Abstract P368: Developing Automated Methods for Disease Subtyping in UK Biobank: An Exemplar Study on Stroke

Abstract

Objective: In UK Biobank (UKB), a large population-based prospective study, cases of many diseases are ascertained through linkage to routinely collected, coded national health datasets. However routinely collected coded data cannot always provide sufficient accuracy or granularity (i.e. sub-phenotypes) for research studies. For example, while ischemic stroke codes appear accurate, the precision for hemorrhagic stroke codes (intracerebral hemorrhage [ICH] and subarachnoid hemorrhage [SAH]) may be as poor as <50%. We investigated whether automated analysis of radiology reports could improve disease subtyping in UKB, using stroke as an exemplar disease. Methods: From a sub-population of 17,249 UKB participants, we ascertained those with an incident stroke code and ≥1 clinical brain scan report. We used automated methods (a combination of natural language processing and clinical knowledge inference) on brain scan reports to assign a stroke subtype (ischemic vs ICH vs SAH) for each participant and assessed performance by precision (positive predictive value) and recall (sensitivity) at both entity and patient levels. Results: Of 225 participants with an incident stroke code, 207 had a relevant brain scan report. Entity level precision and recall ranged from 78% to 100%. Automated methods showed precision (positive predictive value) and recall (sensitivity) at patient level that were very good for ICH (both 89%), good for SAH (both 82%), but, as expected, lower for ischemic stroke (73%, and 64%, respectively), suggesting coded data remains the preferred method for identifying the latter stroke subtype (Table 1). Discussion: Future research should validate these findings in another dataset. Conclusion: Our novel automated method applied to radiology reports provides a feasible, scalable and accurate solution to improve disease subtyping when used in conjunction with administrative coded health data.