Identifying behaviour-related and physiological risk factors for suicide attempts in the UK Biobank

Neurology Information Science Artificial Intelligence Medicine Psychiatry Environmental and Public Health Genetics Clinical Psychology Life Sciences
suicide attempts behaviour risk factors physiological risk factors UK Biobank multifactor analysis

Research Background:

Suicide is a global public health challenge, but there is still significant uncertainty regarding the relationship between behavioral and physiological factors and suicide attempts (SA). Previous studies often focus on limited hypothesized factors such as mental illnesses (e.g., depression), personality and psychological traits (e.g., hopelessness), and social and family factors (e.g., low social support and life stress). This narrow perspective may lead to other risk factors being overlooked. To fill these research gaps, our research team conducted a large-scale systematic analysis and Mendelian randomization analysis to identify potential behavioral and physiological risk factors associated with SA in the UK Biobank dataset.

Source of Research:

This paper was written by the following researchers: Bei Zhang, Jia You, Edmund T. Rolls, Xiang Wang, Jujiao Kang, Yuzhu Li, Ruohan Zhang, Wei Zhang, Huifu Wang, Shitong Xiang, Chun Shen, Yuchao Jiang, Chao Xie, Jintai Yu, Wei Cheng, and Jianfeng Feng. The affiliated institutions include Fudan University, University of Oxford, among others. The paper was published in “Nature Human Behaviour.”

Research Process:

This study estimated polygenic risk scores (PRS) related to SA using data from 334,706 participants in the UK Biobank and conducted a phenome-wide association study (PheWAS) covering 2291 factors. Subsequently, a case-control analysis involving 3558 SA cases and 149,976 controls was conducted to validate the behavioral and physiological risk factors. Further, Mendelian randomization analysis was used to evaluate the potential causal relationships of these factors with SA. Finally, a machine learning classification model based on behavioral factors was constructed to distinguish individuals with or without a history of SA.

Research Details:

a) Workflow of the Study

  1. Estimation of Polygenic Risk Scores (PRS):

    • Data Source: 334,706 participants from the UK Biobank.
    • Analysis Method: PRS was used as a proxy for SA risk, analyzing 2291 factors across behavioral and physiological phenotypes.

  2. Phenome-Wide Association Study (PheWAS):

    • Considered Factors: A total of 2291, divided into 12 categories.
    • Sample Characteristics: 53.59% of participants were female, with an average age of 56.91 years.
    • Statistical Results: Found 246 (63.07%) behavioral phenotypes and 200 (10.41%) physiological phenotypes significantly associated with SA-PRS.

  3. Case-Control Analysis:

    • Participants: 3558 SA cases and 149,976 controls.
    • Results: 83% of behavioral factors and 37% of physiological factors also showed significant associations in the case-control analysis.

  4. Mendelian Randomization Analysis:

    • Supplemented by two-sample MR analysis, validated the potential causal relationships of significant factors related to SA.
    • Results: 57 behavioral factors and 1 biomarker showed causal relationships after Bonferroni correction.

  5. Machine Learning Classification Model:

    • The model based on behavioral factors had an AUC of 0.909±0.006, demonstrating high discrimination accuracy.

b) Main Research Results

  1. PheWAS Analysis Results:

    • A total of 246 behavioral phenotypes (including 11 socio-demographic factors, 43 lifestyle factors, 11 early life and family history factors, 36 physical measurements, 16 cognitive functions, and 129 (including multiple classification associations) mental health factors) showed significant associations with SA-PRS.

  2. Associations of Physiological Phenotypes:

    • Including 20 neuroimaging phenotypes (e.g., gray matter volume and white matter microstructure), 76 blood and metabolic biomarkers (including white blood cell count, blood biochemistry, and nuclear magnetic resonance metabolites), and 104 proteins.

  3. Risk Factors Confirmed by Case-Control Analysis:

    • 277 behavioral phenotypes and 51 physiological phenotypes showed significant associations in the case-control analysis, with mental health factors being the most significant among behavioral phenotypes.

c) Research Conclusions

  1. Identification of Causal Relationships:

    • 57 behavioral factors and 1 biomarker were confirmed as causal factors for SA after Bonferroni correction.

  2. Importance of the Machine Learning Model:

    • The machine learning model based on behavioral factors showed high discrimination accuracy for individuals with or without a history of SA, with mental health and lifestyle variables being highly relevant.

  3. Importance of Mental Health:

    • Depressive symptoms and neuroticism scores were particularly important in predicting SA risk.

Research Highlights:

  1. Comprehensive Analysis Method:

    • Combined large-sample PheWAS analysis with two-sample Mendelian randomization methods.

  2. Identification of Multidimensional Risk Factors:

    • Included not only behavioral factors but also neuroimaging, blood biomarkers, and proteins as physiological factors.

  3. Application of Machine Learning Models:

    • Demonstrated the efficiency of machine learning models based on behavioral factors in SA risk assessment for the first time.

  4. Significant Contribution of Mental Health:

    • Emphasized the core role of mental health factors in SA risk.

Scientific and Practical Value of the Study:

  1. Prevention and Intervention Strategies:

    • The study results provide important basis for targeted prevention and intervention strategies, helping to identify high-risk individuals early.

  2. Policy Making and Public Health:

    • Provide data support for public health policymakers aiming to reduce the incidence of suicidal behavior.

Gender Stratification Analysis:

The study also found 203 behavioral factors and 87 physiological factors significantly associated in females, whereas 199 behavioral factors and 73 physiological factors were significantly associated in males.

Limitations of the Study:

  1. Sample Characteristics:

    • Mainly middle-aged and elderly individuals of European descent, which may limit the generalizability of the findings.

  2. Healthy Volunteer Bias:

    • Volunteers’ health conditions might be better, potentially limiting the extrapolation of the study results.

Future Research Directions:

  1. Longitudinal Data Analysis:

    • Longer time-span longitudinal studies are needed to capture dynamic changes before and after suicidal behavior.

  2. Studies in Different Races and Regions:

    • Expand sample size to include individuals of different racial and cultural backgrounds to improve the generalizability of the findings.

Through the above analysis, this study comprehensively reveals the mechanisms of the influence of behavioral and physiological factors on suicide attempts, providing valuable data support and scientific basis for future prevention and intervention measures.