Scalable Wide and Deep Learning for Computer Assisted Coding

Abstract

In recent years the use of electronic medical records has accelerated resulting in large volumes of medical data when a patient visits a healthcare facility. As a first step towards reimbursement healthcare institutions need to associate ICD-10 billing codes to these documents. This is done by trained clinical coders who may use a computer assisted solution for shortlisting of codes. In this work, we present our work to build a machine learning based scalable system for predicting ICD-10 codes from electronic medical records. We address data imbalance issues by implementing two system architectures using convolutional neural networks and logistic regression models. We illustrate the pros and cons of those system designs and show that the best performance can be achieved by leveraging the advantages of both using a system combination approach.

A Systematic Literature Review of Automated ICD Coding and Classification Systems using Discharge Summaries

“…Apart from that, we found that the majority of the studies (n=31) have focused on the ICD-9-CM classification system, 3 studies [52,3,99] predicted ICD-10-CM codes, 2 studies [39,40] predicted ICD-10-AM and ACHI codes, and 1 study [86] used ICD-10-PCS, 1 study [95] converted ICD-9-CM codes to ICD-10-CM codes an using online resource before assigning them to discharge summaries. This shows that there is a scarcity of studies relevant to other classification systems including Australian classification systems.…”

“…Table 10 shows machine learning and deep learning models that were employed for assigning ICD codes to discharge summaries. Notably, in several studies (Perotte et al [65], Marafino et al [55], Subotin and Davis [86], Ayyar and Oliver [4], Berndorfer and Henriksson [8], Amoia et al [3], Catling et al [13], Baumel et al [5], Kaur and Ginige [39,40], Xu et al [95], Moons et al [58]) authors did not compare their proposed model with any existing study or algorithm; therefore, the third column value is left empty. A brief overview and comparison of studies is presented in the Section 5.…”

“…To summarise, the majority of studies have evaluated the performance of their model using microaveraged F1-score, followed by macro-averaged F1-score, and standard F1-score. [65] Flat SVM, Hierarchy-based SVM -Marafino et al [55] SVM -Subotin and Davis [86] Two-level hierarchical classification -Kavuluru et al [43] SVM, LR, MNB BR, copy transformation, ECC Ayyar and Oliver [4] LSTM -Prakash et al [67] C-MemNN and A-MemNN End-to-End Memory Network, KV-MemNNs Lin et al [52] CNN SVM, RF, GBM Berndorfer and Henriksson [8] Flat SVM and Hierarchical SVM -Amoia et al [3] LR and CNN -Catling et al [13] RNN-GRU -Baumel et al [5] SVM, CBOW, CNN, HA-GRU -Mullenbach et al [61] CAML, DR-CAML CNN, LR, Bi-GRU, Flat SVM [65], HA-GRU [5], C-MemNN [67], [80],…”

Medical code prediction with multi-view convolution and description-regularized label-dependent attention

“…There has been significant work towards the automated coding problem (Perotte et al, 2013;Kavuluru et al, 2015;Wang et al, 2016;Scheurwegs et al, 2017;Prakash et al, 2017;Rajkomar et al, 2018;Amoia et al, 2018). We review some of the recent relevant work.…”

Generating Accurate Electronic Health Assessment from Medical Graph

“…Earlier expert systems were mainly knowledgebased, typically using decision rules. Later, machine learning approaches were developed, mainly used longitudinal electronic health records (EHR) to predict ICD codes (Subotin and Davis, 2014;Amoia et al, 2018), the diagnostic codes assigned to EHRs after each patient's visit or encounter. However, ICD codes are used mainly for billing purposes and have limitations (e.g., incomplete assignment) when used as the gold standard diagnoses labels (O'malley et al, 2005).…”