Study design, location and participants
We carried out a cross-sectional observational study10 involving adults living in urban cities in Brazil. The study population included individuals who declared themselves to be Brazilians or foreign migrants, who were proficient in the language spoken in Brazil (Brazilian Portuguese), and who had lived in the country for at least six months. These participants were aged 18 or older, willing to collaborate in the research and had access to the internet.
The sample was formed using the snowball technique11, a nonprobabilistic approach in which participants were invited to respond to the invitation via the websites of the research partner institutions, emails, WhatsApp and social networks (Facebook, Instagram, and Twitter). In addition, participants were encouraged to invite individuals from their own network of contacts, thus facilitating initial contact for conducting the research, opinion studies and surveys11.
$$n={\left(\frac{{{Z}_{(1-\alpha )} .DP. }_{ }}{d}\right)}^{2}$$
where
"n" is the needed sample size; "Z" is the critical value of the standard normal distribution, which corresponds to the confidence level (1 - α); "SD" represents the standard deviation of the population, a measure of the variability of the data in the population; and "d" is the desired margin of error, the acceptable amount of variation in the research results. The minimum sample size was 1924, considering a 10% loss rate.
Measuring instruments and measures:
The instrument was developed and validated by researchers from the National School of Public Health at the New University of Lisbon (ENSP/UNL)12, adapted for Brazil, and submitted to a validation process using the Delphi technique, which proved to be sufficiently valid and reliable for its application in the Brazilian context. The Delphi technique used in content validation processes involves building consensus among a group of experts; this process is carried out in successive rounds, with the aim of evaluating a given problem or intervention proposal. Originally, the number of rounds used in the Delphi technique was defined by building a consensus based on overcoming/eliminating disagreements13.
Data were collected between August 2020 and January 2023 using an online questionnaire. The recruitment strategy used was the snowball technique and social media advertising on digital platforms (Twitter, Facebook, WhatsApp and the University's own institutional websites). A virtual environment was created so that people could have a better understanding of the Thermometer Project and its intentions. The questionnaire was accessed via a link made available on social networks, which is hosted in the RED-CAP environment.
Statistical analysis
After checking that the database was consistent, the data were analyzed using descriptive statistics and calculated as frequencies (n) and percentages (%) to characterize the profiles of the people who participated in the study and answered the questionnaire. As this was a survey with a predominance of categorical variables, bivariate logistic regression was applied, considering the dependent and independent variables, which were dichotomized.
Logistic regression was performed to verify the associations of the dependent variables, such as self-perceived risk of developing serious illness or complications due to COVID-19 with age, sex, occupation, education, monthly wage income, use of a protective mask, chronic noncommunicable diseases, religion and individual and collective (population) risk of contracting COVID-19 infection; these variables were identified as independent variables, with alternative answers (yes or no).
Modeling was conducted using the backward selection method, the Akaike information criterion (AIC) was used to choose the best model, and the model with the lowest AIC value was chosen 14. For the final model, odds ratios (ORs) were also calculated, with 95% confidence intervals (95% CIs).
After choosing the final model (based on the criterion of the lowest AIC value), the Hosmer–Lemeshow, likelihood ratio, Cox–Snell, Nagelkerke and McFadden tests were carried out to validate the model, and its accuracy was also checked using the area under the curve through the receiver operating characteristic (ROC) curve and their respective 95% CI values15. The data were analyzed using R 3.6.0 software.
Ethical aspects:
The study was approved by the Research Ethics Committee (CEP) of the Sergio Arouca National School of Public Health (ENSP-FIOCRUZ) (CAAE: 32210320.1.0000.5240) and the Ribeirão Preto School of Nursing - EERP-USP (CAEE: 32210320.1.3001.5393), the latter as a result of the research partnership.
All participants who agreed to participate had to electronically sign the informed consent form, which described the risks (minimum participation in the research) and the possibility of withdrawing at any time from the research; additionally, the data were anonymized and stored in a secure environment on the RED-CAP platform.