Medical Coding with Biomedical Transformer Ensembles and Zero/Few-shot Learning

TL;DR

This paper introduces xTARS, a novel ensemble method combining BERT and zero/few-shot learning for medical coding, achieving improved accuracy especially with limited training data, and is deployed in a real-world setting.

Contribution

The paper presents xTARS, a new ensemble approach that enhances medical coding accuracy by integrating BERT with zero/few-shot learning techniques, addressing data scarcity and large label space.

Findings

Outperforms strong baselines in few-shot scenarios

Effective in real-world deployment at Bayer since 2021

Reproducible code released for research community

Abstract

Medical coding (MC) is an essential pre-requisite for reliable data retrieval and reporting. Given a free-text reported term (RT) such as "pain of right thigh to the knee", the task is to identify the matching lowest-level term (LLT) - in this case "unilateral leg pain" - from a very large and continuously growing repository of standardized medical terms. However, automating this task is challenging due to a large number of LLT codes (as of writing over 80,000), limited availability of training data for long tail/emerging classes, and the general high accuracy demands of the medical domain. With this paper, we introduce the MC task, discuss its challenges, and present a novel approach called xTARS that combines traditional BERT-based classification with a recent zero/few-shot learning approach (TARS). We present extensive experiments that show that our combined approach outperforms strong baselines, especially in the few-shot regime. The approach is developed and deployed at Bayer, live since November 2021. As we believe our approach potentially promising beyond MC, and to ensure reproducibility, we release the code to the research community.