Suicide is a global public health challenge, yet considerable uncertainty remains regarding the associations of both behaviour-related and physiological factors with suicide attempts (SA). Here we first estimated polygenic risk scores (PRS) for SA in 334,706 UK Biobank participants and conducted phenome-wide association analyses considering 2,291 factors. We identified 246 (63.07%) behaviour-related and 200 (10.41%, encompassing neuroimaging, blood and metabolic biomarkers, and proteins) physiological factors significantly associated with SA-PRS, with robust associations observed in lifestyle factors and mental health. Further case–control analyses involving 3,558 SA cases and 149,976 controls mirrored behaviour-related associations observed with SA-PRS. Moreover, Mendelian randomization analyses supported a potential causal effect of liability to 58 factors on SA, such as age at first intercourse, neuroticism, smoking, overall health rating and depression. Notably, machine-learning classification models based on behaviour-related factors exhibited high discriminative accuracy in distinguishing those with and without SA (area under the receiver operating characteristic curve 0.909 ± 0.006). This study provides comprehensive insights into diverse risk factors for SA, shedding light on potential avenues for targeted prevention and intervention strategies.

自杀是一项全球公共卫生挑战，但行为相关因素和生理因素与自杀未遂（SA）之间的关系仍然存在相当大的不确定性。在这里，我们首先估计了 334,706 名英国生物银行参与者的 SA 多基因风险评分 (PRS)，并考虑了 2,291 个因素进行了全表组关联分析。我们确定了 246 个（63.07%）行为相关因素和 200 个（10.41%，包括神经影像、血液和代谢生物标志物以及蛋白质）与 SA-PRS 显着相关的生理因素，并在生活方式因素和心理健康方面观察到了强有力的关联。涉及 3,558 个 SA 病例和 149,976 个对照的进一步病例对照分析反映了 SA-PRS 观察到的行为相关关联。此外，孟德尔随机化分析支持 58 个因素对 SA 的潜在因果影响，例如首次性交年龄、神经质、吸烟、整体健康评级和抑郁。值得注意的是，基于行为相关因素的机器学习分类模型在区分有和没有 SA 的人方面表现出很高的判别准确性（受试者工作特征曲线下面积为 0.909 ± 0.006）。这项研究提供了对 SA 的各种危险因素的全面见解，揭示了有针对性的预防和干预策略的潜在途径。