Background Military veterans are at increased risk of firearm injury (FAI) compared to non-military veterans in the USA. To inform effective programs for the prevention of, or rehabilitation from, firearm injuries, it is important to examine patterns of FAI and intent. The USA Veterans Health Administration (VHA) is an integrated, national healthcare system providing care for >9 million eligible military veterans. Thus, VHA healthcare data contain records for a large population in which to examine patterns of FAI in the USA. However, the validity of using VHA healthcare data for identifying FAIs and categorizing intent has not been evaluated.

Objective We measured the accuracy of International Classification of Diseases (ICD) – 9th and 10th Revisions – Clinical Modification external cause codes in VHA data to identify FAIs and their intent.

Methods Among veterans with FAI-related visits to VHA healthcare between 2010–2019 (n=11,380), we manually reviewed electronic health records for a random sample of 600. Records were double-coded by two independent reviewers and coding discrepancies were adjudicated by consensus. Positive predictive value (PPV) was calculated to estimate the accuracy of ICD codes to correctly identify FAIs. Among true FAIs, overall concordance, sensitivity, and specificity were then calculated for categories of intent (assault, self-harm, unintentional, or other) using intent codes, determined from manually-reviewed healthcare records, as the criterion standard.

Results Among 600 manually-coded records, ICD external cause codes correctly identified 527 FAIs (PPV=87.8%). Unintentional FAIs were the most frequent manually-coded and ICD-coded intent (45.5% and 70.0%, respectively). Overall concordance for unintentional FAI was 68.1% (sensitivity = 90.7%; specificity = 49.2%), whereas concordance for assault was 78.6% (sensitivity = 34.9%; specificity = 93.9%) and for self-harm was 92.2% (sensitivity = 58.4%; specificity = 98.6%).

Conclusions ICD external cause codes tended to accurately identify true FAIs among military veterans treated in the VHA healthcare system; however, measures of validity for categorizing intent varied. Concordance was high for self-harm, but less so for other intents. Firearm injury research based on ICD external cause codes should account for potential misclassification of injury cause and intent, at least when used with VHA healthcare data in the USA.