional reaction to crime, a social phenomenon that reduces social interaction and mutual trust among residents, causing a decline in the quality of life in the community or neighborhood [3,4,5].
The perceived urban violence score was constructed using the variables to assess the respondents’ perception of fear, danger, and insecurity of suffering some form of violence in the neighborhood. Participants were asked the following questions: What is the risk of being (1) personally threatened, robbed/mugged; (2) assaulted or threatened with aggression; (3) abducted (kidnapped); (4) hit by a stray bullet; (5) seriously injured or killed; and (6) a victim of police violence. Respondents rated the risk as very high, high, low, or very low. The perceived urban violence score was calculated using the principal components method and ranged from 1.21 to 4.85 (mean ± sd = 2.13 ± 0.69) for women and 1.21 to 4.65 (2.32 ± 0.74) for men.
Potential confounding variables were divided into the following categories:
-
A)
Time residing in the neighborhood in years
-
B)
Demographics: age (18–24 and 25–29 years); sex and marital status: married/living together or divorced/separated and single
-
C)
Schooling: able to read/primary school equivalency; 1st-4th grades not completed; 1st-4th grades completed; 5th-8th grades not completed; 5th-8th grades completed; high school not completed; high school graduate/technical school/attended university; university graduate; post-undergraduate studies.
-
D)
Socioeconomic: socioeconomic position score (SPS), detailed below;
-
E)
Lifestyle and behaviors: 1) smoking: current smoker and non-smoker/former smoker; 2) alcohol consumption: non-drinker, moderate consumption (1–2 times a week and less than five drinks per day), or excessive consumption (≥ three times a week or more than five drinks per day); 3) consumption of fruits and vegetables: defined as consumption of at least one portion 5 days a week for the past 12 months, yes/no; 4) physical activity: physical activity during leisure time ≥ 30 min/day and physical activity during leisure time < 30 min/day (1% of respondents who performed physical activity exercised for <30 min/day).
-
F)
Health condition: 1) whether he/she is satisfied with their own weight: yes/ no; 2) report of chronic disease: yes/no, detailed below;
The socioeconomic position score was constructed using 13 indicators: number of residents per bedroom; housing tenure (rented, owned, loaned, other); and presence or absence (yes/no) of the following items in the household: DVD player, videocassette recorder; cable TV subscription; microwave oven; automatic washing machine; house maid; semi-automatic washing machine; motorcycle; newspaper and/or magazine subscription; computer; internet access; motorcycle; car. The scores were calculated using the principal components method (range: 0–3.39) and were divided into quintiles. Higher scores indicate higher socioeconomic position [6].
The variable self-reported chronic disease included the following illnesses: hypertension, diabetes, asthma, bronchitis, depression, migraine, epilepsy, cancer, chronic digestive disease (ulcer, gastritis), and mental illness (schizophrenia, psychosis, anxiety disorder, bipolar disorder, obsessive-compulsive disorder, panic disorder, anorexia, bulimia). Participants were classified as having a chronic disease if they reported having at least one of the above conditions.
Statistical analyses
The perceived urban violence score was constructed from variables that assessed the respondents’ insecurity and perception of fear and danger of suffering some form of violence in the neighborhood using exploratory factor analysis. Factor analysis is a data reduction method, which is based on the assumption that highly correlated observed variables (indicators, items, or manifest variables) reflect the action of one or more (unobservable) latent variables or factors [25]. By estimating the latent factors, we were able to account for all or most of the variability generated by the observed variables using a few factors [25, 26]. Because variables were ordinal, we used the polychoric correlation matrix in factor analysis to calculate the scores [26].
Variables with a p value ≤ 0.2 on univariate analysis were included in the multivariate analysis. The sampling design was incorporated into the analysis using Stata ‘svy’ command. The strength of the associations was estimated by the odds ratios (OR) and 95% confidence intervals.
To construct the final model, we adopted the hierarchical approach, a sequential process in which the variables entry into the analysis in blocks following the theoretical model presented in Fig. 1 [13]. We use this model to better evaluate how perceived urban violence is associated with health self-assessment, hierarchically adjusting the confounding factors, understanding that they are moderators of this association [27]. Nested models were evaluated using the Wald test. The adjustment of the final model was assessed by the Hosmer Lemeshow test, considering the sampling design.
The evaluated models were: model 1: Urban Violence Perceived Score.
Model 2: model 1 plus years residing in the neighborhood.
Model 3: model 2 plus age, sex (for all participants only) and marital status.
Model 4: model 3 plus schooling and socioeconomic position score.
Model 5: model 4 plus alcohol consumption, smoking, healthy diet and physical activity.
Model 6: model 5 plus satisfied with his/her own weight and reporting of chronic diseases.
All analyses were performed using Stata software version 12.0 (StataCorp, College Station, TX, USA).