Developmental predictors of suicide attempts from childhood to early adulthood: a 15-year prospective cohort study

Suicide attempts in childhood and adolescence are common, recurrent, and the strongest predictor of death by suicide; yet their developmental antecedents remain poorly understood, particularly in low- and middle-income countries. We aimed to identify childhood factors associated with incidence, onset timing, and number of suicide attempts and to quantify the population impact of modifiable risk factors.

We analyzed data from the Brazilian High-Risk Cohort Study, a community-based cohort of 2511 children aged 6-14 years followed for 15 years. Risk factors across genetic, perinatal, sociodemographic, family, adversity, cognitive, and clinical domains were examined using logistic regression (OR), Cox proportional hazards (HR), and quasi-Poisson models (IRR), were examined for three outcomes (cumulative incidence, onset timing, and number of suicide attempts), with all predictors entered simultaneously and bootstrap resampling. Population attributable fractions were calculated for modifiable factors.

Of 2060 participants (48.0% female unweighted, 47.3% weighted), 309 (15.0%) reported suicide attempts over 24,756.5 person-years (mean onset 17.8 years). Female sex (OR 3.06, 95% CI 2.28-4.17; HR 2.88, 2.25-3.69; IRR 3.12, 2.19-4.49) and childhood threat (OR 1.26, 1.08-1.47; HR 1.21, 1.06-1.38; IRR 1.26, 1.05-1.51) were associated with all three outcomes. Caregiver suicide attempt (OR 2.07, 1.40-3.14; HR 1.88, 1.34-2.63) and externalizing disorders (OR 1.52, 1.05-2.15; HR 1.50, 1.07-2.08) were associated with incidence and earlier onset. PRS-depression (IRR 1.39, 1.00-1.94) and gestational diabetes (IRR 2.18, 1.13-3.83) were associated with increased attempt frequency. Child-reported bullying (OR 1.73, 1.22-2.45) and parent-reported physical abuse (OR 1.96, 1.24-3.10) retained association after mutual adjustment in threat-component models. Population attributable fractions were substantial for caregiver suicide attempt (14.1%), caregiver internalizing disorders (11.6%), high childhood threat (>1 SD; 11.5%), and externalizing disorders (10.2%). In lethality analyses, PRS-depression (OR 2.21, 1.09-4.43; HR 1.98, 1.13-3.60; IRR 2.68, 1.43-4.97), childhood threat (HR 1.40, 1.09-1.80; IRR 1.34, 1.03-1.90), and maternal alcohol use in pregnancy (HR 1.92, 1.12-3.26) were associated with medically serious attempts. Prediction models showed modest discrimination (mean area under the curve [AUC] 0.665, SD 0.033, across 200 repeated test splits).

These findings highlight three potentially modifiable or clinically actionable domains - childhood threat, caregiver suicidal behaviour history, and childhood externalizing disorders - that may be amenable to prevention even when genetic liability is present. Despite multi-domain predictors, predictive performance remained limited, consistent with a complementary role for population-level prevention alongside efforts to improve risk prediction.

CNPq, FAPESP, ERC, UK MRC, and Banco Industrial do Brasil S/A. Full grant details are provided in the manuscript.