From Machine Learning to Knowledge-Based Decision Support—A Predictive-Model-Markup-Language-to-Arden-Syntax Transformer for Decision Trees

Abstract

Arden Syntax is an HL7 International standard for the representation and execution of clinical knowledge in knowledge-based clinical decision support (CDS) systems. The predictive model markup language (PMML) specifies a file format for the representation and exchange of statistical and data mining models. To use those machine-learned models in Arden-Syntax-based CDS systems, the PMML files have to be transformed into an Arden Syntax representation. A PMML-to-Arden-Syntax transformer was created to process PMML structures and generate the Arden Syntax code. It employs the extensible stylesheet language transformation to create Arden Syntax medical logic modules (MLMs) out of PMML files. The transformer may create multiple MLM files from a single PMML. Currently the transformer is able to transform decisions tree models only. Its transforming capabilities may be extended to additional models available in PMML format in the future. This approach generated a new way of creating MLMs based on machine learning results, in addition to the traditional method of knowledge design with clinical experts.