An XML Based Framework for Merging Incomplete and Inconsistent Statistical Information from Clinical Trials

Meta-analysis is a vital task for systematically summarizing statistical results from clinical trials that are carried out to compare the effect of one medication (or other treatment) against another. Currently, most meta-analysis activities are done by manually pooling data. This is a very time consuming and expensive task. An automated or even semi-automated tool that can support some of the processes underlying meta-analysis is greatly needed. Furthermore, statistical results from clinical trials are usually represented as sampling distributions (i.e., with the mean value and the SEM). When collecting statistical information from reports on clinical trials, not all reports contain full statistical information (i.e., some do not provide SEMs) whilst traditional meta-analysis excludes trials reports that contain incomplete information, which inevitably ignores many trials that could be valuable. Furthermore, some trials results can be significantly inconsistent with the rest of trials that address the same problem. Therefore, highlighting (resp. removing) such inconsistencies is also very important to reveal (resp. reduce) any potential flaws in some of the trials results. In this paper, we aim to design and develop a framework that tackles the above three issues. We first present an XML-based merging framework that aims to merge statistical information automatically with the potential to add a component to extract clinical trials information automatically. This framework shall consider any valid clinical trial including trials with partial information. We then develop a method to analyze inconsistencies among a collection of clinical trials and if necessary to exclude any trials that are deemed to be illegible. Finally, we use two sets of clinical trials, trials on Type 2 diabetes and on neurocognitive outcomes after off-pump versus on-pump coronary revascularisation, to illustrate our framework. keywords XML, Meta-analysis, Merging, Incomplete information, Sampling distribution, Semantic heterogeneity, Information Extraction