Aernoud Fiolet

343 Accuracy of using routinely collected electronic healthcare data to identify cardiovascular endpoints adjudicated end points as reference standard. Additionally, positive and negative predictive values and Cohen’s kappa values were calculated. The main analysis included all participants from the four participating sites. To quantify the impact of potential misclassification of endpoints when using EHR data, event-free survival times were compared for both endpoint collection methods. Time-to-event was defined as the difference between the date of occurrence of the endpoint and the date of randomization. For patients without an endpoint, follow-up time was censored at the end of trial. Endpoints that occurred before randomization or after the close-out visit were excluded from the analysis. For all endpoints, Kaplan–Meier estimates, and curves of the event-free survival probabilities were calculated to evaluate the timing of endpoint occurrences using the two data collectionmethods. For atrial fibrillation and diabetes mellitus no time to event data were collected, in concordance with the original trial design. For all tests, the threshold for statistical significance was set at a p-value of 0.05. Multiple testing was not corrected for. Per the design of the current study, all analyses were done irrespective of treatment allocation. All analyses were performed using R software, version 4.0.3.19 Ethical oversight, author contributions and conflicts of interest All patients provided written informed consent to participate in the original trial. The trial protocol was approved by a centralized institutional review board (Medical Research Ethics Committees United, Nieuwegein) and registered in a clinical trial registry (LoDoCo2 Australian New Zealand Clinical Trials Registry number, ACTRN12614000093684). All authors contributed to the design of the study. PV,WD, MA, AM, EB, AM andAF collected and managed the data. The text- mining software was provided free of cost to the investigators. The vendors of the text-mining and EHR software had no role in the design or writing of the protocol or manuscript. All authors had unrestricted access to the data and vouch for the completeness and accuracy of the data and analyses.