Development of a Computerized Decision Support System for Treatment Planning and Outcome Prediction in Aneurysmal Subarachnoid Haemorrhage

Purpose The United States Academy of Engineering has stated that the provision of healthcare systems to provide ‘just in time, just for me’ advice at the point of care is one of the great engineering challenges of the 21st Century. Nowhere is the provision of such systems more important than in interventional neurovascular radiology. The rapid pace of change in the treatment of aneurysmal subarachnoid haemorrhage (ASAH) leads to significant problems in the assessment and validation of new therapeutic techniques and allocation of treatment resources. It is incumbent on the physician to assimilate vast quantities of rapidly changing literature in order to make informed treatment decisions and give tailored outcome predictions for individual patients. The current state of the art in computerised decision support system (DSS) are based upon Bayesian networks and the ‘just in time, just for me’ support scenario is now becoming a reality with developments in information management, data analysis and multi-core processor technology. However, the development of such systems is hampered by the techniques necessary to capture expert knowledge. The purpose of this project is to develop and validate a Bayesian DSS to provide real-time clinical advice regarding ASAH to clinicians in both neuroscience centres and primary referral sites. Materials & Methods We have developed a novel language-based software tool that is capable of capturing data from various knowledge bases. This includes sources as disparate as prospective studies and single case reports. It enables the assimilation of these disparate data streams into one searchable repository. The performance of the system to extract knowledge from different data sources was assessed by comparison with six experienced interventional neuroradiologists. Additional system testing looked at comparison between the system and domain experts performance regarding outcome prediction related to aneurysm morphology. Results The system has been populated with 298 papers giving a knowledgebase representing 44,364 patients with a total of 10,644 aneurysms all with associated outcome metrics. We found a mean rate of 125 data variables/publication (range 47-261 S.D. 47). Comparison of domain experts to DSS saw a mean data extraction gain of 35%/paper (range 2%-91% S.D. 22.6) and a mean gain of 8% (range 0-11% S.D. 4.6) over a synthesised domain expert created by amalgamation of all experts. The system identified over 300 separate aneurysmal descriptors with associated outcome metrics as defined in the literature, compared to 32 specific descriptors defined by domain experts.Conclusion We have shown that we can capture and assimilate information regarding ASAH into a single data repository with a system that can improve the neuro-radiologist gold standard. Analysis of the data generated by this tool has allowed us to develop a formal set of descriptive terms pertaining to cerebral aneurysms and define how these relate to the outcome of the disease process. This is an important step in the advancement of e-medicine resources. Such DSS need validation for their accepted use in clinical practice and a system such as the one presented here represents an important early step in their development, validation and general acceptance.