|
Abstract:
|
In order to evaluate the efficacy and safety of diagnostic or prognostic algorithms in clinical practice, formal clinical trials are needed. Such trials can play a key role in evaluating the impact of an institution-level (e.g., hospital-level) policy whereby all eligible patients within an institution would have the algorithm applied to their care. In this case, for example, hospitals would be randomized to implement the new algorithm or to use their current standard of care/decision-making. In such a “cluster randomized trial”, a full picture of the impact of hospital-level implementation of the new algorithm could be assessed through a combination of clinical outcomes together with cost, time, and other such health economic measures. Importantly, negative outcomes such as misdiagnosis, mis-treatment and extended hospital stays could be evaluated. In this talk, we will introduce the cluster randomized trial design and will illustrate the importance of accounting for clustering of outcomes in both the design and analysis phases. We will talk about a range of variants of the design including the parallel-arm, crossover and stepped-wedge design and will consider how such design
|
