International Classification Of Diseases Coding Research Articles

The Association Between Metformin Use and New-Onset ICD Coding of Geographic Atrophy.

Metformin has been suggested to protect against the development of age-related macular degeneration (AMD) in multiple observational studies. However, the association between metformin and geographic atrophy (GA), a debilitating subtype of AMD, has not been analyzed. We conducted a case-control study of patients ages 60years and older with new-onset International Classification of Diseases (ICD) coding of GA in the Merative MarketScan Commercial and Medicare Databases between 2017 and 2021. Cases were matched with propensity scores estimated by age, region, hypertension, and Charlson Comorbidity Index to a control without GA of the same year. Exposure to metformin was assessed for cases and controls in the year prior to their index visit. Conditional multivariable logistic regression, adjusting for AMD risk factors, was used to calculate odd ratios and 95% confidence intervals (CIs). This study design and analysis were repeated in a sample of patients without diabetes. In the full sample, we identified 10,505 cases with GA and 10,502 matched controls without GA. In total, 1149 (10.9%) cases and 1277 (12.2%) controls were exposed to metformin, and in multivariable regression, metformin decreased the odds of new-onset ICD coding of GA by 12% (95% CI, 0.79-0.99). In the sample of patients without diabetes, we identified 7611 cases with GA and 7608 matched controls without GA. Twenty-nine (0.4%) cases and 63 (0.8%) controls were exposed to metformin, and in multivariable regression, metformin decreased the odds of new-onset ICD coding of GA by 47% (95% CI, 0.33-0.83). Metformin may hold promise as a noninvasive, alternative agent to prevent the development of GA. This finding is notable due to shortcomings in recently approved therapeutics for GA and metformin's overall ease of use and few adverse effects. Additional studies are required to explore our findings further and motivate a clinical trial.

Creating a computer assisted ICD coding system: Performance metric choice and use of the ICD hierarchy

Objective:Machine learning methods hold the promise of leveraging available data and generating higher-quality data while alleviating the data collection burden on healthcare professionals. International Classification of Diseases (ICD) diagnoses data, collected globally for billing and epidemiological purposes, represents a valuable source of structured information. However, ICD coding is a challenging task. While 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.This study aims to explore the evaluation and construction of more effective Computer Assisted Coding (CAC) systems using generic approaches, focusing on the use of ICD hierarchy, medication data and a feed forward neural network architecture. Methods:We conduct comprehensive experiments using the MIMIC-III clinical database, mapped to the OMOP data model. Our evaluations encompass various performance metrics, alongside investigations into multitask, hierarchical, and imbalanced learning for neural networks. Results:We 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. Our findings highlight that selectively cherry-picking ICD codes diminish retrieval performance without performance improvement over the selected subset. We show that optimizing for metrics such as NDCG and AUPRC outperforms traditional F1-based metrics in ranking performance. We observe that Neural Network training on different ICD levels simultaneously offers minor benefits for ranking and significant runtime gains. However, our models do not derive benefits from hierarchical or class imbalance correction techniques for ICD code retrieval. Conclusion:This study offers valuable insights for researchers and healthcare practitioners interested in developing and evaluating CAC systems. Using 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. Our study underscores the importance of metric selection and challenges existing practices related to ICD code sub-setting for model training and evaluation.

Modelling long medical documents and code associations for explainable automatic ICD coding

Quick and accurate International Classification of Diseases (ICD) code assignment is vital for billing, reimbursement and medical research. Owing to the labour-intensive and error-prone nature of manual coding, automatic ICD coding using deep learning methods has flourished. However, this task is still challenging because of (1) the interpretability of coding, (2) lengthy clinical documents and (3) long-tail label distribution. In the current study, we propose a novel automatic ICD coding framework to address these issues. First, a biomedical-specific pre-trained language model, Clinical-Longformer, is used as an encoder, which generates meaningful representations of long clinical documents by injecting rich medical knowledge and capturing long-distance dependence among tokens. Second, a decoding architecture that combines the multi-synonym attention mechanism, hierarchical curriculum learning and distribution-balance loss is designed to perform ICD code prediction. The decoder improves the tail-end performance by fully capturing code associations in terms of semantics, structure and co-occurrence. In addition, the label-wise attention mechanism provides the interpretability of prediction. Experimental results on benchmark MIMIC-III datasets indicate that our model achieves higher F1 scores than previous state-of-the-art baselines. Our suggested model is expected to serve as an aid in improving the efficiency of manual ICD coding and to offer insights for other long text classification tasks with multiple label associations.

Evaluating the coding accuracy of type 2 diabetes mellitus among patients with non-alcoholic fatty liver disease

BackgroundNon-alcoholic fatty liver disease (NAFLD) describes a spectrum of chronic fattening of liver that can lead to fibrosis and cirrhosis. Diabetes has been identified as a major comorbidity that contributes to NAFLD progression. Health systems around the world make use of administrative data to conduct population-based prevalence studies. To that end, we sought to assess the accuracy of diabetes International Classification of Diseases (ICD) coding in administrative databases among a cohort of confirmed NAFLD patients in Calgary, Alberta, Canada.MethodsThe Calgary NAFLD Pathway Database was linked to the following databases: Physician Claims, Discharge Database, National Ambulatory Care Reporting System, Pharmaceutical Information Network database, Laboratory, and Electronic Medical Records. Hemoglobin A1c and diabetes medication details were used to classify diabetes groups into absent, prediabetes, meeting glycemic targets, and not meeting glycemic targets. The performance of ICD codes among these groups was compared to this standard. Within each group, the total numbers of true positives, false positives, false negatives, and true negatives were calculated. Descriptive statistics and bivariate analysis were conducted on identified covariates, including demographics and types of interacted physicians.ResultsA total of 12,012 NAFLD patients were registered through the Calgary NAFLD Pathway Database and 100% were successfully linked to the administrative databases. Overall, diabetes coding showed a sensitivity of 0.81 and a positive predictive value of 0.87. False negative rates in the absent and not meeting glycemic control groups were 4.5% and 6.4%, respectively, whereas the meeting glycemic control group had a 42.2% coding error. Visits to primary and outpatient services were associated with most encounters.ConclusionDiabetes ICD coding in administrative databases can accurately detect true diabetic cases. However, patients with diabetes who meets glycemic control targets are less likely to be coded in administrative databases. A detailed understanding of the clinical context will require additional data linkage from primary care settings.

Associations Between Autoimmune Disease and the Development of Age-Related Macular Degeneration.

The pathogenesis of age-related macular degeneration (AMD) likely implicates the dysregulation of immune response pathways. Several studies demonstrate that the pathogenic elements of AMD resemble those of autoimmune diseases, yet the association between AMD development and most autoimmune diseases remain unexplored. We conducted a case-control analysis of patients ages 55 and older with new-onset International Classification of Diseases (ICD) coding of dry, wet, or unspecified AMD between 2005 and 2019 in the Merative MarketScan Commercial and Medicare Databases. The diagnosis of an autoimmune disease was defined by an outpatient or inpatient claim with a relevant ICD code in the 12months before the index visit. Conditional multivariable logistic regression, adjusted for AMD risk factors, was used to calculate odd ratios and 95% confidence intervals. We identified 415,027 cases with new-onset ICD coding for AMD matched with propensity scores to 414,853 controls. In total, 16.1% of cases and 15.9% of controls were diagnosed with any autoimmune disease. The diagnosis of any autoimmune disease did not affect the odds of new-onset ICD coding for AMD in multivariable regression (OR = 1.01; 95% CI, 0.999-1.02). Discoid lupus erythematosus (OR = 1.29; 95% CI, 1.12-1.48), systemic lupus erythematosus (SLE) (OR = 1.21; 95% CI, 1.15-1.27), giant cell arteritis (OR = 1.19; 95% CI, 1.09-1.30), Sjogren's syndrome (OR = 1.17; 95% CI, 1.09-1.26), and Crohn's disease (OR = 1.13; 95% CI, 1.06-1.22) increased the odds of a new-onset ICD coding for AMD. Most autoimmune diseases do not affect the odds of developing AMD but several common autoimmune disorders such as SLE and Crohn's disease were associated with modestly increased odds of AMD. Further studies are needed to validate and investigate the underlying mechanisms of these associations.

Automatic ICD-10-CM coding via Lambda-Scaled attention based deep learning model

The International Classification of Diseases (ICD) serves as a global healthcare administration standard, with one of its editions being ICD-10-CM, an enhanced diagnostic classification system featuring numerous new codes for specific anatomic sites, co-morbidities, and causes. These additions facilitate conveying the complexities of various diseases. Currently, ICD-10 coding is widely adopted worldwide. However, public hospitals in Pakistan have yet to implement it and automate the coding process. In this research, we implemented ICD-10-CM coding for a private database and named it Clinical Pool of Liver Transplant (CPLT). Additionally, we proposed a novel deep learning model called Deep Recurrent-Convolution Neural Network with a lambda-scaled Attention module (DRCNN-ATT) using the CPLT database to achieve automatic ICD-10-CM coding. DRCNN-ATT combines a bi-directional long short-term memory network (bi-LSTM), a multi-scale convolutional neural network (MS-CNN), and a lambda-scaled attention module. Experimental results demonstrate that deep recurrent convolutional neural network (DRCNN) faces attention score vanishing problem with a standard attention module for automatic ICD coding. However, adding a lambda-scaled attention module resolves this issue. We evaluated DRCNN-ATT model using two distinct datasets: a private CPLT dataset and a public MIMIC III top 50 dataset. The results indicate that the DRCNN-ATT model outperformed various baselines by generating 0.862 micro F1 and 0.25 macro F1 scores on CPLT dataset and 0.705 micro F1 and 0.655 macro F1 scores on MIMIC III top 50 dataset. Furthermore, we also deployed our model for automatic ICD-10-CM coding using ngrok and the Flask APIs, which receives input, processes it, and then returns the results.

Hyperbolic graph convolutional neural network with contrastive learning for automated ICD coding

The International Classification of Diseases (ICD) is a widely used criterion for disease classification, health monitoring, and medical data analysis. Deep learning-based automated ICD coding has gained attention due to the time-consuming and costly nature of manual coding. The main challenges of automated ICD coding include imbalanced label distribution, code hierarchy and noisy texts. Recent works have considered using code hierarchy or description for better label representation to solve the problem of imbalanced label distribution. However, these methods are still ineffective and redundant since they only interact with a constant label representation. In this work, we introduce a novel Hyperbolic Graph Convolutional Network with Contrastive Learning (HGCN-CL) to solve the above problems and the shortcomings of the previous methods. We adopt a Hyperbolic graph convolutional network on ICD coding to capture the hierarchical structure of codes, which can solve the problem of large distortions when embedding hierarchical structure with graph convolutional network. Besides, we introduce contrastive learning for automatic ICD coding by injecting code features into text encoder to generate hierarchical-aware positive samples to solve the problem of interacting with constant code features. We conduct experiments on the public MIMIC-III and MIMIC-II datasets. The results on MIMIC III show that HGCN-CL outperforms previous state-of-art methods for automatic ICD coding, which achieves a 2.7% and 3.6% improvement respectively compared to previous best results (Hypercore). We also provide ablation experiments and hierarchy visualization to verify the effectiveness of components in our model.

Designing NLP applications to support ICD coding: an impact analysis and guidelines to enhance baseline performance when processing patient discharge notes

Financial costs are a major concern in the healthcare system, with medical billing and coding playing a key role in facilitating transactions and financing procedures. Billing involves filing claims with insurance companies and requires scrutiny of clinical summaries and electronic health records to correctly match diagnoses, prescriptions, and procedures to standardized codes. Accuracy in assigning International Classification of Diseases (ICD) codes is critical to proper reimbursement of care. Incorrect codes waste time and resources, and cause administrative and financial problems for hospitals, insurance companies and patients. Manual medical coding is a labor-intensive and error-prone process that creates additional administrative burden and inconvenience for hospitals, insurance companies, and patients. To simplify the process, clinical records are often processed to automatically identify and extract clinical concepts and corresponding ICD codes. Deep learning and natural language processing techniques have shown promise in a variety of tasks but applying them to medical coding has been challenging. Accurate coding requires a deep understanding of medical terminology, context, and guidelines that may be difficult to capture with traditional deep learning methods. Although deep learning shows promise in healthcare, its specific impact on ICD coding is not fully understood, and translating scalable deep learning methods into practical improvements in ICD coding remains a challenge. Evaluating deep learning models under the scenarios of real-world coding and comparing them to established practice is critical to determining their true effectiveness. In this work, we address the automation of ICD coding by highlighting pitfalls and contrasting different perspectives. We investigated automatic ICD coding using baseline machine learning models, with a focus on identifying ICD-9 codes in discharge notes from Medical Information Mart for Intensive Care (MIMIC) database. A thorough evaluation of different models and approaches is crucial to avoid over-reliance on any method. Our findings show that simpler methods can achieve comparable results to deep learning models while still requiring fewer computational resources.

Combining transformer-based model and GCN to predict ICD codes from clinical records

Automatic International Classification of Diseases (ICD) coding is a method of automatically classifying diseases through a computer program based on rules of etiology and clinical presentation, and representing them through codes, which are widely used to assist in medical reimbursement and reporting of patient health status. With the application of machine learning and deep learning, the accuracy of automatic ICD coding methods has improved considerably. However, this has been accompanied by problems such as insufficient pre-training of text in the models and increased computational complexity along with improved prediction accuracy. In this work we propose an approach called TF-GCN to counter this problem. Firstly, a more accurate and concise feature representation is obtained by feature extraction of both clinical records and ICD codes through the transformer-based model. Secondly, the node features, document features, and relationships between them in the obtained clinical records are input to the GCN for training. Next, a pseudo labeling attention mechanism is added to eliminate the noise generated in the feature extraction process. Finally, the features of the clinical records are compared with the features of the ICD codes for similarity to obtain the classification results. This can not only reduce computational redundancy, but also obtain more accurate classification features. In the real-world MIMIC-III dataset, we compare the proposed algorithm with 11 automatic ICD coding methods to validate the performance of TF-GCN. According to experimental findings, our suggested strategy outperforms the standard evaluation metrics Mif (0.589), MiAUC (0.989), and P@8 (0.758).

Few-shot ICD coding with knowledge transfer and evidence representation

The task of automatic ICD (International Classification of Diseases) coding involves allocating appropriate ICD codes to electronic health records. Due to the long-tailed distribution of ICD codes, current methods perform poorly on rare diseases, also known as few-shot codes. Consequently, we resort to transfer learning as a solution to address the challenge of ICD coding with limited instances. In our paper, we examine the opportunities in few-shot ICD coding and propose a new solution, the evidence-representation-based meta-network (EPEN). Our model has two key innovations: (i) we design evidence representation for diseases based on the observation that the same disease can have different symptoms among individuals, and (ii) we construct a meta-network to memorize category knowledge from common diseases and apply it to rare diseases. Many experiments show that our EPEN solution performs better than the previous methods for both frequently occurring ICD codes and infrequently occurring ICD codes (few-shot codes). Furthermore, EPEN exhibits improved stability in performance, as evidenced by an improvement in both the mean and range of the F1-score.

Artificial Intelligence Algorithm with ICD Coding Technology Guided by Embedded Electronic Medical Record System in Medical Record Information Management

ObjectiveThe study aims to discuss the coding technology of international classification of diseases (ICD) and the application of embedded electronic medical record (EMR) system through clinical diagnosis selection and medical record information management by Artificial Intelligence Algorithm. MethodsAccording to the embedded medical record input interface, the embedded medical record information knowledge system is established, and the data are randomly extracted from the knowledge base. The data are randomly extracted from the EMR database to collect patient medical records and disease diagnosis code data on the first page of 8 clinical departments of endocrinology, oncology, obstetrics and gynecology, ophthalmology, burns, orthopedics, neurosurgery and cardiovascular medicine for statistical analysis. The natural language processing-bidirectional recurrent neural network (NLP-BIRNN) algorithm is used to optimize the medical records. ResultsThe coders are unfamiliar with the basic rule of the main diagnosis selection, and are unclear about the classification of disease codes. They do not code according to the main diagnosis principles. When there is a complication, the previous disease is mistaken as the main diagnosis. The coding of many kinds of diseases is not detailed and accurate. There is no merge coding for diseases that should be merged. The undiagnosed diseases are coded according to the diagnosed diseases. The diseases are not coded according to different conditions. The diseases are not coded according to the specific classification. The diseases are not coded according to cause. Postoperative complications codes are inaccurate. The filling of the disease diagnosis is incomplete. The name of the disease diagnosis site is inaccurate. Coding selection is too general. The solutions to these problems are as follows. Coders and medical staff should strengthen communication. The training of coders and medical staff in this aspect of knowledge should be strengthened. The corresponding incentive measures should be formulated to increase the importance of coders and medical staff on ICD coding technology. BIRNN is compared with convolutional neural network (CNN) and recurrent neural network (RNN) in accuracy, symptom precision, and symptom recall, and BIRNN proposed has higher values. Conclusionthe errors of coding staff and medical staff are reduced in the main diagnosis and coding, and the pathological language reading under artificial intelligence algorithm provides certain convenience for the diagnosis and treatment of diseases.

Use of clinical classifications software to address ICD coding transition in large healthcare databases analyzed via high-dimensional propensity scores.

The EUPAS26595 study characterized the rate of acute renal failure (ARF) in patients exposed to levetiracetam versus other antiepileptic drugs using healthcare claims data and a high-dimensional propensity score (hd-PS) for confounding adjustment. The data contained several coding systems by design and an update in International Classification of Diseases (ICD) coding dictionary. Such coding heterogeneity can affect the performance of hd-PS, and manually coding harmonization is not feasible. Our objective was to explore the impact of code aggregation via Clinical Classifications Software (CCS) on the analysis of a large claims-based database using hd-PS. Patients with epilepsy, who were new-users of an antiepileptic drug, were identified from the IBM® MarketScan® Research Databases. We used CCS categories to harmonize coding and compared the results with other alternatives. Incidence rate ratios (IRRs) were computed using modified Poisson regression model with a robust variance estimator. For January 2008-October 2015 (before ICD update), 34 833 eligible patients initiated levetiracetam and 52 649 initiated a comparator drug; IRR (95% CI) for ARF for the hd-PS analysis was 1.34 (0.72-2.50) without CCS categories and 1.30 (0.71-2.39) with CCS categories. For January 2008-December 2017 (including ICD coding change), 45 672 eligible patients initiated levetiracetam and 64 664 initiated a comparator drug; IRR (95% CI) for the hd-PS analysis was 1.34 (0.78-2.29) without CCS categories and 1.37 (0.80-2.34) with CCS categories. Using single-level CCS categories to overcome differences in coding provides consistent results and can be used in studies that use large claims data and hd-PS for adjustment.

Automatic International Classification of Diseases Coding via Note-Code Interaction Network with Denoising Mechanism.

Clinical notes are comprehensive files containing explicit information about a patient's visit. However, accurately assigning medical codes from clinical documents can be a persistent challenge due to the complexity of clinical data and the vast range of medical codes. Moreover, the large volume of medical records, the noisy medical records, and the uneven quality of coders all negatively impact the quality of the final codes. Deep learning technology has recently been integrated into automatic International Classification of Diseases (ICD) coding tasks to improve accuracy. Nevertheless, the imbalanced class problem, the complexness of code associations, and the noise in lengthy records still restrict the advancement of ICD coding tasks in deep learning. Thus, we present the Note-code Interaction Denoising Network (NIDN) that employs the self-attention mechanism to pull critical semantic features in electronic medical records (EMRs). Our model utilizes the label attention mechanism for retaining code-specific text expression. We introduce Clinical Classifications Software coding for multitask learning, capturing the functional relationships of medical coding to oblige in model prediction. To minimize the impact of noise on model prediction and improve the label distribution imbalance, a denoising module is introduced to filter noise. Our practical consequences indicate that the model NIDN exceeds competitive models on a third version of Medical Information Mart for Intensive Care data set.

Multigranularity Label Prediction Model for Automatic International Classification of Diseases Coding in Clinical Text.

International Classification of Diseases (ICD) serves as the foundation for generating comparable global disease statistics across regions and over time. The process of ICD coding involves assigning codes to diseases based on clinical notes, which can describe a patient's condition in a standard way. However, this process is complicated by the vast number of codes and the intricate taxonomy of ICD codes, which are hierarchically organized into various levels, including chapter, category, subcategory, and its subdivisions. Many existing studies focus solely on predicting subcategory codes, ignoring the hierarchical relationships among codes. To address this limitation, we propose a multitask learning model that trains multiple classifiers for different code levels, while also capturing the relations between coarser and finer-grained labels through a reinforcement mechanism. Our approach is evaluated on both English and Chinese benchmark dataset, and we demonstrate that our method achieves competitive performance with baseline models, particularly in terms of macro-F1 results. These findings suggest that our approach effectively leverages the hierarchical structure of ICD codes to improve disease code prediction accuracy. Analysis of attention mechanism shows that multigranularity attention of our model captures crucial feature of input text on different granularity levels, which can provide reasonable explanations for the prediction results.

Multi-label Few-shot ICD Coding as Autoregressive Generation with Prompt.

Automatic International Classification of Diseases (ICD) coding aims to assign multiple ICD codes to a medical note with an average of 3,000+ tokens. This task is challenging due to the high-dimensional space of multi-label assignment (155,000+ ICD code candidates) and the long-tail challenge - Many ICD codes are infrequently assigned yet infrequent ICD codes are important clinically. This study addresses the long-tail challenge by transforming this multi-label classification task into an autoregressive generation task. Specifically, we first introduce a novel pretraining objective to generate free text diagnoses and procedures using the SOAP structure, the medical logic physicians use for note documentation. Second, instead of directly predicting the high dimensional space of ICD codes, our model generates the lower dimension of text descriptions, which then infers ICD codes. Third, we designed a novel prompt template for multi-label classification. We evaluate our Generation with Prompt (GPsoap) model with the benchmark of all code assignment (MIMIC-III-full) and few shot ICD code assignment evaluation benchmark (MIMIC-III-few). Experiments on MIMIC-III-few show that our model performs with a marco F130.2, which substantially outperforms the previous MIMIC-III-full SOTA model (marco F1 4.3) and the model specifically designed for few/zero shot setting (marco F1 18.7). Finally, we design a novel ensemble learner, a cross-attention reranker with prompts, to integrate previous SOTA and our best few-shot coding predictions. Experiments on MIMIC-III-full show that our ensemble learner substantially improves both macro and micro F1, from 10.4 to 14.6 and from 58.2 to 59.1, respectively.

Artificial intelligence for prediction of International Classification of Disease codes

Background: The automatic coding of electronic medical records with ICD (International Classification of Diseases) codes is an area of interest due to its potential in improving efficiency and streamlining processes such as billing and outcome tracking. artificial intelligence (AI), and particularly convolutional neural networks (CNN), have been suggested as a possible mechanism for automatic coding. To this end, a rapid review has been undertaken in order to assess the current use of CNN in predicting ICD codes from electronic medical records. Methods: After screening PubMed, IEEE Xplore, Scopus, and Google Scholar, 11 studies were analyzed for the use of CNN in predicting ICD codes. We used artificial intelligence and ICD prediction as keywords in the search strategy. Results: The analysis yielded a recommendation to further explore and research CNN frameworks as a promising lead to automatic ICD coding when paired with word embedding and/or neural transfer learning, while keeping research open to a wide variety of AI techniques. Conclusion: CNN frameworks are promising for the prediction of ICD codes from clinical notes. Bangabandhu Sheikh Mujib Medical University Journal 2023;16(2): 118-123

Using EHR data and machine learning approach to facilitate the identification of patients with lung cancer from a pan-cancer cohort.

e13552 Background: Real-world data from Electronic Health Records (EHR) have been widely used for patient identification to build study cohorts for clinical research. Traditionally, diagnosis codes in the EHR, such International Classification of Diseases (ICD), are used to identify the target patients. However, the accuracy of this approach is dependent on the accuracy of ICD coding, with potential errors especially for tumor types that are frequent locations for metastases (which may contribute to mis-coding). In this study, we attempted to develop a Machine Learning (ML) based approach on EHR data to improve the accuracy of identification of patients with lung cancer. Methods: We used survey respondents in the Enhanced, EHR-facilitated Cancer Symptom Control (E2C2, NCT03892967) cluster-randomized trial at Mayo Clinic as our initial pan-cancer cohort. E2C2 includes adults receiving Medical Hematology/Oncology care for a solid or liquid tumor at Mayo Clinic. We collected cancer diagnoses from the individually abstracted Mayo Clinic Cancer Registry to annotate the cancer type for the patients. Lung cancer related ICD-9 (162.X) and ICD-10 (C34.X) codes were used to build a search query on the Mayo’s EHR to find target patients from the E2C2 cohort, and to investigate the ICD-based lung cancer patient identification performance. Diagnosis, radiation oncology treatment (CPT 77261 - 77799), and antineoplastic drug administration data were collected from EHR as variables. Logistic Regression (LR), Support vector machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGB) were selected to build models for lung cancer patient identification. 10-fold cross-validation was implemented to assess the models. Precision, Recall, F1 Score, and area under the curve (AUC) were selected to measure the performance. Results: We collected 13,893 patients with a specific cancer diagnosis, and 1,394 were identified as having lung cancer. The identification performance across different methods is shown. The ICD-based method only had 0.65 precision. It means we collected a lot of false positive cases (other cancer but no lung cancer patient), as we mentioned in the background. SVM gets the best precision results, but its recall and F1 score are not good enough. XGB shows the best F1 Score and AUC, which also means this method achieved the best and most balanced performance. Conclusions: In this study, we found that XGB-based methods achieved the best identification performance for lung cancer. In the future, we will investigate if this is also true for the identification of other cancer types. [Table: see text]

Development of a machine learning algorithm based on administrative claims data for identification of ED anaphylaxis patient visits.

Epidemiologic studies of anaphylaxis commonly rely on International Classification of Diseases (ICD) codes to identify anaphylaxis cases, which may lead to suboptimal epidemiologic classification. We sought to develop and assess the accuracy of a machine learning algorithm using ICD codes and other administrative data compared with ICD code-only algorithms to identify emergency department (ED) anaphylaxis visits. We conducted a retrospective review of ED visits from January 2013 to September 2017. Potential ED anaphylaxis visits were identified using 3 methods: anaphylaxis ICD diagnostic codes (method 1), ICD symptom-based codes with or without a code indicating an allergic trigger (method 2), and ICD codes indicating a potential allergic reaction only (method 3). Amachine learning algorithm was developed from administrative data, and test characteristics were compared with ICD code-only algorithms. A total of 699 of 2191 (31.9%) potential ED anaphylaxis visits were classified as anaphylaxis. The sensitivity and specificity of method 1 were 49.1% and 87.5%, respectively. Method 1 used in combination with method 2 resulted in a sensitivity of 53.9% and a specificity of 68.7%. Method 1 used in combination with method 3 resulted in a sensitivity of 98.4% and a specificity of 15.1%. The sensitivity and specificity of the machine learning algorithm were 87.3% and 79.1%, respectively. ICD coding alone demonstrated poor sensitivity in identifying cases of anaphylaxis, with venom-related anaphylaxis missing 96% of cases. The machine learning algorithm resulted in a better balance of sensitivity and specificity and improves upon previous strategies to identify ED anaphylaxis visits.

Development and external validation of automated ICD-10 coding from discharge summaries using deep learning approaches

ObjectivesTo develop an automated international classification of diseases (ICD) coding tool using natural language processing (NLP) and discharge summary texts from Thailand. Materials and methodsThe development phase included 15,329 discharge summaries from Ramathibodi Hospital from January 2015 to December 2020. The external validation phase included Medical Information Mart for Intensive Care III (MIMIC-III) data. Three algorithms were developed: naïve Bayes with term frequency-inverse document frequency (NB-TF-IDF), convolutional neural network with neural word embedding (CNN-NWE), and CNN with PubMedBERT (CNN-PubMedBERT). In addition, two state-of-the-art models were also considered; convolutional attention for multi-label classification (CAML) and pretrained language models for automatic ICD coding (PLM-ICD). ResultsThe CNN-PubMedBERT model provided average micro- and macro-area under precision-recall curve (AUPRC) of 0.6605 and 0.5538, which outperformed CNN-NWE (0.6528 and 0.5564), NB-TF-IDF (0.4441 and 0.3562), and CAML (0.6257 and 0.4964), with corresponding differences of (0.0077 and −0.0026), (0.2164 and 0.1976), and (0.0348 and 0.0574), respectively. However, CNN-PubMedBERT performed less well relative to PLM-ICD, with corresponding AUPRCs of 0.7202 and 0.5865. The CNN-PubMedBERT model was externally validated using two subsets of MIMIC-III; MIMIC-ICD-10, and MIMIC-ICD-9 datasets, which contained 40,923 and 31,196 discharge summaries. The average micro-AUPRCs were 0.3745, 0.6878, and 0.6699, corresponding to directly predictive MIMIC-ICD-10, MIMIC-ICD-10 fine-tuning, and MIMIC-ICD-9 fine-tuning approaches; the average macro-AUPRCs for the corresponding models were 0.2819, 0.4219 and 0.5377, respectively. DiscussionCNN-PubMedBERT performed second-best to PLM-ICD, with considerable variation observed between average micro- and macro-AUPRC, especially for external validation, generally indicating good overall prediction but limited predictive value for small sample sizes. External validation in a US cohort demonstrated a higher level of model prediction performance. ConclusionBoth PLM-ICD and CNN-PubMedBERT models may provide useful tools for automated ICD-10 coding. Nevertheless, further evaluation and validation within Thai and Asian healthcare systems may prove more informative for clinical application.

Learning from undercoded clinical records for automated International Classification of Diseases (ICD) coding.

To develop an unbiased objective for learning automatic coding algorithms from clinical records annotated with only partial relevant International Classification of Diseases codes, as annotation noise in undercoded clinical records used as training data can mislead the learning process of deep neural networks. We use Medical Information Mart for Intensive Care III as our dataset. We employ positive-unlabeled learning to achieve unbiased loss estimation, which is free of misleading training signal. We then utilize reweighting mechanism to compensate for the imbalance between positive and negative samples. To further close the performance gap caused by poor quality annotation, we integrate the supervision provided by the automatic annotation tool Medical Concept Annotation Toolkit which can ease the heavy burden of manual validation. Our benchmarking results show that positive-unlabeled learning with reweighting outperforms competitive baseline methods over a range of missing label ratios. Integrating supervision provided by annotation tool further boosted the performance. Considering the annotation noise and severe imbalance, unbiased loss estimation and reweighting mechanism are both important for learning from undercoded clinical records. Unbiased loss requires the estimation of false negative ratios and estimation through trained models is practical and competitive. The combination of positive-unlabeled learning with reweighting and supervision provided by the annotation tool is a promising solution to learn from undercoded clinical records.