Estimating the Number of Self-Harm Events Not Identified by Encounter Diagnoses in Health Records

BackgroundPublic health and research assessment of self-harm or suicide attempts must sometimes rely on health system records. Previous research regarding accuracy of encounter or billing diagnoses to identify self-harm or suicide attempt has yielded mixed results. MethodsUsing data from a large pragmatic trial of outreach to prevent suicide attempt, investigators identified ICD-10-CM diagnostic codes in categories where suicide attempt or self-harm might be missed or misclassified: injuries and poisonings coded as having "undetermined intent, injuries and poisonings coded as accidental, and injuries with no coding of intent. For injury and poisoning events with these selected codes, relevant text was extracted from full-text electronic health records and rated by a blinded panel of investigators as indicating or not indicating self-harm ResultsDiagnostic codes selected for review include 43 codes for injuries and poisonings with undetermined intent, 26 codes for accidental injuries and poisonings, and 46 codes for injuries without coding of intent. Among injury and poisoning events in this sample, full-text records were available for review for 60 coded as undetermined intent, 173 coded as accidents, and 288 with no coding of intent. After review of text, raters judged 46 (77%), 152 (88%), and 275 (95%) respectively as having adequate text documentation for confident rating of self-harm intent. Among those with adequate text for confident rating, the proportions of events classified as due to self-harm were 19% for injuries and poisonings coded as undetermined intent, 6% for injuries and poisonings coded as accidents, and 15% for injuries without coding of intent. ConclusionsIn this sample of people at high risk for self-harm or suicide attempt, we low rates of misclassified self-harm events among injuries and poisonings coded as having undetermined intent, accidental intent, or no coding of intent. In this sample, inclusion of all misclassified events identified through this search process would increase the total number of self-harm events identified by approximately 10%.