Clinical Outcome Prediction from Admission Notes using Self-Supervised Knowledge Integration

Outcome prediction from clinical text can prevent doctors from overlooking possible risks and help hospitals to plan capacities. We simulate patients at admission time, when decision support can be especially valuable, and contribute a novel admission to discharge task with four common outcome prediction targets: Diagnoses at discharge, procedures performed, in-hospital mortality and length-of-stay prediction. The ideal system should infer outcomes based on symptoms, pre-conditions and risk factors of a patient. We evaluate the effectiveness of language models to handle this scenario and propose clinical outcome pre-training to integrate knowledge about patient outcomes from multiple public sources. We further present a simple method to incorporate ICD code hierarchy into the models. We show that our approach improves performance on the outcome tasks against several baselines. A detailed analysis reveals further strengths of the model, including transferability, but also weaknesses such as handling of vital values and inconsistencies in the underlying data.

PDF Abstract EACL 2021 PDF EACL 2021 Abstract

Datasets


Introduced in the Paper:

Clinical Admission Notes from MIMIC-III

Used in the Paper:

MIMIC-III MedQuAD

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Mortality Prediction Clinical Admission Notes from MIMIC-III BioBERT Base AUROC 82.55 # 2
Mortality Prediction Clinical Admission Notes from MIMIC-III BERT Base AUROC 81.13 # 3
Medical Diagnosis Clinical Admission Notes from MIMIC-III CORe AUROC 83.54 # 1
Medical Diagnosis Clinical Admission Notes from MIMIC-III BioBERT Base AUROC 82.81 # 2
Medical Diagnosis Clinical Admission Notes from MIMIC-III BERT Base AUROC 82.08 # 3
Length-of-Stay prediction Clinical Admission Notes from MIMIC-III CORe AUROC 72.53 # 1
Length-of-Stay prediction Clinical Admission Notes from MIMIC-III BioBERT Base AUROC 71.59 # 2
Length-of-Stay prediction Clinical Admission Notes from MIMIC-III BERT Base AUROC 70.40 # 3
Medical Procedure Clinical Admission Notes from MIMIC-III CORe AUROC 88.37 # 1
Medical Procedure Clinical Admission Notes from MIMIC-III BioBERT Base AUROC 86.36 # 2
Medical Procedure Clinical Admission Notes from MIMIC-III BERT Base AUROC 85.84 # 3
Mortality Prediction Clinical Admission Notes from MIMIC-III CORe AUROC 84.04 # 1

Methods


No methods listed for this paper. Add relevant methods here