An Autonomous Algorithm for Generating and Merging Clinical Algorithms

Abstract

Procedural knowledge in medicine uses to come expressed as isolated sentences in Clinical Practice Guidelines (CPG) that describe how to act in front of specific health-care situations. Although CPGs gather all evidence available on concrete medical problems, their direct application has been proved to have some limitations. One of these limitations occurs when they have to be applied on co-morbid patients suffering from several simultaneous and mutually related diseases. In such cases, health-care professionals have to follow the indications of multiple CPGs and solve their interactions as they appear in the treatment of concrete patients. Clinical Algorithms (CA) are schematic models of the procedures appearing in a CPG. They are used to organize and summarize the recommendations contained in CPGs. Here, we extend a knowledge-based algorithm to merge CAs with a machine learning procedure to relax the knowledge dependence of that algorithm. The resulting algorithm has been tested on health-care data provided by the SAGESSA Group on hypertension patients. The results obtained prove that it is a good approach to the generation of CA from data though several improvements at the levels of prediction and medical interpretation are possible. Furthermore, the learned knowledge from the data generation process can be reused to improve the results of the merging process for similar diseases.