Explainable AI Enhances Precision in Clinical Trial Participant Selection for Depression
Overview
NetraAI, an explainable AI platform, significantly improved prediction of treatment response in a Phase II ketamine trial for treatment-resistant depression by identifying high-effect-size patient subpopulations. It outperformed traditional machine learning models by 25-30% in accuracy and achieved near-perfect specificity using a minimal set of clinical and MRI features.
Background
Clinical trial failures often result from patient heterogeneity diluting treatment effects, leading to low statistical power and ambiguous outcomes. Traditional subgroup analyses rely on univariate methods that are underpowered and typically post-hoc, limiting their utility in prospective trial enrichment. Artificial intelligence offers promise for identifying responder subgroups but faces challenges with small sample sizes and model interpretability. NetraAI addresses these issues by combining dynamical systems modeling, long-range memory feature selection, and large language model-generated insights to discover clinically meaningful patient Personas for precision enrichment.
Data Highlights
Metric
NetraAI
Traditional ML
Predictive Accuracy Improvement
~25-30%
Baseline
Predictive AUC Improvement (Clinical Variables)
+0.32
Baseline
Accuracy (MRI Features)
95%
Lower
Specificity (MRI Features)
100%
Lower
Key Findings
NetraAI identified a 10-variable clinical model improving predictive AUC by 0.32 over standard ML models.
An 8-MRI feature model from NetraAI achieved 95% accuracy and 100% specificity in detecting responders.
NetraAI outperformed traditional ML by approximately 25-30% in predictive accuracy for treatment outcomes.
The platform uses explainable, interpretable patient subpopulations (“Personas”) based on minimal variable combinations for clinical feasibility.
NetraAI’s approach effectively handles the small n, large p problem typical in early-phase trials without univariate variable pruning.
Large language model integration translates findings into transparent, clinically actionable inclusion/exclusion criteria.
Clinical Implications
NetraAI’s explainable AI approach enables prospective identification of patients most likely to benefit from therapy, improving trial enrichment and potentially increasing success rates. Its minimal variable sets facilitate practical screening during trial enrollment, supporting personalized medicine strategies in psychiatry and other therapeutic areas. Adoption of such platforms may reduce late-stage trial failures by addressing patient heterogeneity more effectively.
Conclusion
NetraAI demonstrates that explainable dynamical AI can leverage small, high-dimensional datasets to uncover clinically meaningful responder subgroups, enhancing predictive accuracy and specificity beyond traditional methods. This precision enrichment strategy holds promise for improving clinical trial outcomes and advancing personalized treatment approaches.
References
Original Article -- Utilizing Explainable AI for Enhanced Precision in Clinical Trial Participant Selection
