Bridging the Version Gap: Multi-version Training Improves ICD Code Prediction, Especially for Rare Codes

2026-05-17

AI Summary

This study addresses the challenges of automatic ICD coding from clinical documents, particularly the heterogeneity across ICD versions and the difficulty in recognizing rare codes, which are exacerbated by the long-tailed label distribution. Existing approaches are typically confined to a single ICD version and struggle with generalization. To overcome these limitations, this work proposes the first cross-version joint training strategy—simultaneously leveraging data from multiple ICD versions such as ICD-9 and ICD-10—to build a version-agnostic coding model that enhances generalization without increasing model complexity. By incorporating an improved label-level attention mechanism to effectively fuse multi-version annotations, the method achieves a 27% improvement in micro F1 score on 18K rare ICD-10 codes and significantly boosts macro-level performance on 8K frequent codes, all while reducing the total number of model parameters.

Abstract

Clinical coding maps clinical documentation to standardized medical codes, an essential yet time-consuming administrative task that could benefit from automation. Current models on ICD coding are typically optimized for codes from a specific ICD version. However, in reality, ICD systems evolve continuously, and different versions are adopted across time periods and regions. Moreover, ICD coding suffers from the long-tail problem, and rare code performance can be a bottleneck for developing implementable models. We examine whether it is viable to train version-independent models by combining data annotated in different ICD versions, which may help address these challenges. We add ICD-9 data to the training of a modified label-wise attention model for ICD-10 prediction, and find that despite the version mismatch, adding ICD-9 yields a 27% increase in micro F1 for 18K rare ICD codes compared to training on ICD-10 alone. On 8K frequent ICD-10 codes, the multi-version training also substantially improves macro metrics, with far fewer model parameters.

Problem

• ICD code prediction
• version gap
• rare codes
• clinical coding
• long-tail problem

Innovation

• multi-version training
• ICD code prediction
• rare code enhancement
• version-independent modeling
• clinical coding automation