Using Published Medical Results and Non-homogenous Data in Rule Learning (original) (raw)

Many factors limit researchers from accessing studies' original data sets. As a result, much medical and healthcare research is based off of systematic reviews and meta-analysis of published results. However, when research involves the use of aggregated data from multiple studies, traditional machine learning-based means of analysis cannot be used. This paper describes diversity of data and results available in published manuscripts, and relates them to a rule learning method that can be applied to build classification and predictive models from such input. The method can be used to support meta-analysis and systematic reviews. Two application areas are used to illustrate the discussed issues: diagnosis of liver diseases in patients with metabolic syndrome, and detection of polycystic ovary syndrome.