Bridging the Trust Gap: Clinician-Validated Hybrid Explainable AI for Maternal Health Risk Assessment in Bangladesh

TL;DR

The paper tackles the trust gap in machine learning–driven maternal health risk assessment in resource-constrained settings by proposing a hybrid explainable AI framework that fuses ante-hoc fuzzy logic with post-hoc SHAP/LIME explanations. It validates the approach with 14 Bangladeshi clinicians on a dataset of $N=1{,}014$ records, achieving $88.67\%$ accuracy and a ROC-AUC of $0.9703$, while showing a strong clinician preference for hybrid explanations ($71.4\%$) and a moderate level of trust ($54.8\%$ yes). Key findings include SHAP identifying healthcare access as the primary predictor and the fuzzy risk score ranking third, supporting clinical knowledge integration, as well as qualitative insights on missing data such as obstetric history and gestational age, and practical deployment barriers like connectivity and training. The work demonstrates that integrating interpretable fuzzy rules with feature-importance explanations improves utility and trust for maternal health decision support, offering concrete guidance for XAI deployment in LMIC healthcare settings.

Abstract

While machine learning shows promise for maternal health risk prediction, clinical adoption in resource-constrained settings faces a critical barrier: lack of explainability and trust. This study presents a hybrid explainable AI (XAI) framework combining ante-hoc fuzzy logic with post-hoc SHAP explanations, validated through systematic clinician feedback. We developed a fuzzy-XGBoost model on 1,014 maternal health records, achieving 88.67% accuracy (ROC-AUC: 0.9703). A validation study with 14 healthcare professionals in Bangladesh revealed strong preference for hybrid explanations (71.4% across three clinical cases) with 54.8% expressing trust for clinical use. SHAP analysis identified healthcare access as the primary predictor, with the engineered fuzzy risk score ranking third, validating clinical knowledge integration (r=0.298). Clinicians valued integrated clinical parameters but identified critical gaps: obstetric history, gestational age, and connectivity barriers. This work demonstrates that combining interpretable fuzzy rules with feature importance explanations enhances both utility and trust, providing practical insights for XAI deployment in maternal healthcare.

Figures (3)

• Figure 1: Examples of fuzzy rules used in the ante-hoc XAI layer. The rules combine clinical parameters using logical operators to produce interpretable risk assessments.
• Figure 2: Comparison of XAI methods: (a) SHAP global feature importance, (b) LIME local explanations, and (c) Fuzzy rule-based explanations. The hybrid approach combines all three for comprehensive explanations.
• Figure 3: Summary of clinician validation results: (a) Explanation preferences across cases, (b) Trust levels distribution, and (c) Clarity ratings by case complexity.