Frequently Asked Questions

Baseline suicidal ideation on PHQ-9 item 9 was the strongest predictor of suicide attempts over 24 weeks in this sample of distressed Brazilian health care professionals. In the multivariable Cox model, compared with no ideation, the hazard ratio was 39.58 (95% CI, 14.03–111.64; P<.001) for ideation occurring nearly every day, 7.67 (95% CI, 2.32–25.35; P<.001) for more than half the days, and 3.17 (95% CI, 1.42–7.07; P=.005) for several days, showing a clear dose-response relationship.

Yes. Sleep problems remained independently associated with suicide attempts even after adjustment for suicidal ideation and other baseline factors. In the multivariable model, hazard ratios were 17.39 (95% CI, 2.05–147.46; P=.009) for severe sleep disturbance, 12.10 (95% CI, 1.55–94.24; P=.017) for moderate sleep disturbance, and 9.96 (95% CI, 1.23–80.43; P=.031) for mild sleep disturbance.

The highest-risk subgroup was participants with nearly every day suicidal ideation plus severe sleep problems, with a 24-week suicide attempt probability of 57.1%. By comparison, the 24-week attempt probabilities were 42.3% for nearly every day ideation alone, 12.3% for severe sleep problems alone, and 1.2% for neither factor (log-rank P<.001).

Yes, but the pattern depended on how interaction was tested. The multiplicative interaction between sleep problems and suicidal ideation was not significant (χ2=11.34, P=.253), but additive interaction analysis suggested substantial synergy: the relative excess risk due to interaction was 11.51, the attributable proportion was 0.40, and the synergy index was 1.66.

Among 2,708 participants in this analysis, event rates were 1.0% for neither severe sleep nor nearly every day ideation, 22.2% for ideation only, 3.0% for severe sleep only, and 33.3% for both. The authors interpreted this as indicating that 40% of the combined effect was due to synergy.

Nearly every day suicidal ideation was the only baseline factor that remained independently associated with psychiatric hospitalization after multivariable adjustment. Its hazard ratio was 8.11 (95% CI, 3.10–21.18; P<.001), while depression, anxiety, sleep problems, and life satisfaction lost significance in the adjusted model.

In univariate analyses, nearly every day ideation (HR=9.88, 95% CI, 5.01–19.48; P<.001), severe depression (HR=3.02, 95% CI, 1.66–5.49; P<.001), severe anxiety (HR=2.09, 95% CI, 1.24–3.53; P=.006), and severe sleep problems (HR=2.55, 95% CI, 1.25–5.17; P=.009) were associated with hospitalization, but only nearly every day ideation persisted after adjustment.

PHQ-9 item 9 had sensitivity of 71.7%, specificity of 77.8%, positive predictive value of 12.9%, and negative predictive value of 98.3% for predicting suicide attempts in the intention-to-treat sample. The authors concluded that, in this setting, the item was more useful for ruling out near-term risk than for ruling it in.

Among the 46 participants who attempted suicide in the intention-to-treat sample, baseline item 9 responses were distributed as 28.3% “not at all,” 30.4% “several days,” 8.7% “more than half the days,” and 32.6% “nearly every day.”

No. Although PHQ-9 item 9 showed a strong association with later suicide attempts, the study found a positive predictive value of only 12.9%, meaning most participants with a positive screen did not attempt suicide during follow-up. The authors stated that PHQ-9 item 9 should not be used as a standalone decision tool and should be supplemented with a more comprehensive suicide risk assessment incorporating multiple risk domains.

Yes. In the multivariable model, male sex was associated with higher suicide attempt risk with a hazard ratio of 2.08 (95% CI, 1.01–4.26; P=.046). This association was also present in inverse probability weighted analyses, with a hazard ratio of 2.71 (95% CI, 1.27–5.76; P=.010).

The authors noted that this finding differs from usual epidemiologic patterns for suicide attempts and advised caution, stating that it may reflect selection effects among help-seeking men or pandemic-specific occupational stressors and warrants replication before influencing clinical practice.

The study developed a clinical risk score based on suicidal ideation, sleep severity, and sex, assigning 3 points for ideation frequency, 2 points for sleep severity, and 1 point for male sex. The score achieved an AUC of 0.83 (95% CI, 0.77–0.89), indicating good discrimination in this sample.

• Low risk (0–2 points): 0.8% attempt rate
• Moderate risk (3–5 points): 4.2% attempt rate
• High risk (6–8 points): 18.7% attempt rate
• Very high risk (≥9 points): 45.3% attempt rate

The authors stated that prospective validation is needed before implementation.

In the full enrolled sample of 3,329 participants, 53 participants (1.59%) reported at least 1 suicide attempt during follow-up and 66 participants (2.14%) had at least 1 psychiatric hospitalization. There were 72 total suicide attempt events and 73 total hospitalization events over 24 weeks.

From the 3,087 participants assessed at baseline, the intention-to-treat sample included 2,815 participants with complete baseline data on key variables. In that intention-to-treat sample, 46 participants (1.63%) reported at least 1 suicide attempt, with 64 total suicide attempt events.

This was a nationwide, investigator-initiated, single-center, randomized, unblinded, pragmatic controlled trial in Brazil using data from the TelePSI COVID-19 project, with remote interventions delivered through digital platforms. Participants were health care professionals or trainees seeking psychological support and were followed with self-report assessments at baseline, 4 weeks, 12 weeks, and 24 weeks.

For this analysis, the authors used Cox proportional hazards models to examine associations between baseline predictors and later suicide attempts or psychiatric hospitalizations, adjusting for treatment allocation and using an intention-to-treat approach. Because the analysis of risk factors was observational, the findings identify associations rather than proving causation.

The main limitations were self-reported outcomes and predictors, observational risk-factor analyses, lack of objective sleep measures, possible residual confounding, and substantial attrition. The authors specifically noted that prior suicide attempts, trauma history, and substance use disorders were not measured, and these are important potential confounders.

They also reported that 56% of the sample was lost during follow-up and could not exclude the possibility that losses were related to suicide or psychiatric hospitalization. However, inverse probability weighting reduced imbalance, with all standardized differences below 0.1 after weighting, and the key findings remained significant in both weighted and unweighted models.