Explainable Prediction of Medical Codes from Clinical Text

Clinical notes are text documents that are created by clinicians for each patient encounter. They are typically accompanied by medical codes, which describe the diagnosis and treatment. Annotating these codes is labor intensive and error prone; furthermore, the connection between the codes and the text is not annotated, obscuring the reasons and details behind specific diagnoses and treatments. We present an attentional convolutional network that predicts medical codes from clinical text. Our method aggregates information across the document using a convolutional neural network, and uses an attention mechanism to select the most relevant segments for each of the thousands of possible codes. The method is accurate, achieving precision@8 of 0.71 and a Micro-F1 of 0.54, which are both better than the prior state of the art. Furthermore, through an interpretability evaluation by a physician, we show that the attention mechanism identifies meaningful explanations for each code assignment

PDF Abstract NAACL 2018 PDF NAACL 2018 Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Medical Code Prediction MIMIC-III CAML Macro-AUC 89.5 # 10
Micro-AUC 98.6 # 8
Macro-F1 8.8 # 9
Micro-F1 53.9 # 10
Precision@8 70.9 # 10
Precision@15 56.1 # 10
Medical Code Prediction MIMIC-III DR-CAML Macro-AUC 89.7 # 9
Micro-AUC 98.5 # 10
Macro-F1 8.6 # 10
Micro-F1 52.9 # 11
Precision@8 69.0 # 11
Precision@15 54.8 # 11
Medical Code Prediction MIMIC-III SVM Micro-F1 44.1 # 12
Medical Code Prediction MIMIC-III Logistic Regression Macro-AUC 56.1 # 14
Micro-AUC 93.7 # 14
Macro-F1 1.1 # 15
Micro-F1 27.2 # 16
Precision@8 54.2 # 15
Precision@15 41.1 # 14
Medical Code Prediction MIMIC-III CNN Macro-AUC 80.6 # 13
Micro-AUC 96.9 # 13
Macro-F1 4.2 # 12
Micro-F1 41.9 # 13
Precision@8 58.1 # 14
Precision@15 44.3 # 13
Medical Code Prediction MIMIC-III Bi-GRU Macro-AUC 82.2 # 12
Micro-AUC 97.1 # 12
Macro-F1 3.8 # 13
Micro-F1 41.7 # 14
Precision@8 58.5 # 13
Precision@15 44.5 # 12

Methods