Accuracy of ICD-10AM coding in neurologist confirmed transient ischemic attack (TIA).

Background: Diagnostic coding is frequently used in stroke research. However, few data exist regarding the accuracy of Transient Ischemic Attack (TIA) codes in Australia. Aim(s): To determine the accuracy of ICD-10 Australian Modification (AM) diagnostic codes assigned to patients with neurologist diagnosed TIA. Method(s): ICD-10AM codes were identified for patients with suspected TIA treated at Monash Medical Centre, Australia (2004-2007). Coding accuracy was assessed against final neurologist diagnosis using the 24-hour criterion (clinical TIA). Assignment of a TIA code of G45.3, G45.8 or G45.9 to a patient with neurologist confirmed TIA was considered true positive and TIA mimics were considered false positives. Result(s): ICD-10AM codes were available for 336/488 (68.9%; 95% Confidence Interval 64.6%-72.8%) patients. Clinical TIA was confirmed in 209/336 (62.2%; 56.9%-67.2%). Evaluation of ICD-10AM codes for clinical TIA revealed: sensitivity 0.76 (0.69-0.81), specificity 0.43 (0.34-0.51), Positive Predictive Value 0.68 (0.62-0.74), Negative Predictive Value 0.51 (0.42-0.61). Mean age was slightly higher in those assigned a TIA code (66.4+/-13.7) compared with a non-TIA code (62.3+/-15.6; p < 0.02). Patients with symptoms lasting less than one hour were more likely to be assigned a TIA code than those with symptoms lasting more than one hour (p < 0.001). Discussion(s): ICD10AM codes are sensitive for clinical TIA, but insufficiently specific. These data may help to estimate measurement error when using administrative datasets for TIA research.