Needle in a Haystack: Generating Audit Hypotheses for Clinical Audits of Hospitals

The purpose of the research is to identify actionable audit hypotheses, i.e., potential abnormality with respect to drugs prescribed, procedures undertaken or lab tests performed from clinical audit perspective. Structural and novel graphical representations are derived from data of a real hospital. 20 anomaly detection techniques, a novel association rule mining technique and four anomaly detection techniques on graphical data are applied to determine audit hypotheses from various perspectives. On correlating the details it is observed that the proposed methods capture inconsistencies or abnormalities on the data from multiple perspectives.We show that the anomaly detection techniques point out useful and actionable audit hypotheses in the clinical data.