A hybrid decision support system using rule-based and AI methods: the OnCATs knowledge-based framework.

Abstract

Clinical decision support systems (CDSS) can improve evidence-based oncology care, but many rely on opaque AI models that limit transparency and reproducibility. Rule-based approaches provide interpretability but often lack adaptability, a critical issue in prostate cancer where decisions depend on tumor stage, PSA, Gleason score, comorbidities, and life expectancy. Bridging explainability and adaptability is essential for trustworthy decision support. To develop and evaluate OnCATs, a modular, explainable, hybrid-ready CDSS that encodes prostate cancer management guidelines in a machine-readable and auditable format. Evidence from 23 international guidelines was formalized into a JSON-based rule base executed through a forward-chaining inference engine. OnCATs supports three decision layers: (1) risk stratification, (2) treatment-pathway recommendation, and (3) prescription-level assistance for radiotherapy, brachytherapy, androgen deprivation therapy, and surgery. Feasibility was tested using ten published case reports. Performance was assessed with precision, recall, F1 scores, and descriptive concordance. OnCATs achieved perfect concordance for risk stratification (precision = 1.00, recall = 1.00, F1 = 1.00). Treatment-pathway concordance was 0.80 (F1 = 0.80). Prescription-level agreement ranged from 0.67 to 0.75 (mean F1 = 0.71). Divergences primarily reflected simplified life-expectancy modeling and incomplete case data. OnCATs demonstrates that transparent, rule-based reasoning can reproduce guideline-defined prostate cancer decisions with traceability. Limitations include the small sample size and reliance on secondary data. OnCATs operationalizes multi-source guidelines into an explainable, modular CDSS, providing a reproducible foundation for future integration of probabilistic and machine-learning methods.