Skip to main content
WorldCist'18 - 6th World Conference on Information Systems and Technologies

Full Program »

Miscoding alerts within hospital datasets: an unsupervised machine learning approach

The appropriate funding of hospital services may depend upon grouping hospital cases into Diagnosis Related Groups (DRGs). DRGs rely on the quality of clinical data held in administrative healthcare databases, mainly proper diagnoses and procedure codes. This work proposes a methodology based on unsupervised machine learning and statistical methods to generate alerts of suspect cases of upcoding and under-coding in healthcare administrative databases. DRG coded hospital cases from an administrative database were split into homogeneous patient subgroups by applying decision tree-based algorithms. The proportion of specific diagnosis and procedure codes were compared within targeted subgroups to identify hospitals with abnormal distributions. Preliminary results indicate that the proposed methodology has the potential of automatically identify suspect upcoding or under-coding cases, as well as other relevant types of discrepancies regarding coding practices. Nevertheless, additional evaluation under the medical perspective must be incorporated into the algorithm.

Julio Souza
Faculty of Medicine of the University of Porto
Portugal

João Vasco Santos
Faculty of Medicine of the University of Porto
Portugal

Fernando Lopes
Faculty of Medicine of the University of Porto
Portugal

João Viana
Faculty of Medicine of the University of Porto
Portugal

Alberto Freitas
Faculty of Medicine of the University of Porto
Portugal

 

Powered by OpenConf®
Copyright ©2002-2017 Zakon Group LLC