Don't Let Notes Be Misunderstood: A Negation Detection Method for Assessing Risk of Suicide in Mental Health Records

摘要: Mental Health Records (MHRs) contain freetext documentation about patients’ suicide and suicidality. In this paper, we address the problem of determining whether grammatic variants (inflections) word “suicide” are affirmed or negated. To achieve this, populate annotate a dataset with over 6,000 sentences originating from large repository MHRs. The resulting has high InterAnnotator Agreement ( 0.93). Furthermore, develop propose negation detection method that leverages syntactic features text 1 . Using parse trees, build set basic rules rely on minimum domain knowledge render as binary classification (affirmed vs. negated). Since overall goal is to identify patients who expected be at risk suicide, focus evaluation positive (affirmed) cases determined by our classifier. Our approach yields recall (sensitivity) value 94.6% for an accuracy 91.9%. We believe can integrated other clinical Natural Language Processing tools in order further advance information extraction capabilities.