Background: Acute bleeding is a major public health problem. Tranexamic acid is an antifibrinolytic drug that reduces bleeding by inhibiting blood clot breakdown. Several clinical trials have assessed the effects of tranexamic acid for bleeding. To provide a valid and reliable estimate of the treatment effect, a trial must be well designed with a suitable outcome measure. Due to similarities in pathophysiology and the types of interventions used to treat acute bleeding, clinical trials assessing tranexamic acid often evaluate similar outcomes regardless of the site or cause of bleeding. These trials provide a rich resource for assessing the suitability of different outcome measures. They also deepen our understanding of the natural history of bleeding and the mechanism of action of tranexamic acid. This project aims to inform the choice of outcome measures by exploring the effects of tranexamic acid on bleeding in large clinical trial datasets. Methods: The effects of tranexamic acid on acute bleeding were assessed by applying a range of methodological approaches to large clinical trial datasets. The tendency for non-differential misclassification of outcomes to cause bias towards the null was exploited as a tool to study the biological effects of tranexamic acid. The impact of misclassification was investigated by varying assumptions about the empirical induction period and locating the least diluted measure of effect. This allowed hypotheses about the biological effects of tranexamic acid to be refined and the selection of outcome measures better able to capture these effects. Descriptive and multivariable analyses of baseline characteristics and the timing and frequency of various outcome events in patients with acute bleeding were used to investigate the natural history of bleeding. Main findings: An outcome must have the potential to be affected by the trial intervention, be amenable to unbiased measurement, be sufficiently common, and be clinically relevant and important to patients. There is a window of opportunity for a treatment to exert its effects. Inappropriate assumptions about the time from causation to detection (the empirical induction period) can cause non-differential outcome misclassification. Tranexamic acid is most effective when given soon after bleeding onset and appears to work mainly by reducing bleeding on the day of onset. If we intervene too late in the disease process, when the outcome is inevitable or the targeted biological pathways have ceased, there will be no potential for benefit. When the outcome measure includes events that fall outside of the etiologically relevant period or biological pathway, the effect estimate is diulted towards the null. Impact of work: My work informed the selection of the primary outcome measure in the CRASH-3, HALT-IT and WOMAN-2 trials. By generating new insights into the natural history of acute bleeding and the mechanism of action of tranexamic acid, it also helped the implementation of the trial results, potentially contributing to improvements in patient care, and has influenced research on haemostatic treatments more generally. Strengths and weaknesses: The large, high-quality trial datasets comprised over 70,000 patients with almost complete follow-up and little missing data, providing reliable effect estimates, and allowing meaningful subgroup and sensitivity analyses and the assessment of different outcome measures. The biological effect of tranexamic acid and impact of dilution from outcome misclassification is consistent with biology and across multiple trials. Some measurement error in the timing of events is inevitable, and some results are imprecise so chance cannot be ruled out as a potential alternative explanation. Implications for future research: Dilution from outcome misclassification is a common issue in clinical trials, with different implications in superiority trials compared to equivalence or non-inferiority trials. Given the rising cost of research, trials need to be efficient and cost effective. The use of appropriate outcome measures that capture the biological effect of a treatment can reduce non-differential misclassification, increasing statistical power. New information generated by trials as they are underway can be used to inform adaptive trial design, including the choice of outcome measures. Conclusions: The concepts presented in this thesis could be applied to clinical trials in other disease areas and might help to inform the choice of outcome measures and generate knowledge on the cause-effect relationship between study interventions and outcome.