Investigation of Protective and Risk Factors for Aboriginal Perinatal Mental Health

Title: Investigating Protective and Risk Factors and Predictive Insights for Aboriginal Perinatal Mental Health: Explainable Artificial Intelligence Approach
Authors:
- Guanjin Wang1,2, PhD
- Hachem Bennamoun1, MIT
- Wai Hang Kwok3, PhD
- Jenny Paola Ortega Quimbayo1, MIT
- Bridgette Kelly2, RMT
- Trish Ratajczak2, RMT
- Rhonda Marriott2, PhD
- Roz Walker2, PhD
- Jayne Kotz2, PhD
Affiliations:
- 1School of Information Technology, Murdoch University, Perth, Australia
- 2Ngangk Yira Institute for Change, Murdoch University, Perth, Australia
- 3School of Nursing and Midwifery, Edith Cowan University, Perth, Australia
Corresponding Author: Guanjin Wang, PhD
- Contact: 90 South St Murdoch WA Perth, 6150 Australia
- Phone: 61 89360735
- Email: Guanjin.Wang@murdoch.edu.au

Perinatal depression and anxiety greatly affect maternal and infant health.
Potential severe outcomes include:
- Preterm birth
- Suicide
Aboriginal women are at heightened risk due to
- Long-term impacts of colonization
- Cultural disruption
Successful co-design of the Baby Coming You Ready (BCYR) model for holistic, strengths-based assessment has been implemented, replacing traditional risk-based screening.
Issue: Overreliance on psychological risk scores by some health professionals can overlook cultural strengths and protective factors.

To investigate explainable artificial intelligence (XAI) techniques for developing predictive models of psychological distress in Aboriginal mothers that identify protective and risk factors.
Emphasis on culturally-informed assessments and transparent decision-making.

Deidentified data from 293 Aboriginal mothers participating in the BCYR program from September 2021 to June 2023 across six healthcare services in Perth, WA.
Initial dataset contained numerous variables, ultimately reducing to 20 predictors after feature selection and expert input.
The Kessler-5 scale (K5) used to assess psychological distress.
Used various machine learning models:
- Random Forest (RF)
- CatBoost (CB)
- Light Gradient-Boosting Machine (LightGBM)
- Extreme Gradient Boosting (XGBoost)
- K-Nearest Neighbor (KNN)
- Support Vector Machine (SVM)
- Explainable Boosting Machine (EBM)
Post hoc explanation techniques such as Shapley Additive Explanations (SHAP) were utilized to interpret model predictions.

EBM outperformed others with performance metrics:
- Accuracy: 0.849 (95% CI 0.8170-0.8814)
- F1-score: 0.771 (95% CI 0.7169-0.8245)
- Area Under Curve (AUC): 0.821 (95% CI 0.7829-0.8593)
RF's results were also notable:
- Accuracy: 0.829 (95% CI 0.7960-0.8617)
- F1-score: 0.736 (95% CI 0.6859-0.7851)
- AUC: 0.795 (95% CI 0.7581-0.8318)
Key influential factors identified included sentiments around loneliness and self-blame, alongside family pride and management of daily life.

XAI models can effectively predict psychological distress in Aboriginal mothers, offering transparent explanations of influential factors which may guide healthcare decisions for improved mental health interventions.

Explainable AI, perinatal mental health, AI-assisted decision-making, Aboriginal women, psychological risk, machine learning, protective factors, risk factors, cultural strengths

Perinatal depression and anxiety (PNDA) adversely affect maternal and infant health.
Associated with serious outcomes like suicidal thoughts and self-harm during and post-pregnancy.
Studies indicate correlations between PNDA and various adverse outcomes:
- Preterm birth, stillbirth, low birth weight
- Long-term effects on infant relationships

Many Aboriginal women enjoy strong social connections; however, population-level risk remains elevated due to:
- Effects of colonization
- Persistent inequities and trauma
Statistics indicating higher vulnerabilities:
- Aboriginal women have a 38% higher chance for depression during pregnancy
- 79% more likely to face mental health challenges whilst pregnant
- 30% post-birth complexities compared to non-Aboriginal women.

Current perinatal care often culturally insensitive:
- Risk-focused assessments lead to alienation in Aboriginal women
The BCYR program was developed as a strengths-based, culturally safe, and holistic alternative.
Successful engagements led to positive outcomes in maternal trust and accurate health assessments, but cultural incompetence among some health practitioners remained an issue.

Data collected from the BCYR pilot program included six service locations in WA.
Ethical clearance was achieved through multiple institutional committees.

Analysis focusing on seven machine learning models using tenfold cross-validation to validate performance.

Table display of model performance demonstrating RF and EBM as superior in predictive metrics:
- Best performance metrics across accuracy and recall for EBM and RF were highlighted.

Key factors influencing predictions were identified using local and global interpretative models to provide insights into risk and protective factors, shaping future assessments.

The robustness of the EBM and RF predictive models indicates the effectiveness of ensemble models in perinatal mental health predictions.
Feature importance analysis revealed critical insights into factors contributing to mental health risks.

Acknowledgment of limitations including nonrandom sampling and inherent biases. - Recommendations include expanding dataset diversity and improving the clarity of AI outputs for clinical applications.