Performance evaluation for semantic-based risk factors extraction from clinical narratives

Precision medicine is the new perspective in healthcare that requires a personalized diagnosis and treatment plan for a patient. This new approach mandates that clinical decision support system (CDSS), an essential element of preventive and precision medicine processes, uses state-of-the-art technologies in terms of clinical knowledge extraction. To assist a physician's precise prognosis using CDSS, it is important that a patient's health data is properly and automatically analyzed. The unstructured part of the data in electronic health records (EHR) is critical, as it may contain hidden risk factors. We propose a new approach for CDSS to extract risk factors concepts from the clinical narratives using natural language processing techniques (NLP) and semantic web technologies (SWT). We evaluate our model using a case study dataset of patients' records with venous thromboembolism (VTE). Our model extracts risk factors of VTE to make a prognosis. Results of proposed technique yielded precision of 85% and recall of 84% to identify and extract risk factors concepts.