Recent Advances in AI for Automated ICD Coding: A Systematic Literature Review

Clinical Snapshot

Key Findings

• P Value: Not applicable (narrative systematic review without statistical pooling)

• Effect Size: Not quantitatively pooled; narrative synthesis indicates models achieve higher accuracy on frequent code subsets compared to full label spaces, with substantial architectural diversity and variable performance across dataset configurations

• Primary Outcome: Performance of AI models for automated ICD code assignment from unstructured clinical text, assessed across multiple architectures and dataset configurations

• Nnt Or Sensitivity: No NNT, sensitivity, or specificity values pooled across studies; individual study performance metrics (F1-score, precision, recall) are described qualitatively with considerable inter-study variation noted

• Confidence Interval: Not reported (no meta-analysis conducted)

Clinical Application

Currently limited by lack of external validation, poor performance on rare codes, interpretability deficits, and absence of standardised benchmarks. Real-world deployment requires substantial further development, regulatory approval, and prospective clinical validation before routine use Australian relevance is limited at present. Australia uses ICD-10-AM (Australian Modification) administered through the Australian Consortium for Classification Development (ACCD), which differs from ICD-10-CM used in the majority of studies reviewed (predominantly MIMIC-III/IV, a US dataset). The Australian Digital Health Agency's National Clinical Terminology Service and IHPA (Independent Hospital Pricing Authority) casemix funding frameworks depend on accurate AR-DRG coding derived from ICD-10-AM, meaning that US-trained models cannot be directly applied without substantial local adaptation. The TGA would need to evaluate any AI-based coding tool as a Software as a Medical Device (SaMD) if it influences clinical decisions. RACGP and ACHI standards for clinical documentation would need to be incorporated into any Australian implementation. The 5P research agenda proposed by the authors is directly relevant to Australian priorities, particularly Population Diversity (including Indigenous health data) and Practical Integration with My Health Record infrastructure. Healthcare institutions seeking to implement or evaluate AI-assisted ICD coding systems for clinical documentation, particularly for high-volume discharge summary and EHR coding workflows

Abstract

International Classification of Diseases (ICD) codes enable correct billing, insurance reimbursement, and healthcare analytics. However, manual coding is time-consuming, expensive, and error-prone, creating bottlenecks in clinical workflow and limiting scalability. Artificial intelligence (AI) has emerged as a promising solution for automated ICD code assignment from unstructured clinical text. This systematic review explores the current state of automated ICD coding research, examining models applied to diverse clinical documents including discharge summaries, electronic health records, nursing notes, and pathology reports. Following PRISMA guidelines, we searched six databases for studies published between 2019 and 2024, selecting 54 relevant studies from 4,280 initial citations. Our analysis reveals the use of diverse datasets, preprocessing techniques, and feature extraction methods, alongside a clear evolution from traditional machine learning to deep learning approaches, with substantial architectural diversity across convolutional, recurrent, transformer, and hybrid models. Performance varies considerably across dataset configurations, with models achieving higher accuracy on frequent code subsets compared to full label spaces. However, critical gaps persist: overreliance on single-language, single-institution datasets limits generalizability; difficulties in predicting rare codes remain unresolved; lack of model interpretability undermines clinical trust; and inconsistent evaluation protocols hinder meaningful comparison. To address these challenges, we propose a 5P evidence-grounded research agenda: Population Diversity, Performance Robustness, Prediction of Rare Codes, Provenance Transparency, and Practical Integration. These findings underscore AI's potential to transform ICD coding while highlighting the need for standardized benchmarks, rigorous external validation, multilingual datasets, and explainable architectures to enable equitable and effective deployment in real-world healthcare systems.

References

1. 1.Khalid, A. R., Ali, H., Fathima, Kounen, S. C., Kouayep, & Kim, H.-C. (2026). Recent advances in AI for automated ICD coding: A systematic literature review. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02429-7