Efficient Research Design

Edward C F Wilson, Miranda Mugford, Garry Barton, Lee Shepstone

In designing economic evaluations alongside clinical trials, analysts are frequently faced with alternative methods of collecting the same data, the extremes being top-down ("gross costing") and bottom-up ("micro-costing") approaches. A priori, bottom-up approaches may be considered superior to top-down approaches but are also more expensive to collect and analyze. In this article, we use value-of-information analysis to estimate the efficient mix of observations on each method in a proposed clinical trial. By assigning a prior bivariate distribution to the 2 data collection processes, the predicted posterior (i.e., preposterior) mean and variance of the superior process can be calculated from proposed samples using either process. This is then used to calculate the preposterior mean and variance of incremental net benefit and hence the expected net gain of sampling. We apply this method to a previously collected data set to estimate the value of conducting a further trial and identifying the optimal mix of observations on drug costs at 2 levels: by individual item (process A) and by drug class (process B). We find that substituting a number of observations on process A for process B leads to a modest £35,000 increase in expected net gain of sampling. Drivers of the results are the correlation between the 2 processes and their relative cost. This method has potential use following a pilot study to inform efficient data collection approaches for a subsequent full-scale trial. It provides a formal quantitative approach to inform trialists whether it is efficient to collect resource use data on all patients in a trial or on a subset of patients only or to collect limited data on most and detailed data on a subset.