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.

Hierarchical label-wise attention transformer model for explainable ICD coding.

International Classification of Diseases (ICD) coding plays an important role in systematically classifying morbidity and mortality data. In this study, we propose a hierarchical label-wise attention Transformer model (HiLAT) for the explainable prediction of ICD codes from clinical documents. HiLAT firstly fine-tunes a pretrained Transformer model to represent the tokens of clinical documents. We subsequently employ a two-level hierarchical label-wise attention mechanism that creates label-specific document representations. These representations are in turn used by a feed-forward neural network to predict whether a specific ICD code is assigned to the input clinical document of interest. We evaluate HiLAT using hospital discharge summaries and their corresponding ICD-9 codes from the MIMIC-III database. To investigate the performance of different types of Transformer models, we develop ClinicalplusXLNet, which conducts continual pretraining from XLNet-Base using all the MIMIC-III clinical notes. The experiment results show that the F1 scores of the HiLAT + ClinicalplusXLNet outperform the previous state-of-the-art models for the top-50 most frequent ICD-9 codes from MIMIC-III. Visualisations of attention weights present a potential explainability tool for checking the face validity of ICD code predictions.

Using a Simple Prescription Gap to Determine Warfarin Discontinuation Can Lead to Substantial Misclassification.

BACKGROUND: Warfarin remains widely used and a key comparator in studies of other direct oral anticoagulants. As longer-than-needed warfarin prescriptions are often provided to allow for dosing adjustments according to international normalized ratios (INRs), the common practice of using a short allowable gap between dispensings to define warfarin discontinuation may lead to substantial misclassification of warfarin exposure. We aimed to quantify such misclassification and determine the optimal algorithm to define warfarin discontinuation. METHODS: We linked Medicare claims data from 2007 to 2014 with a multicenter electronic health records system. The study cohort comprised patients ≥65 years with atrial fibrillation and venous thromboembolism initiating warfarin. We compared results when defining warfarin discontinuation by (1) different gaps (3, 7, 14, 30, and 60 days) between dispensings and (2) having a gap ≤60 days or bridging larger gaps if there was INR ordering at least every 42 days (60_INR). Discontinuation was considered misclassified if there was an INR ≥2 within 7 days after the discontinuation date. RESULTS: Among 3,229 patients, a shorter gap resulted in a shorter mean follow-up time (82, 95, 117, 159, 196, and 259 days for gaps of 3, 7, 14, 30, 60, and 60_INR, respectively; p < 0.001). Incorporating INR (60_INR) can reduce misclassification of warfarin discontinuation from 68 to 4% (p < 0.001). The on-treatment risk estimation of clinical endpoints varied significantly by discontinuation definitions. CONCLUSION: Using a short gap between warfarin dispensings to define discontinuation may lead to substantial misclassification, which can be improved by incorporating intervening INR codes.

Comparing reliability of ICD-10-based COVID-19 comorbidity data to manual chart review, a retrospective cross-sectional study.

International Statistical Classification of Disease and Related Health Problems, 10th Revision codes (ICD-10) are used to characterize cohort comorbidities. Recent literature does not demonstrate standardized extraction methods. OBJECTIVE: Compare COVID-19 cohort manual-chart-review and ICD-10-based comorbidity data; characterize the accuracy of different methods of extracting ICD-10-code-based comorbidity, including the temporal accuracy with respect to critical time points such as day of admission. DESIGN: Retrospective cross-sectional study. MEASUREMENTS: ICD-10-based-data performance characteristics relative to manual-chart-review. RESULTS: Discharge billing diagnoses had a sensitivity of 0.82 (95% confidence interval [CI]: 0.79-0.85; comorbidity range: 0.35-0.96). The past medical history table had a sensitivity of 0.72 (95% CI: 0.69-0.76; range: 0.44-0.87). The active problem list had a sensitivity of 0.67 (95% CI: 0.63-0.71; range: 0.47-0.71). On day of admission, the active problem list had a sensitivity of 0.58 (95% CI: 0.54-0.63; range: 0.30-0.68)and past medical history table had a sensitivity of 0.48 (95% CI: 0.43-0.53; range: 0.30-0.56). CONCLUSIONS AND RELEVANCE: ICD-10-based comorbidity data performance varies depending on comorbidity, data source, and time of retrieval; there are notable opportunities for improvement. Future researchers should clearly outline comorbidity data source and validate against manual-chart-review.

Prevention and Management of Pseudoverrucous Lesions: A Review and Case Scenarios.

GENERAL PURPOSE: To present the associated risk factors, prevention measures, and assessment and management of pseudoverrucous lesions specific to a surgically created ileal conduit, as well as three clinical scenarios illustrating this condition. TARGET AUDIENCE: This continuing education activity is intended for physicians, physician assistants, nurse practitioners, and nurses with an interest in skin and wound care. LEARNING OBJECTIVES/OUTCOMES: After participating in this educational activity, the participant will:1. Define pseudoverrucous lesions.2. Identify the risk factors for stoma complications such as pseudoverrucous lesions.3. Select the appropriate routine care procedures to teach patients following stoma creation to help prevent pseudoverrucous lesions.4. Choose the recommended treatment options for patients who develop pseudoverrucous lesions.

Pseudoverrucous lesions are a late peristomal complication that occurs most commonly in people with urinary stomas. Impairment of the peristomal skin can result in pouching system leaks that can translate into odor, embarrassment, and diminished quality of life. Prevention is key to maintaining smooth, dry skin and intact psyche. Treatment revolves around outpatient postoperative follow-up, refitting the pouching system to eliminate moisture impacting the peristomal area, modification of pouching system wear time, acidification of the urine, and intensive education. This review includes three case scenarios to support early, intermediate, and late-stage intervention guidelines. Some interventions were successful; one case remains unresolved.

La trazabilidad en el sistema logístico de medicamentos en Cuba y el uso de las tecnologías de auto-identificación / Traceability in the logistics system of medicines in Cuba, use of auto-identification

La trazabilidad es la capacidad para rastrear la historia, aplicación o ubicación de un objeto bajo consideración. En el ámbito farmacéutico, el rastreo y seguimiento de los medicamentos, incluyendo las vacunas y otros medicamentos biológicos, a lo largo de la cadena de suministro constituye un requisito obligatorio establecido por las autoridades sanitarias a nivel internacional, que se exige en mayor o menor magnitud en las reglamentaciones vigentes. En este artículo se analiza el sistema de codificación y clasificación en el sector de la salud y su estado actual en la cadena de suministro de medicamentos de Cuba. Se presenta un procedimiento para la implementación de las tecnologías de auto-identificación e intercambio electrónico de datos, mediante el uso de GS1 en el sistema de codificación y clasificación empleado en el sector de salud, que permita la trazabilidad en toda la cadena de suministro en Cuba(AU)

Traceability is the capability to track the history, application or location of an object under consideration. In the pharmaceutical field, the tracking and monitoring of medicines, including vaccines and other biological medicines, along the supply chain constitutes a mandatory requirement established by the sanitary authorities at an international level, which is demanded to a greater or lesser extent in the regulations in force. This research was carried out involving different links in the drug supply chain in Cuba, ranging from drug suppliers, drug distribution company, to healthcare centers and pharmacies. An analysis is carried out on the current coding and classification system, detecting the ineffectiveness of the identification of the drugs as the main deficiency. A procedure is proposed for the implementation of the auto-identification and electronic data interchange technologies using GS1 in the coding and classification system used in the health sector that allows traceability throughout the supply chain in Cuba(AU)

Comparison of ICD-9-CM to ICD-10-CM Crosswalks Derived by Physician and Clinical Coder vs. Automated Methods.

Purpose: To evaluate whether automated methods are sufficient for deriving ICD-10-CM algorithms by comparing ICD-9-CM to ICD-10-CM crosswalks from general equivalence mappings (GEMs) with physician/clinical coder-derived crosswalks. Patients and methods: Forward mapping was used to derive ICD-10-CM crosswalks for 10 conditions. As a sensitivity analysis, forward-backward mapping (FBM) was also conducted for three clinical conditions. The physician/coder independently developed crosswalks for the same conditions. Differences between the crosswalks were summarized using the Jaccard similarity coefficient (JSC). Results: Physician/coder crosswalks were typically far more inclusive than GEMs crosswalks. Crosswalks for peripheral artery disease were most dissimilar (JSC: 0.06), while crosswalks for mild cognitive impairment (JSC: 1) and congestive heart failure (0.85) were most similar. FBM added ICD-10-CM codes for all three conditions but did not consistently increase similarity between crosswalks. Conclusion: The GEMs and physician/coder algorithms rarely aligned fully; human review is still required for ICD-9-CM to ICD-10-CM crosswalk development.

Coding practice in national and regional kidney biopsy registries.

BACKGROUND: Kidney biopsy registries all over the world benefit research, teaching and health policy. Comparison, aggregation and exchange of data is however greatly dependent on how registration and coding of kidney biopsy diagnoses are performed. This paper gives an overview over kidney biopsy registries, explores how these registries code kidney disease and identifies needs for improvement of coding practice. METHODS: A literature search was undertaken to identify biopsy registries for medical kidney diseases. These data were supplemented with information from personal contacts and from registry websites. A questionnaire was sent to all identified registries, investigating age of registries, scope, method of coding, possible mapping to international terminologies as well as self-reported problems and suggestions for improvement. RESULTS: Sixteen regional or national kidney biopsy registries were identified, of which 11 were older than 10 years. Most registries were located either in Europe (10/16) or in Asia (4/16). Registries most often use a proprietary coding system (12/16). Only a few of these coding systems were mapped to SNOMED CT (1), older SNOMED versions (2) or ERA-EDTA PRD (3). Lack of maintenance and updates of the coding system was the most commonly reported problem. CONCLUSIONS: There were large gaps in the global coverage of kidney biopsy registries. Limited use of international coding systems among existing registries hampers interoperability and exchange of data. The study underlines that the use of a common and uniform coding system is necessary to fully realize the potential of kidney biopsy registries.

Artificial Intelligence for Unstructured Healthcare Data: Application to Coding of Patient Reporting of Adverse Drug Reactions.

Adverse drug reaction (ADR) reporting is a major component of drug safety monitoring; its input will, however, only be optimized if systems can manage to deal with its tremendous flow of information, based primarily on unstructured text fields. The aim of this study was to develop an automated system allowing to code ADRs from patient reports. Our system was based on a knowledge base about drugs, enriched by supervised machine learning (ML) models trained on patients reporting data. To train our models, we selected all cases of ADRs reported by patients to a French Pharmacovigilance Centre through a national web-portal between March 2017 and March 2019 (n = 2,058 reports). We tested both conventional ML models and deep-learning models. We performed an external validation using a dataset constituted of a random sample of ADRs reported to the Marseille Pharmacovigilance Centre over the same period (n = 187). Here, we show that regarding area under the curve (AUC) and F-measure, the best model to identify ADRs was gradient boosting trees (LGBM), with an AUC of 0.93 (0.92-0.94) and F-measure of 0.72 (0.68-0.75). This model was run for external validation showing an AUC of 0.91 and a F-measure of 0.58. We evaluated an artificial intelligence pipeline that was found able to learn how to identify correctly ADRs from unstructured data. This result allowed us to start a new study using more data to further improve our performance and offer a tool that is useful in practice to efficiently manage drug safety information.

Coding of Sexual Assault by Emergency Physicians: A Nationally Representative Study.

INTRODUCTION: Sexual assault is a public health problem that affects many Americans and has multiple long-lasting effects on victims. Medical evaluation after sexual assault frequently occurs in the emergency department, and documentation of the visit plays a significant role in decisions regarding prosecution and outcomes of legal cases against perpetrators. The American College of Emergency Physicians recommends coding such visits as sexual assault rather than adding modifiers such as "alleged." METHODS: This study reviews factors associated with coding of visits as sexual assault compared to suspected sexual assault using the 2016 Nationwide Emergency Department Sample. RESULTS: Younger age, female gender, a larger number of procedure codes, urban hospital location, and lack of concurrent alcohol use are associated with coding for confirmed sexual assault. CONCLUSION: Implications of this coding are discussed.

Explainable automated coding of clinical notes using hierarchical label-wise attention networks and label embedding initialisation.

BACKGROUND: Diagnostic or procedural coding of clinical notes aims to derive a coded summary of disease-related information about patients. Such coding is usually done manually in hospitals but could potentially be automated to improve the efficiency and accuracy of medical coding. Recent studies on deep learning for automated medical coding achieved promising performances. However, the explainability of these models is usually poor, preventing them to be used confidently in supporting clinical practice. Another limitation is that these models mostly assume independence among labels, ignoring the complex correlations among medical codes which can potentially be exploited to improve the performance. METHODS: To address the issues of model explainability and label correlations, we propose a Hierarchical Label-wise Attention Network (HLAN), which aimed to interpret the model by quantifying importance (as attention weights) of words and sentences related to each of the labels. Secondly, we propose to enhance the major deep learning models with a label embedding (LE) initialisation approach, which learns a dense, continuous vector representation and then injects the representation into the final layers and the label-wise attention layers in the models. We evaluated the methods using three settings on the MIMIC-III discharge summaries: full codes, top-50 codes, and the UK NHS (National Health Service) COVID-19 (Coronavirus disease 2019) shielding codes. Experiments were conducted to compare the HLAN model and label embedding initialisation to the state-of-the-art neural network based methods, including variants of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). RESULTS: HLAN achieved the best Micro-level AUC and F1 on the top-50 code prediction, 91.9% and 64.1%, respectively; and comparable results on the NHS COVID-19 shielding code prediction to other models: around 97% Micro-level AUC. More importantly, in the analysis of model explanations, by highlighting the most salient words and sentences for each label, HLAN showed more meaningful and comprehensive model interpretation compared to the CNN-based models and its downgraded baselines, HAN and HA-GRU. Label embedding (LE) initialisation significantly boosted the previous state-of-the-art model, CNN with attention mechanisms, on the full code prediction to 52.5% Micro-level F1. The analysis of the layers initialised with label embeddings further explains the effect of this initialisation approach. The source code of the implementation and the results are openly available at https://github.com/acadTags/Explainable-Automated-Medical-Coding. CONCLUSION: We draw the conclusion from the evaluation results and analyses. First, with hierarchical label-wise attention mechanisms, HLAN can provide better or comparable results for automated coding to the state-of-the-art, CNN-based models. Second, HLAN can provide more comprehensive explanations for each label by highlighting key words and sentences in the discharge summaries, compared to the n-grams in the CNN-based models and the downgraded baselines, HAN and HA-GRU. Third, the performance of deep learning based multi-label classification for automated coding can be consistently boosted by initialising label embeddings that captures the correlations among labels. We further discuss the advantages and drawbacks of the overall method regarding its potential to be deployed to a hospital and suggest areas for future studies.

Use of a Clinical Electrocardiographic Database to Enhance Atrial Fibrillation/Atrial Flutter Identification Algorithms Based on Administrative Data.

Background Administrative data have limited sensitivity for case finding of atrial fibrillation/atrial flutter (AF/AFL). Linkage with clinical repositories of interpreted ECGs may enhance diagnostic yield of AF/AFL. Methods and Results We retrieved 369 ECGs from the institutional Marquette Universal System for Electrocardiography (MUSE) repository as validation samples, with rhythm coded as AF (n=49), AFL (n=50), or other competing rhythm diagnoses (n=270). With blinded, duplicate review of ECGs as the reference comparison, we compared multiple MUSE coding definitions for identifying AF/AFL. We tested the agreement between MUSE diagnosis and reference comparison, and calculated the sensitivity and specificity. Using a data set linking clinical registries, administrative data, and the MUSE repository (n=11 662), we assessed the incremental diagnostic yield of AF/AFL by incorporating ECG data to administrative data-based algorithms. The agreement between MUSE diagnosis and reference comparison depended on the coding definitions applied, with the Cohen κ ranging from 0.57 to 0.75. Sensitivity ranged from 60.6% to 79.1%, and specificity ranged from 93.2% to 98.0%. A coding definition with AF/AFL appearing in the first 3 ECG statements had the highest sensitivity (79.1%), with little loss of specificity (94.5%). Compared with the algorithms with only administrative data, incorporating ECG data increased the diagnostic yield of preexisting AF/AFL by 14.5% and incident AF/AFL by 7.5% to 16.1%. Conclusions Routine ECG interpretation using MUSE coding is highly specific and moderately sensitive for AF/AFL detection. Inclusion of MUSE ECG data in AF/AFL case identification algorithms can identify cases missed using administrative data-based algorithms alone.

Utility of 30-Day Readmission Metrics in the Burn Population: Navigating Quality Metric Limitations in Special Populations.

Hospital readmission data may be a useful tool in identifying risk factors leading to higher costs of care or poorer overall outcomes. Several studies have emerged utilizing these datasets to examine the trauma and burn population, which have been unable to distinguish planned from unplanned readmissions. The 2014 Nationwide Readmissions Database was queried for 363 burn-specific ICD-9 DX codes and filtered by age and readmission status to capture the adult burn population. Additionally, burn-related excision and grafting procedures were filtered from 25 ICD-9 SG codes to distinguish planned readmissions. A total of 26,719 burn patients were identified with 781 all-cause unscheduled 30-day readmissions. Further filtering by burn-related excision and grafting procedures then identified 468 patients undergoing a burn-related excision and grafting procedure on readmission, reducing the dataset to 313 patients and identifying up to 60% of readmissions as possibly improperly coded planned readmissions. From this dataset, nonoperative management on initial admission was found to have the strongest correlation with readmission (OR 5.00; 3.33-7.14). Notably corrected data, when stratified by annual burn patient admission volume, identified a significant likelihood of readmission (OR 4.57; 2.15-9.70) of centers receiving the lowest annual number of burn patients, which was not identified in the unfiltered dataset. Healthcare performance statistics may be a powerful metric when utilized appropriately; however, these databases must be carefully applied to small and special populations. This study has determined that as many as 60% of burn patient readmissions included in prior studies may be improperly coded planned readmissions.