Automatic Analysis of Electronic Discharge Letters as a Means to Evaluate the Continuity of Information and of Patient Care

Joint Commission International standard 3.2 on Access to Care and Continuity of Care states that discharge letters should contain information about follow-up instructions of doctors to patients. We developed a text mining system to analyze a collection of 413 discharge letters of heart failure patients and checked their compliance with standard 3.2. We built a domain-specific ontology and a thesaurus and mined the collection with CASOS AutoMap. After validation, the system sensitivity was 0.484; specificity was 0.834; positive predictive value was 0.555; negative predictive value was 0.790. Improving these results requires more powerful natural language processing tools, but text mining seems a promising way to evaluate the continuity of information and of care.