@article{fdi:010089920,
  title = {{C}reating a computer assisted {ICD} coding system : performance metric choice and use of the {ICD} hierarchy},
  author = {{M}arcou, {Q}. and {B}erti-{E}quille, {L}aure and {N}ovelli, {N}.},
  editor = {},
  language = {{ENG}},
  abstract = {{O}bjective : {M}achine learning methods hold the promise of leveraging available data and generating higher-quality data while alleviating the data collection burden on healthcare professionals. {I}nternational {C}lassification of {D}iseases ({ICD}) diagnoses data, collected globally for billing and epidemiological purposes, represents a valuable source of structured information. {H}owever, {ICD} coding is a challenging task. {W}hile numerous previous studies reported promising results in automatic {ICD} classification, they often describe input data specific model architectures, that are heterogeneously evaluated with different performance metrics and {ICD} code subsets.{T}his study aims to explore the evaluation and construction of more effective {C}omputer {A}ssisted {C}oding ({CAC}) systems using generic approaches, focusing on the use of {ICD} hierarchy, medication data and a feed forward neural network architecture. {M}ethods : {W}e conduct comprehensive experiments using the {MIMIC}-{III} clinical database, mapped to the {OMOP} data model. {O}ur evaluations encompass various performance metrics, alongside investigations into multitask, hierarchical, and imbalanced learning for neural networks. {R}esults: {W}e introduce a novel metric, , tailored to the {ICD} coding task, which offers interpretable insights for healthcare informatics practitioners, aiding them in assessing the quality of assisted coding systems. {O}ur findings highlight that selectively cherry-picking {ICD} codes diminish retrieval performance without performance improvement over the selected subset. {W}e show that optimizing for metrics such as {NDCG} and {AUPRC} outperforms traditional {F}1-based metrics in ranking performance. {W}e observe that {N}eural {N}etwork training on different {ICD} levels simultaneously offers minor benefits for ranking and significant runtime gains. {H}owever, our models do not derive benefits from hierarchical or class imbalance correction techniques for {ICD} code retrieval. {C}onclusion : {T}his study offers valuable insights for researchers and healthcare practitioners interested in developing and evaluating {CAC} systems. {U}sing a straightforward sequential neural network model, we confirm that medical prescriptions are a rich data source for {CAC} systems, providing competitive retrieval capabilities for a fraction of the computational load compared to text-based models. {O}ur study underscores the importance of metric selection and challenges existing practices related to {ICD} code sub-setting for model training and evaluation.},
  keywords = {},
  booktitle = {},
  journal = {{J}ournal of {B}iomedical {I}nformatics},
  volume = {152},
  numero = {},
  pages = {104617 [10 ]},
  ISSN = {1532-0464},
  year = {2024},
  DOI = {10.1016/j.jbi.2024.104617},
  URL = {https://www.documentation.ird.fr/hor/fdi:010089920},
}