Point of Care RandOmisation Systems for Performing Embedded Comparative Effectiveness Trials Of Routine Treatments

Every day, doctors and nurses make hundreds of decisions about treatments - like when to start or stop them, or how frequently to give them. Ideally, decisions are based on gold standard evidence from Randomised Controlled Trials (RCTs). Unfortunately, for many treatments little or no evidence exists and clinicians must use knowledge and experience to decide what is best. As clinicians are all different, this leads to random variation in how treatments are given to patients. For example, magnesium is routinely given in intensive care to prevent abnormal heart rhythms. There is little evidence supporting this, and clinicians vary in how they administer magnesium. Traditional RCTs might be used to examine whether more magnesium is better than less magnesium, but this method is inefficient and expensive for investigating multiple comparative treatment questions. Clinical trials are becoming more efficient by using existing hospital computer systems to run them. However, research teams continue to perform tasks like randomisation manually. For questions like magnesium supplementation, which occur daily, this is labour intensive and infeasible. Hospital computer systems also possess mechanisms for prompting and alerting clinicians for particular decisions, reminding them of best practices, warning them of potential problems. These systems may be modified to allow clinicians to randomise patients, under specific conditions. The investigators propose to assess whether modified computer prompts can be used to highlight the magnesium supplementation decision to clinicians. These would prompt the clinician to evaluate the uncertainty around giving or withholding magnesium in that instance. If in agreement that the optimal decision is unclear, clinicians can choose to randomise the patient within a predetermined trial structure. If the clinician knows better, they may override the prompt and continue with their preference. In both cases, the system learns from the decision