Are decision trees a feasible knowledge representation to guide extraction of critical information from randomized controlled trial reports?

BACKGROUND This paper proposes the use of decision trees as the basis for automatically extracting information from published randomized controlled trial (RCT) reports. An exploratory analysis of RCT abstracts is undertaken to investigate the feasibility of using decision trees as a semantic structure. Quality-of-paper measures are also examined. METHODS A subset of 455 abstracts (randomly selected from a set of 7620 retrieved from Medline from 1998 - 2006) are examined for the quality of RCT reporting, the identifiability of RCTs from abstracts, and the completeness and complexity of RCT abstracts with respect to key decision tree elements. Abstracts were manually assigned to 6 sub-groups distinguishing whether they were primary RCTs versus other design types. For primary RCT studies, we analyzed and annotated the reporting of intervention comparison, population assignment and outcome values. To measure completeness, the frequencies by which complete intervention, population and outcome information are reported in abstracts were measured. A qualitative examination of the reporting language was conducted. RESULTS Decision tree elements are manually identifiable in the majority of primary RCT abstracts. 73.8% of a random subset was primary studies with a single population assigned to two or more interventions. 68% of these primary RCT abstracts were structured. 63% contained pharmaceutical interventions. 84% reported the total number of study subjects. In a subset of 21 abstracts examined, 71% reported numerical outcome values. CONCLUSION The manual identifiability of decision tree elements in the abstract suggests that decision trees could be a suitable construct to guide machine summarisation of RCTs. The presence of decision tree elements could also act as an indicator for RCT report quality in terms of completeness and uniformity.