An Empirical Evaluation of Deep Learning for ICD-9 Code Assignment using MIMIC-III Clinical Notes

Code assignment is important on many levels in the modern hospital, from ensuring accurate billing process to creating a valid record of patient care history. However, the coding process is tedious, subjective, and requires medical coders with extensive training. The objective of this study is to evaluate the performance of deep learning based systems to automatically map clinical notes to medical codes. We applied the state-of-the-art deep learning methods such as Recurrent Neural Networks and Convolution Neural Networks on MIMIC-III dataset. Experiments show that the deep-learning-based methods outperform other conventional machine learning methods. Our evaluations are focused on end-to-end learning methods without manually defined rules. From our evaluations, the best models are able to predict the top 10 ICD-9 codes with 69.57% F1 and 89.67% accuracy; the top 10 ICD-9 categories with 72.33% F1 and 85.88% accuracy.