Design and implementation of an intelligent framework for supporting evidence-based treatment recommendations in precision oncology

BACKGROUNDThe advances in genome sequencing technologies have provided new opportunities for delivering targeted therapy to patients with advanced cancer. However, these high-throughput assays have also created a multitude of challenges for oncologists in treatment selection, demanding a new approach to support decision-making in clinics. METHODSTo address this unmet need, this paper describes the design of a symbolic reasoning framework using the method of hierarchical task analysis. Based on this framework, an evidence-based treatment recommendation system was implemented for supporting decision-making based on a patients clinicopathologic and biomarker profiles. RESULTSThis intelligent framework captures a six-step sequential decision process: (1) concept expansion by ontology matching, (2) evidence matching, (3) evidence grading and value-based prioritisation, (4) clinical hypothesis generation, (5) recommendation ranking, and (6) recommendation filtering. The importance of balancing evidence-based and hypothesis-driven treatment recommendations is also highlighted. Of note, tracking history of inference has emerged to be a critical step to allow rational prioritisation of recommendations. The concept of inference tracking also enables the derivation of a novel measure -- level of matching -- that helps to convey whether a treatment recommendation is drawn from incomplete knowledge during the reasoning process. CONCLUSIONSThis framework systematically encapsulates oncologists treatment decisionmaking process. Further evaluations in prospective clinical studies are warranted to demonstrate how this computational pipeline can be integrated into oncology practice to improve outcomes.