Suicidal behavior during the Covid-19 pandemic: Results from cross-sectional surveys of Brazilian adults from 2020 to 2022

2.1 Participants

The final sample consisted of responses to three cross-sectional surveys in the years 2020 (n = 4,803), 2021 (n = 1,477), and 2022 (n = 1,466), totaling 7,746 adults. The average age was 34.0 years (SD = 12.9; Min = 18 and Max = 70), with the majority being female (87.6%; n = 6,785), having a higher level of education (83.8%; n = 6,494), and identifying as white individuals (55.0%; n = 4,261). Participants were residents of approximately 1,300 Brazilian cities. The Northeast region was predominant (44.3%; n = 3,433), followed by the Southeast (34.3%; n = 2,657), South (12.3%; n = 950), Central-West (5.9%; n = 454), and North (3.3%; n = 252). Table 1 presents the distribution of all the collected sociodemographic data.

2.2 Instruments

The Fear of Covid-19 Scale (FCV-19S) (Ahorsu et al., 2020) is a unidimensional scale containing seven items using a Likert scale from 1 (strongly disagree) to 5 (strongly agree). The total score is obtained by summing the items, with totals ranging from 7 to 35 points. In the adaptation study for Brazilian Portuguese, satisfactory psychometric results were found (Faro et al., 2022), and the level of fear was classified into three categories: 7 to 19 points as mild fear, 20 to 26 points as moderate, and 27 points and above as severe. In this study, the Cronbach’s alpha (α) was 0.87, and McDonald’s omega (ω) was 0.88.

The Generalized Anxiety Disorder Scale - 7 (GAD-7) (Spitzer et al., 2006) consists of seven items, which are answered on a scale from 0 (not at all) to 3 (nearly every day) points. The total score ranges from 0 to 21, and the level of anxiety is classified as minimal (0-4 points), mild (5-9 points), moderate (10-14 points), and severe (15-21 points). The psychometric properties and validity evidence for this scale in Brazil, assessed in its Brazilian Portuguese version, are considered satisfactory (Moreno et al., 2016). In this study, satisfactory internal consistency indices were found (α = 0.89; ω = 0.89).

The Patient Health Questionnaire (PHQ-9) (Kroenke et al., 2001) consists of 9 items using a scale from 0 (not at all) to 3 (nearly every day). The total score ranges from a minimum of 0 to a maximum of 27. The stratification of scores proposes four categories: 0-4 as minimal, 5-9 as mild, 10-14 as moderate, and 15-27 as severe. The scale has demonstrated satisfactory psychometric characteristics in the Brazilian population in its Brazilian Portuguese version (Bergerot et al., 2014). In this study, the Cronbach’s alpha and McDonald’s omega were both 0.90.

All Likert scales were analyzed using the simple summation of responses, followed by classification in accordance with the indicated references. Additionally, we ensured that the scoring adhered to established guidelines to maintain the integrity of our analysis.

The sociodemographic questionnaire includes questions about the participant’s sex (female or male), age (in years), skin color (white, black, parda [mixed ethnic heritage]), education level (up to elementary school, high school, or higher school), and city of residence. Participants were asked if at any point during the pandemic they had thoughts, plans, and/or attempts to end their own life, with the following response options: (0) Never, (1) I had only a brief passing thought, (2) I had a plan to end my life at least once, but I didn’t attempt it, (3) I attempted to end my life, but didn’t really want to die, and (4) I attempted to end my life and actually wanted to die.

2.3 Procedures

Participant recruitment was conducted through convenience sampling, using social media platforms such as Facebook and Instagram. Posts on these platforms included links directing users to the research survey. Individuals who clicked on the link and voluntarily chose to participate became part of the study sample. This approach allowed the researchers to engage participants who were willing to take part in the research of their own accord. Only individuals aged 18 or older and those who agreed to the informed consent form, were eligible for the study. The consent was obtained electronically through the survey platform before the questionnaire was made available, which was presented at the beginning of the online questionnaire. Participation was voluntary, and anonymity and confidentiality were guaranteed in accordance with ethical standards. Data collection took place during three different periods of the pandemic in Brazil: June 2020, March 2021, and May 2022. The current findings have been reported in a preprint (Faro et al., 2024).

2.4 Data Analyses

The Jamovi software (jamovi project, 2025, version 2.2.5) was used for data analyses. A hierarchical multinomial logistic regression was conducted with suicidal behavior as the dependent variable. This variable was recoded into the following groups:(1) absence of suicidal behavior; (2) presence of suicidal ideation; (3) suicide planning and/or attempt. The explanatory variables included the year of data collection (2020, 2021, 2022), and the levels of fear of Covid-19 (7-9 mild, 20-26 moderate, 27-35 severe), anxiety (0-4 no symptoms, 5-9 mild, 10-14 moderate, 15-21 severe), and depression (0-9 no symptoms, 10-14 mild, 15-19 moderate, 20-27 severe), added in this order to the model. Model fit evaluation criteria included the initial and final -2 log likelihood, the significance criteria (p < 0.05) of the odds ratios (OR), the model’s predictive ability (correctly predicted cases, with values desired to be above 50%), the X² (expected to be significant at p < 0.05), the Pearson Goodness of Fit X² (expected value to be non-significant), and the Pseudo R-Square index (higher values indicate better fit). Odds ratios (OR) values below 1 were converted using the formula 1/OR for standardized result reporting.

2.5 Ethical consideration

All procedures performed in this study involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments. The study was approved by the National Commission of Research Ethics (CONEP) (Approval number: 3.955.180, Date: 01/05/2020). The consent was obtained electronically through the survey platform before the questionnaire was made available, which was presented at the beginning of the online questionnaire. Participation was voluntary, and anonymity and confidentiality were guaranteed in accordance with ethical standards.

3.1 Multinomial logistic regression

The final model ( Table 3) demonstrated an acceptable fit at all levels (p < 0.001). The model explained 20.7% of the variance (Nagelkerke’s R² = 0.207), and the correct predictive capacity had a total value of 34%. All the variables included remained as statistically significant in the final model.

Block 1: Year

The first block, which included the year of data collection, demonstrated an adequate fit indicator [χ² = 56.4 (4); p < 0.001]. In 2022, there was an additive impact, increasing the odds of being in the group with suicidal ideation by 16.8% (OR = 1.168; p = 0.038) and nearly doubling the odds of suicide planning and/or attempt (OR = 1.953; p < 0.001) compared to the year 2020 (OR = 1.221; p < 0.039). There was no statistically significant difference in suicidal ideation between 2021 and 2020 (p ≤ 0.05).

Block 2: Fear of Covid-19

The second block also showed an adequate fit indicator for the FCV-19S scale [χ² = 95.4 (4); p <0.001]. Individuals with severe fear (OR = 1.876; p < 0.001) or moderate fear (OR = 1.402; p < 0.001) had a higher chance of being in the group with suicidal ideation compared to those with mild fear. In this block, the year variable remained significant for 2022, and there was an increase in the odds for both suicidal ideation and suicide planning and/or attempts.

Block 3: Anxiety

In this block, there were also appropriate fit indicators for the GAD-7 scale [χ² = 987.5 (6); p < 0.001]. However, in this block, the direction of the FCV-19S had reversed compared to the second block. The severe fear group had a lower chance of being in the groups with suicidal ideation and suicide planning and/or attempts when compared to mild fear by 31.7% (1/OR = 1.317; p = 0.002) and 88.6% (1/OR = 1.886; p < 0.001), respectively. The year variable was only significant for regarding suicide planning and/or attempts for 2022, showing a reduction in odds compared to the second block by almost two times to 56.2%, when comparing the 2022 to 2020 (OR = 1.562; p < 0.001).

The anxiety variable increased the odds of suicidal ideation from three and a half times at the mild anxiety level (OR = 3.486; p < 0.001) to over 13 times at the severe anxiety level (OR = 13.515; p < 0.001). For suicide planning and/or attempts, this range was from four and a half times at the mild anxiety level (OR = 4.485; p < 0.001) to over 41 times at the severe anxiety level (OR = 41.533; p < 0.001), compared to the group without anxiety.

Block 4: Depression

The fit indicators were significant for the PHQ-9 scale [χ² = 1015.5 (6); p < 0.001]. The year 2022 had an increase of 64.4% for suicide planning and/or attempts (OR = 1.644; p < 0.001) compared to the year 2020. The fear variable remained significant in the following models. The more severe the fear, the lower the chance of being among those with suicidal ideation, with values of 43.6% (1/OR = 1.436; p < 0.001) for severe fear compared to mild fear. The same pattern was observed for suicide planning and suicidal ideation, with the values of 1/OR increasing from 1.886 to over two times (1/OR = 2.247; p < 0.001) for severe fear compared to mild fear. Moderate fear remained stable in terms of values from the third to the fourth block.

In the fourth block, the adjusted odds of the anxiety variable decreased compared to the third block, specifically for suicidal ideation. The results ranged from two times (OR = 2.183; p < 0.001) to approximately three times (OR = 2.757; p < 0.001). For suicide planning and/or attempts, the behavior of the variable was the same, with results between two times (OR = 2.253; p < 0.001) and over three times (OR = 3.577; p < 0.001).

With the addition of the depression variable in this block and the adjustment of odds for suicidal ideation, the probabilities ranged from almost three times (OR = 2.831; p < 0.001) to over ten times (OR = 10.660; p < 0.001) in comparison to the group without depression. The same pattern occurred when considering the suicide planning and/or attempts group, with the odds increasing from three times (OR = 3.021; p < 0.001) to over 40 times (OR = 40.363; p < 0.001) when compared to the group without depression.

In summary, the cumulative impact of 2022 compared to 2020 was significant for suicidal behaviors in the first block. In the second block, the exposure factor of fear of Covid-19 was noteworthy, as the group of people with severe fear exhibited lower chances of displaying suicidal behaviors than those with mild fear. In the third block, the fear of Covid-19 reversed the direction of the relationship with the inclusion of anxiety and became a protective factor for the suicide behaviors. Moreover, the presence of anxiety in this block increased the chances of displaying suicide planning and/or attempts from over four times to 41 times compared to the group without anxiety. Finally, in the fourth block, the odds of anxiety were adjusted compared to the previous block. Considering the suicide planning and/or attempts group, the presence of depression increased the odds from three times to over 40 times compared to the group without depression.

Evaluations of hierarchical multinomial logistic analyses and various indicators (VIF, tolerance, residuals, and the discrepancy of OR in relation to their confidence intervals) did not indicate unsatisfactory quality of the findings. Figure 1 represents the last block of analysis, containing the adjusted ORs and their respective confidence intervals.

Figure 1. Odds ratio and confidence intervals of the fourth block (last model) - Multinomial logistic regression for suicidal behavior (Brazil, n = 7,746).

Notes. Graphical representations of OR and 95% CI in block 4 by variable. Panel 1A: Year of Pandemic. Panel 1B: Fear of Covid-19. Panel 1C: Anxiety Symptomatology. Panel 1D: Depression Symptomatology. “a” indicates statistically non-significant associations; all other values are statistically significant.

4.1 Predictive variables of the suicidal behavior

Suicidal ideation increased by 16.8% between 2020 and 2022, and suicide planning/attempt nearly doubled. These findings align with research warning of delayed mental health consequences of the pandemic (Bigoni et al., 2022; Dubé et al., 2021; Efstathiou et al., 2022; Gramacho and Turgeon, 2021). Initially, fear of Covid-19 was associated with higher suicidal ideation. However, once anxiety was included in the regression model, fear became a protective factor. This suppressor effect suggests that while persistent fear may initially be harmful, it may promote adaptive behaviors when controlling for anxiety (Harper et al., 2021; Lin et al., 2021; Smith et al., 1992). The fear variable maintained its relevance across years, although with lower explanatory power.

Anxiety was strongly associated with suicidal ideation and behavior, increasing odds by up to tenfold for ideation and over 40 times for planning/attempting suicide. Given that nearly one-third of adults experienced anxiety during the pandemic (Delpino et al., 2022), this reinforces the urgent need for targeted mental health support.

Depression emerged as the most significant predictor. Its inclusion in the model reduced the impact of year, fear, and anxiety variables. Suicidal ideation odds ranged from three to ten times higher depending on symptom severity; for planning/attempting suicide, the risk increased by over 40 times. These findings highlight the need to prioritize depression in suicide prevention strategies (Ali et al., 2022; Daniali et al., 2023; García-Iglesias et al., 2022; Mann et al., 2005; Nock et al., 2008b).

This study supports global findings on the impact of the pandemic on mental health and suicidal behavior (Mann et al., 2005; Robinson et al., 2022; Vandamme et al., 2023). The model’s total predictive capacity reached 34% when combining all variables. Notably, fear showed dual roles—risk factor when isolated, protective when anxiety was included—requiring nuanced interpretation.

Limitations include the use of a convenience, online sample and the inability to control for variables like age, gender, and socioeconomic status due to sample imbalance. However, the large sample (from over 1,300 cities) partially mitigates concerns about representativeness.

Although Covid-19 is no longer considered a public health emergency (World Health Organization, 2023d), its mental health repercussions persist (Penninx et al., 2022; World Health Organization, 2023d). Suicide prevention remains essential. Individuals with depressive and anxiety symptoms—key predictors in this study—should receive prioritized support, in line with WHO guidelines (World Health Organization, 2022).

Informed consent statement

All participants were informed about the objectives of the study and provided their consent prior to participation. Consent was obtained electronically through the survey platform before the questionnaire was made available. Participation was voluntary, and anonymity and confidentiality were guaranteed in accordance with ethical standards.