This is a cross-sectional epidemiological study conducted in five municipalities in the state of Bahia in 2011 and 2012, with health workers from medium complexity/primary care services.
The study population was defined by a previous survey of the workers at the Municipal Health Secretariats of the five municipalities, where the study took place. The number and type of health services available, number of employees, and their occupations, as well as the geographic area in which each service was located, were delimited. Based on the list of all workers and estimates of health outcomes, the sample size of the original study was defined (based on the outcome that resulted in the largest sample size). After this, a representative sample of health workers was obtained by random and stratified sampling, and proportionally organized per geographical areas (coverage areas of the Family Health Care units), complexity level in the health services network (primary or medium complexity care), and professional category. The sampling population comprised all previously identified health workers of each municipality. Then, the randomly selected workers were contacted at their respective work environments and invited to be part of the study.
Inclusion criteria were the following: workers who were in effective professional practice, who accepted to participate in the study, and who had been working for at least 6 months at the health unit. We try to find the raffled workers and conduct the interview by three attempts. Those who refused to participate in the study, those who were away from their work in the period of data collection, or those who were not found in the three attempts were replaced with respect to the geographical area, level of complexity of the service, group occupation, and sex.
Considering that the sample had not been calculated for the purpose of this study, in order to verify whether the multicenter study had the power to evaluate the association between work psychosocial aspects and dissatisfaction, the sample was recalculated using job dissatisfaction as the outcome of interest. Such calculation was performed using the software OpenEpi, version 3.03a. Based on the total population of health workers in the five municipalities (N = 4278), the prevalence of the event of interest was at 44.9% (Fadel, Carvalho, Arcieri, Saliba, & Garbin, 2008), confidence interval at 95.0%, absolute error at 2%, and estimate rate for losses/refusals at 20%. Finally, the estimated number of the sample was 1834 workers. A total of 3084 workers were investigated, thus providing the study with sufficient power for the intended analysis.
Data collection has used a structured questionnaire, based on the literature review, previously tested in a pilot study. In order to standardize the methodological procedures adopted at each location, a manual of procedures was prepared and workshops were held to train and prepare the interviewers.
Job dissatisfaction was assessed by the following question of the Job Content Questionnaire (JCQ): Are you satisfied with your job? Response options included a Likert scale with score from 1 to 4 with the following response options: 1, I am not satisfied at all; 2, I am not satisfied; 3, I am satisfied; and 4, I am very satisfied. For the purposes of analysis, the variable was dichotomized in satisfied (answers 3 and 4) and dissatisfied (answers 1 and 2).
This type of scale to evaluate dissatisfaction is feasible because satisfaction and dissatisfaction represent opposite meanings of the same phenomenon. Satisfaction with work is caused by multiple factors, and this multicausality of the events must be considered. However, studies have demonstrated the feasibility of measuring job satisfaction with a single question. These approaches have shown a positive correlation of measurements made from multidimensional (multiquestions) questionnaires. In addition, one question is more sensitive to capture variations in job dissatisfaction (Ommen et al., 2009; Wanous, Reichers, & Hudy, 1997).
The two main exposure variables were considered as follows: the “demand-control model” measured by the JCQ and the “effort-reward imbalance” obtained from the ERI scale.
The DCM variable was measured by combining job aspects, psychological demand, and control, and the median was considered as a cutoff point for defining categories as high and low. Based on these levels, four specific work situations have been established and considered various risks to worker’s health as follows : low strain (high control/low demand), active job (high control/high demand), passive job (low control/low demand), and high strain (low control/high demand) (Araújo & Karasek, 2008). DCM score was calculated from the ratio between demand and control (D/C) dimensions in order to make DCM comparable to ERI and, thus, combine these two models. By using the mean as a cutoff point, workers were classified as exposed and non-exposed, based on the dichotomous categorization of the scores obtained (values ≤ mean = not exposed, values > mean = exposed) (Griep et al., 2011).
The ERI model was developed from a self-administered questionnaire with Likert-scale responses (1, “I strongly disagree”; 2, “I disagree”; 3, “I agree”; 4, “I strongly agree”). The mean was also used as a cutoff point for defining the proposed scales in this model: effort (high/low), reward (high/low), and overcommitment (present/absent). The effort-reward imbalance indicator was obtained by the following formula: ERI = e/(r × c), where e refers to the sum of the effort items, r is the sum of the reward items, and c is a correction factor. The results were categorized as “balance” (values ≤ 1) and “imbalance” (values > 1)7. Here, workers were classified as exposed and not exposed using 1 as the baseline value (values ≤ 1 = not exposed, values > 1 = exposed) (Griep et al., 2011).
The analysis considered the association between job dissatisfaction and the partial, full, and partial (combined) occupational stress models. Thus, the following five models were evaluated: partial DCM (psychological demand and control over one’s own work), full DCM (demand-control and “social support” in the work), partial ERI (effort and reward), full ERI (including the dimension “overcommitment”), and the combined models (partial DCM combined with partial ERI).
The following covariables, known to be related to job dissatisfaction, were also included in the analyses: socio-demographic characteristics (sex, age, skin color, marital status, children, and income) and professional information (professional practicing time, employment bond, work shift, compatibility with the job description, weekly working hours, having another job, and having labor rights, like vacation and thirteenth salary) based on the literature (Carrillo-Garcia et al., 2013; Lapischies, Jardim, & Kantorski, 2014; Ribeiro et al., 2014; Tambasco, da Silva, Pinheiro, & Gutierrez, 2017).
For the statistical analysis of the data, the studied population was initially described, followed by bivariate, stratified, and multivariate analyses. The bivariate analysis calculated the prevalence ratios (PR) and 95% confidence intervals (CI). Also, p values were calculated using Pearson’s chi-square test to evaluate the statistical significance of the associations. The Statistical Program for Social Sciences 24.0 (SPSS 24.0) and OpenEpi 3.0 were used in this phase.
In order to investigate confounding factors, variations equal to or greater than 20% between crude and adjusted associations per each covariable of interest, as well as theoretical evidences, were considered. As none of the covariables investigated has varied greater than 20%, to select the confounding variables, we applied the theoretical knowledge based on literature review. These variables were added to adjust the final model.
The logistic regression was applied in the multivariate analysis, which generates odds ratio (OR) as a measure of association. Also, the robust Poisson regression method was used to properly estimate the prevalence ratios (PRs) and their respective 95% confidence intervals (Coutinho, Scazufca, & Menezes, 2008; Francisco et al., 2008). The software STATA 12.0 was used in this step.
The final model’s quality assurance and fit model were assessed by the goodness-of-fit test (Hosmer and Lemeshow), ROC curve analysis, and pattern of influential observations.
The regulatory ethical resolution 466/2012 of the Brazilian National Health Council was complied with at all stages of the research, and the study was approved by the Ethical Institutional Review Board (IRB) under the protocol number 081/2009 (CAE 0086.0.059.000. -09).