Statistical Selector of the Best Multiple ICD-coding Method

The International Classification of Diseases 10th version (ICD-10) is one of the standard and most important disease classifications. Since computerized ICD-10 coding systems have drawn a great deal of attention in the medical field, a great number of different coding systems have been proposed. This paper proposes a hybrid architecture of different coding systems. First, given an input disease name, three coding systems output codes with their confidence scores. A C4.5-based system selector then selects the best output by using both input statistics and the confidence score from each system. The experimental results demonstrated that the selector significantly boosts the overall performance (+3.4 points).