AUTOMATIC ICD TO IMPROVE DIAGNOSIS CODING ACCURACY: A LITERATURE STUDY

Abstract

ICD coding is usually done by a coder who assigns the ICD code according to the Doctor's clinical diagnosis. However, because coders need to master specific skills, such as knowledge in the field of medicine, coding rules, and medical terminology, manual coding can be costly, time-consuming, and inefficient. Based on this, developing a computationally accurate approach to automatic ICD encoding is imperative. This literature study aims to provide an overview of automatic ICD in terms of the dataset and classification method used. The literature study results show that automatic ICD research to improve the accuracy of diagnosis coding has been done and is still a challenge today. Most of the datasets used are the public MIMIC-III dataset, while automatic ICD classification is the current research trend with deep learning. The deep learning algorithms that are widely used include CNN, RNN, and LSTM. The resulting accuracy based on the dataset and classification method is very diverse. Future research still has many opportunities to contribute and improve the correct classification method to improve automatic ICD accuracy.