Large-scale, pragmatic randomized trials in the era of big data, precision medicine and machine learning. Valid and necessary, or outdated and a waste of resources?

During the past decade, more and more large-scale pragmatic clinical trials have been carried out in orthopedic trauma surgery. This trend is fueled by the common belief that the larger the numbers in a trial, the broader the eligibility criteria, and the less strict the regimentation of local treatment standards by protocol, the more trustworthy the findings would be. However, it must also be taken into account that the precision of an outcome measurement does not depend on the sample size alone, but the homogeneity of the studied population. Consequently, a small trial with stringent entry and assessment criteria may offer similarly precise answers as a large trial with less strict entry and assessment criteria because of the basic mathematical correlation between standard deviation and standard error of the mean. There is now a lively and controversial debate about the role of randomized controlled trials (RCT) in an era of stratified medicine driven by the ever increasing understanding and clinical measurability of molecular pathways, making a certain intervention more effective in patients who show a distinct genetic variant. Cluster and pattern recognition by artificial intelligence (AI) and its methodological variety applied to huge datasets and population-based cohorts further propel the spiral of knowledge. Advanced adaptive RCT concepts like enrichment designs, basket and bucket trials, master protocols etc. were developed to combine classic principles of the scientific method with big data, the latter of which have not arrived yet in trauma care. In spite of all biomedical and methodological achievements made, surprisingly such key questions remain unanswered as a) is a certain treatment causally responsible for making a difference in patient-centered outcomes compared to placebo, a control treatment, or the standard of care, b) do the results of a controlled experiment are relevant enough to change clinical practice, and c) under which conditions and assumptions shall we conduct large-scale pragmatic RCTs, focused confirmatory RCTs, or personalized analyses with or without AI support. (c) 2022 Elsevier Ltd. All rights reserved.