A total of 1,088 suicides occurred in the Brazilian Federal District between 2005 and 2014. In 308 cases, the toxicological examination was not performed, therefore excluded from the present study. Thus, 780 cases comprised the sample (344 with positive results for PASs use before dying and 436 with negative results), as shown in Diagram S1 (Supplementary Material).

Figure 1 shows the Brazilian Federal District population and suicide rate of those who used PASs normalized by the minimum value of each category as a function of the year and the best linear fits and respective equations. The suicide rate of those who used PASs was 10 times higher than the population growth rate (p = 0.0087, in slope comparison using GLIMMIX procedure).

The descriptive analysis of the data (Table 1) showed that the difference in the parameters analyzed within each of these variables – Day of the week, Period, Place, Age group, Sex, Level of education, Previous SA?, Recent behavior change?, What has changed?, Skin color and BMI – were significantly relevant (p < 0.05, χ² test for specified proportions). On the other hand, Day of the month, Season, and Marital status were non-significant.

Analyzing the relative risk (last column in Table 1), we can establish the probability of suicide within each investigated variable, dividing the value of one parameter by another. For example, the probability of a person using PASs and dying by suicide on the weekend was 1.6-fold greater than those who die by suicide on the weekday; at home was 2.8-fold greater than in any other place, and being a man was 5.6-fold greater than being a woman.

Table 2 shows the descriptive analysis for other variables. Considering employment, the most vulnerable groups were tradespeople and liberal professionals (29.89%), and unemployed individuals (28.80%). The two primary reasons for the suicide were drug addiction (34.75%) and crime of passion (28.75%). It is noteworthy that family members may have been informed of more than one reason, and all of them were considered; therefore, the number of times the term appeared was counted. Hanging was the method of suicide chosen by more than half of the individuals (62.79%), followed by using a firearm (17.44%). Among those who used a firearm, the head region was the most common anatomical site chosen (86.15%).

In the sample studied, 71.68% used alcohol, 56.19% used other drugs, and 22.09% used both. Most individuals who used PASs (except alcohol) used only one type of them (83.05%), and the most common was cocaine (56.25%). Among those who used alcohol before dying by suicide, the mean dose was 1.49 g of alcohol per liter of blood, whereas nearly 35% had a blood alcohol concentration between 1.5 g/l and 2.99 g/l.

Table 3 shows Pearson’s correlation coefficients for the linear correlation between pairs of variables. The most relevant correlations are presented below and discussed in the next section. Younger people were more likely to die by suicide on the weekend, whereas older people were more likely on a weekday. Over time, individuals died by suicide more during the day, average age increased (see inset of Table 3), and more individuals had recently demonstrated a behavior change. Married people were more likely to die by suicide at night, whereas unmarried people during the day. With the increase in age, people who died by suicide were married, had white skin color, had high blood alcohol concentration, and did not use other drugs. While men chose to hang as the primary method of suicide, women preferred jumping from a height or poison. Individuals who died by suicide in their own homes preferred the hanging method. Individuals with previous SA were less likely to use PASs before dying by suicide. Those who recently changed their behavior before dying by suicide were more likely to have a higher blood alcohol concentration. Individuals who jumped from a height were more likely to be under the influence of a PAS (except alcohol).

2-tailed test of significance used; acorrelation significant at 0.05 level1. Day of the month; 2. Day of the week; 3. Month; 4. Year; 5. Period; 6. Age; 7. Sex; 8. Level of education; 9. Marital status; 10. Suicide in own home; 11. Previous attempt?; 12. Behavior changed recently?; 13. Skin color; 14. Body Mass Index; 15. Hanging; 16. Firearm; 17. Jumping from height; 18. Poison; 19. Blood Alcohol Content (BAC); 20. Drug use

The odds ratios show the risk of dying by suicide among individuals with and without PASs according to the variables shown in Table 4. Statistically significant differences were found for seven of the 17 variables studied. Individuals who used PASs before dying by suicide were more likely to have a fatal SA on weekends. Individuals who did not use PASs (except alcohol) were more likely to die by suicide during the day, whereas those who used PASs were more likely to die by suicide at night. Among the age groups, the elderly and adolescents were more likely not to have used PASs (except alcohol), and women were more likely not to have used any PASs before dying by suicide. Among occupations, construction workers were more likely to use PASs before committing suicide. Individuals whose reason for committing suicide was drug addiction were more likely to use drugs before committing suicide. Individuals with a mental disorder were less likely to use PASs (except alcohol). Individuals who chose poison as the suicide method were less likely to be under the influence of PASs before dying by suicide.

The PCA yielded five uncorrelated components (Table S2, in Supplementary Material), enabling a global, synthetic data view. These components had eigenvalues of 1.61, 1.43, 1.29, 1.28, and 1.17 and explained 9.47%, 8.40%, 7.60%, 7.56%, and 6.90% of the total variance in the initial data, respectively. The components were characterized as follows:

• Component 1 – suicide method: men, young people, individuals with a lower education level, and individuals with black/pardo skin color chose hanging; women, older people, individuals with a higher education level, and white-skinned individuals chose other suicide methods;
• Component 2 – drug group: individuals with higher levels of schooling and younger people preferred using PASs (except alcohol); older people and those with lower levels of education chose alcohol;
• Component 3 – suicidal thinking (characteristics of individuals who have non-fatal SA); preferred suicide at night, away from home, and at the beginning of the month; low loadings for alcohol use and other drug use and opposite directions to the previous attempt suggest that recidivism and suicide do not depend on drug use, which confirms the belief in suicide;
• Component 4 – behavioral traits: individuals who had a recent behavioral change due to delusion, depression, etc., tended to commit suicide outside the home, during the day, and in the first months of the year; loading associated with recent behavior change was in the opposite direction of marital status, which means that the unexpected change in behavior predominantly occurred among single people;
• Component 5 – premeditation: over the years, suicide occurred more during the day and at the beginning of the month, independent of the dosage of PASs used (except alcohol); these tendencies indicate a higher degree of forethought, which may explain the higher increase in suicides compared to the rise in the population.

Figure S1 (Three-dimensional PCA graph considering the first three principal components, in Supplementary Material) presents the three-dimensional PCA graph considering the first three main components, reinforcing the discussion established in the previous paragraph (individual component). Hanging (Hang.), sex (Gen.), and skin color (Skin) are vectors with approximately the same dimension positioned in the right hemisphere of the figure and are therefore correlated. Moreover, these vectors are in the opposite direction (negatively correlated) to the level of education (Edu) and age (Age), justifying the discussion related to Component 1 (suicide method). Drug use (Drug), level of education (Edu), and age (Age) are on the left side of the graph, validating Component 2 (drug group). Previous attempts (Prev. Att.?) and period (Period) are directed to the central lower part of the graph and are positively correlated. Moreover, these vectors are opposite to place (Resid.) and day of the month (day-m), corroborating what has been discussed for suicidal thinking (Component 3).

Figure 2 presents a 4D PCA graph. The three axes correspond to the principal components with a higher coefficient for the variable year, and a fourth dimension (color) corresponds to the year.

The graph of the scores shows that the data are very dispersed due to randomness and the peculiarities of each suicide event. However, using the degradation, although considerable dispersion was maintained over the historical series, the spatial position of the data changed, with lower values, on average, for Component 3 and higher values for Components 2 and 5. This indicates that the suicide characteristics among PASs users changed over time, albeit slowly. Correlating this position change of the dispersion of the points with the factors attributed to the components, one may infer that the increase in suicide associated with the alcohol use (PC2) was accompanied by increases in the forethought of suicide (PC5) and the number of individuals who had previously SA (PC3). There is no way to establish a relationship among these three trends since, algebraically, the components are orthogonal (independent of one another). Therefore, this change in characteristics may be related to different groups that shaped this change in the individuals who died by suicide and had used PASs before. Furthermore, the low accumulated variance in the first five components (39.93%) supports the notion that suicide is a dependent, multivariate event that the analysis of a few effects cannot fully explain.