Explainable Prediction of Medical Codes from Clinical Text
By James Mullenbach, Sarah Wiegreffe, Jon D. Duke, Jimeng Sun, Jacob Eisenstein. 2018
692 citations recorded by the source corpus.
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.
Subfields: Artificial Intelligence, Molecular Biology, Information Extraction, Medical & Health AI
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Chunyuan Li
- Lawrence Carin
- Noah A. Smith
- Jianfeng Gao
- Dinghan Shen
- Ricardo Henao
- Yizhe Zhang
- Mohit Bansal
In Molecular Biology
- Ilias Chalkidis
- Ion Androutsopoulos
- Jimeng Sun
- Edward Choi
- Cao Xiao
- Maria de la Iglesia-Vayá
- Jose María Salinas
- Xiang Dai
In Information Extraction
- Ilias Chalkidis
- Ion Androutsopoulos
- Noah A. Smith
- Prodromos Malakasiotis
- Nikolaos Aletras
- Sarah Wiegreffe
- Greg Durrett
- Daniel Klein
In Medical & Health AI
- Cao Xiao
- Ilias Chalkidis
- Jimeng Sun
- Edward Choi
- Ion Androutsopoulos
- Maria de la Iglesia-Vayá
- Jose María Salinas
- Aurelia Bustos
Citations in the corpus
- Attention is not Explanation (2019, neutral): “, (Xu et al., 2015; Choi et al., 2016; Lei et al., 2017; Martins and Astudillo, 2016; Xie et al., 2017; Mullenbach et al., 2018).”
- Attention is not not Explanation (2019, background): “Attention mechanisms (Bahdanau et al., 2014) are nowadays ubiquitous in NLP, and their suitability for providing explanations for model predictions is a topic of high interest (Xu et al., 2015; Rocktäschel et al., 2015; Mullenbach et al., 2018; Thorne et al., 2019; Serrano and Smith, 2019).”
- Deep Weighted Averaging Classifiers (2018, methodology): “For text datasets, we use a shallow convolutional model with attention [34].”
- Joint Embedding of Words and Labels for Text Classification (2018, methodology): “To compare with previous work, we follow (Shi et al., 2017; Mullenbach et al., 2018), and preprocess a dataset consisting of the most common 50 labels.”
- A Survey of the State of Explainable AI for Natural Language Processing (2020, methodology): “The approaches found in this survey vary in several dimensions, including the number of humans involved (ranging from 1 (Mullenbach et al., 2018) to 25 (Sydorova et al., 2019) humans), as well as the high-level task that they were asked to perform (including rating the explanations of a single…”
- PadChest: A large chest x-ray image dataset with multi-label annotated reports (2019, methodology): “where u` is a vector parameter for label ` and SoftMax(x) = exp(x) ∑ ∀i exp(xi) , where exp(x) is the element-wise exponentiation of the vector x as described in [34].”
- Opportunities and challenges in developing deep learning models using electronic health records data: a systematic review (2018, methodology): “18,20 Unsupervised embedding Several other unsupervised learning methods besides AEs have been applied to EHR concept representations.”
- Cross-Domain Transfer of Generative Explanations Using Text-to-Text Models (2021, background): “In the medical domain specifically, 1-dimensional CNNs with label-conditional attention have been explored for explainable ICD code prediction from discharge summaries [31].”
- Neural Legal Judgment Prediction in English (2019, methodology): “The Label-Wise Attention Network (Mullenbach et al., 2018) has been shown to be robust in multi-label classification.”
- Why Attention is Not Explanation: Surgical Intervention and Causal Reasoning about Neural Models (2020, background): “This approach resulted in a variety of visualizations and other attempts at model explanation being explored based, in part or in whole, on attention weights (Xu et al., 2015a; Mullenbach et al., 2018; Yang et al., 2019).”
- Cross-Loss Influence Functions to Explain Deep Network Representations (2020, background): “…mechanisms have been proposed, including interpretable neural network approaches (Wang et al., 2016), natural language generation of explanations (Mullenbach et al., 2018; Wiegreffe and Pinter, 2019; Chen et al., 2021), and many others Hoffman et al. (2018); Holzinger et al. (2020); Linardatos et…”
- Revisiting Transformer-based Models for Long Document Classification (2022, methodology): “Therefore, in this paper, we follow previous work (Mullenbach et al., 2018; Chalkidis et al., 2022) and consider a subset of frequent labels in MIMIC-III and ECtHR. numeric-only tokens in Mullenbach et al. (2018).”
- Large-Scale Multi-Label Text Classification on EU Legislation (2019, methodology): “As a second contribution, we experiment with several neural classifiers on EURLEX57K, including the Label-Wise Attention Network of Mullenbach et al. (2018), called CNN-LWAN here, which was reported to achieve state of the art performance in LMTC on medical records.”
- Federated Learning for ICD Classification with Lightweight Models and Pretrained Embeddings (2025, unknown): “The research focus was then shifted to more sophisticated techniques, like attention mechanisms (e.g., CAML, DR-CAML) and label-aware embeddings (e.g., LEAM), which improved the processing of multi-label outputs and rare codes [4,7].”