Background <p>Increasingly research studies use HES (Hospital Episode Statistics) data to report clinical outcomes. No data exists on the performance of individual ICD-10 (International Classification of Diseases 10th Revision) diagnostic codes for identifying major bleeding events. Data from the ROBOT-ACS study provide a unique opportunity to assess this compared to conventionally adjudicated bleeding by standard definitions.</p> Methods <p>A secondary analysis was performed on the 1172 HES records from ROBOT-ACS follow up data containing bleeding or anaemia codes. The 213 adjudicated major bleeds in ROBOT-ACS served as the gold standard comparator. Individual bleeding codes, and groups by type, were assessed for their positive predictive value (PPV).</p> Results <p>The PPV of most codes for major bleeding were poor, generally &lt; 50%. The best performing group of codes were relating to intracranial haemorrhage. 26 of 213 adjudicated bleeding events in ROBOT-ACS would be missed if anaemia codes were not considered.</p> Conclusions <p>The performance of diagnostic ICD-10 codes in HES, without further interrogation, for determining major bleeding events is poor. Whilst sensitivity is likely to be favourable, differentiating major bleeding is challenging. Options for using HES data to determine bleeding in cardiovascular studies would be either a hybrid approach, with HES screening followed by records review, or creating a new definition of significant bleeding using data more readily available in HES.</p> Trial registration <p>ROBOT-ACS is registered on clinicaltrials.gov. Unique identifier: NCT02484924. Registered 30/6/2015.</p>