BETA: a comprehensive benchmark for computational drug-target prediction.

Internal validation is the most popular evaluation strategy used for drug-target predictive models. The simple random shuffling in the cross-validation, however, is not always ideal to handle large, diverse and copious datasets as it could potentially introduce bias. Hence, these predictive models cannot be comprehensively evaluated to provide insight into their general performance on a variety of use-cases (e.g. permutations of different levels of connectiveness and categories in drug and target space, as well as validations based on different data sources). In this work, we introduce a benchmark, BETA, that aims to address this gap by (i) providing an extensive multipartite network consisting of 0.97 million biomedical concepts and 8.5 million associations, in addition to 62 million drug-drug and protein-protein similarities and (ii) presenting evaluation strategies that reflect seven cases (i.e. general, screening with different connectivity, target and drug screening based on categories, searching for specific drugs and targets and drug repurposing for specific diseases), a total of seven Tests (consisting of 344 Tasks in total) across multiple sampling and validation strategies. Six state-of-the-art methods covering two broad input data types (chemical structure- and gene sequence-based and network-based) were tested across all the developed Tasks. The best-worst performing cases have been analyzed to demonstrate the ability of the proposed benchmark to identify limitations of the tested methods for running over the benchmark tasks. The results highlight BETA as a benchmark in the selection of computational strategies for drug repurposing and target discovery.

FedFSA: Hybrid and federated framework for functional status ascertainment across institutions.

INTRODUCTION: Patients' functional status assesses their independence in performing activities of daily living, including basic ADLs (bADL), and more complex instrumental activities (iADL). Existing studies have discovered that patients' functional status is a strong predictor of health outcomes, particularly in older adults. Depite their usefulness, much of the functional status information is stored in electronic health records (EHRs) in either semi-structured or free text formats. This indicates the pressing need to leverage computational approaches such as natural language processing (NLP) to accelerate the curation of functional status information. In this study, we introduced FedFSA, a hybrid and federated NLP framework designed to extract functional status information from EHRs across multiple healthcare institutions. METHODS: FedFSA consists of four major components: 1) individual sites (clients) with their private local data, 2) a rule-based information extraction (IE) framework for ADL extraction, 3) a BERT model for functional status impairment classification, and 4) a concept normalizer. The framework was implemented using the OHNLP Backbone for rule-based IE and open-source Flower and PyTorch library for federated BERT components. For gold standard data generation, we carried out corpus annotation to identify functional status-related expressions based on ICF definitions. Four healthcare institutions were included in the study. To assess FedFSA, we evaluated the performance of category- and institution-specific ADL extraction across different experimental designs. RESULTS: ADL extraction performance ranges from an F1-score of 0.907 to 0.986 for bADL and 0.825 to 0.951 for iADL across the four healthcare sites. The performance for ADL extraction with impairment ranges from an F1-score of 0.722 to 0.954 for bADL and 0.674 to 0.813 for iADL across four healthcare sites. For category-specific ADL extraction, laundry and transferring yielded relatively high performance, while dressing, medication, bathing, and continence achieved moderate-high performance. Conversely, food preparation and toileting showed low performance. CONCLUSION: NLP performance varied across ADL categories and healthcare sites. Federated learning using a FedFSA framework performed higher than non-federated learning for impaired ADL extraction at all healthcare sites. Our study demonstrated the potential of the federated learning framework in functional status extraction and impairment classification in EHRs, exemplifying the importance of a large-scale, multi-institutional collaborative development effort.

MedTator: a serverless annotation tool for corpus development.

SUMMARY: Building a high-quality annotation corpus requires expenditure of considerable time and expertise, particularly for biomedical and clinical research applications. Most existing annotation tools provide many advanced features to cover a variety of needs where the installation, integration and difficulty of use present a significant burden for actual annotation tasks. Here, we present MedTator, a serverless annotation tool, aiming to provide an intuitive and interactive user interface that focuses on the core steps related to corpus annotation, such as document annotation, corpus summarization, annotation export and annotation adjudication. AVAILABILITY AND IMPLEMENTATION: MedTator and its tutorial are freely available from https://ohnlp.github.io/MedTator. MedTator source code is available under the Apache 2.0 license: https://github.com/OHNLP/MedTator. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Development of Parkinson Disease and Its Relationship with Incidentally Discovered White Matter Disease and Covert Brain Infarction in a Real-World Cohort.

OBJECTIVE: This study aimed to examine the relationship between covert cerebrovascular disease, comprised of covert brain infarction and white matter disease, discovered incidentally in routine care, and subsequent Parkinson disease. METHODS: Patients were ≥50 years and received neuroimaging for non-stroke indications in the Kaiser Permanente Southern California system from 2009 to 2019. Natural language processing identified incidentally discovered covert brain infarction and white matter disease and classified white matter disease severity. The Parkinson disease outcome was defined as 2 ICD diagnosis codes. RESULTS: 230,062 patients were included (median follow-up 3.72 years). A total of 1,941 Parkinson disease cases were identified (median time-to-event 2.35 years). Natural language processing identified covert cerebrovascular disease in 70,592 (30.7%) patients, 10,622 (4.6%) with covert brain infarction and 65,814 (28.6%) with white matter disease. After adjustment for known risk factors, white matter disease was associated with Parkinson disease (hazard ratio 1.67 [95%CI, 1.44, 1.93] for patients <70 years and 1.33 [1.18, 1.50] for those ≥70 years). Greater severity of white matter disease was associated with increased incidence of Parkinson disease(/1,000 person-years), from 1.52 (1.43, 1.61) in patients without white matter disease to 4.90 (3.86, 6.13) in those with severe disease. Findings were robust when more specific definitions of Parkinson disease were used. Covert brain infarction was not associated with Parkinson disease (adjusted hazard ratio = 1.05 [0.88, 1.24]). INTERPRETATION: Incidentally discovered white matter disease was associated with subsequent Parkinson disease, an association strengthened with younger age and increased white matter disease severity. Incidentally discovered covert brain infarction did not appear to be associated with subsequent Parkinson disease. ANN NEUROL 2022;92:620-630.

Stratifying Future Stroke Risk with Incidentally Discovered White Matter Disease Severity and Covert Brain Infarct Site.

BACKGROUND: Covert cerebrovascular disease (CCD) includes white matter disease (WMD) and covert brain infarction (CBI). Incidentally discovered CCD is associated with increased risk of subsequent symptomatic stroke. However, it is unknown whether the severity of WMD or the location of CBI predicts risk. OBJECTIVES: The aim of this study was to examine the association of incidentally discovered WMD severity and CBI location with risk of subsequent symptomatic stroke. METHOD: This retrospective cohort study includes patients aged ≥50 years old in the Kaiser Permanente Southern California health system who received neuroimaging for a nonstroke indication between 2009 and 2019. Incidental CBI and WMD were identified via natural language processing of the neuroimage report, and WMD severity was classified into grades. RESULTS: A total of 261,960 patients received neuroimaging; 78,555 patients (30.0%) were identified to have incidental WMD and 12,857 patients (4.9%) to have incidental CBI. Increasing WMD severity is associated with an increased incidence rate of future stroke. However, the stroke incidence rate in CT-identified WMD is higher at each level of severity compared to rates in MRI-identified WMD. Patients with mild WMD via CT have a stroke incidence rate of 24.9 per 1,000 person-years, similar to that of patients with severe WMD via MRI. Among incidentally discovered CBI patients with a determined CBI location, 97.9% are subcortical rather than cortical infarcts. CBI confers a similar risk of future stroke, whether cortical or subcortical or whether MRI- or CT-detected. CONCLUSIONS: Increasing severity of incidental WMD is associated with an increased risk of future symptomatic stroke, dependent on the imaging modality. Subcortical and cortical CBI conferred similar risks.

Stratifying future stroke risk with incidentally-discovered white matter disease severity and covert brain infarct site.

BACKGROUND: Covert cerebrovascular disease (CCD) includes white matter disease (WMD) and covert brain infarction (CBI). Incidentally-discovered CCD is associated with increased risk of subsequent symptomatic stroke. However, it is unknown whether the severity of WMD or the location of CBI predicts risk. OBJECTIVES: To examine the association of incidentally-discovered WMD severity and CBI location with risk of subsequent symptomatic stroke. METHOD: This retrospective cohort study includes patients 50 years old in the Kaiser Permanente Southern California health system who received neuroimaging for a non-stroke indication between 2009-2019. Incidental CBI and WMD were identified via natural language processing of the neuroimage report, and WMD severity was classified into grades. RESULTS: 261,960 patients received neuroimaging; 78,555 (30.0%) were identified to have incidental WMD, and 12,857 (4.9%) to have incidental CBI. Increasing WMD severity is associated with increased incidence rate of future stroke. However, the stroke incidence rate in CT-identified WMD is higher at each level of severity compared to rates in MRI-identified WMD. Patients with mild WMD via CT have a stroke incidence rate of 24.9 per 1,000 person-years, similar to that of patients with severe WMD via MRI. Among incidentally-discovered CBI patients with a determined CBI location, 97.9% are subcortical rather than cortical infarcts. CBI confers a similar risk of future stroke, whether cortical or subcortical, or whether MRI- or CT-detected. CONCLUSIONS: Increasing severity of incidental WMD is associated with an increased risk of future symptomatic stroke, dependent on the imaging modality. Subcortical and cortical CBI conferred similar risks.

External Validation of Natural Language Processing Algorithms to Extract Common Data Elements in THA Operative Notes.

BACKGROUND: Natural language processing (NLP) systems are distinctive in their ability to extract critical information from raw text in electronic health records (EHR). We previously developed three algorithms for total hip arthroplasty (THA) operative notes with rules aimed at capturing (1) operative approach, (2) fixation method, and (3) bearing surface using inputs from a single institution. The purpose of this study was to externally validate and improve these algorithms as a prerequisite for broader adoption in automated registry data curation. METHODS: The previous NLP algorithms developed at Mayo Clinic were deployed and refined on EHRs from OrthoCarolina, evaluating 39 randomly selected primary THA operative reports from 2018 to 2021. Operative reports were available only in PDF format, requiring conversion to "readable" text with Adobe software. Accuracy statistics were calculated against manual chart review. RESULTS: The operative approach, fixation technique, and bearing surface algorithms all demonstrated perfect accuracy of 100%. By comparison, validated performance at the developing center yielded an accuracy of 99.2% for operative approach, 90.7% for fixation technique, and 95.8% for bearing surface. CONCLUSION: NLP algorithms applied to data from an external center demonstrated excellent accuracy in delineating common elements in THA operative notes. Notably, the algorithms had no functional problems evaluating scanned PDFs that were converted to "readable" text by common software. Taken together, these findings provide promise for NLP applied to scanned PDFs as a source to develop large registries by reliably extracting data of interest from very large unstructured data sets in an expeditious and cost-effective manner.

Real-time risk prediction of colorectal surgery-related post-surgical complications using GRU-D model.

BACKGROUND: Post-surgical complications (PSCs) have been an increasing concern for hospitals in light of Medicare penalties for 30-day readmissions. PSCs have become a target for quality improvement and benchmark for the healthcare system. Over half (60 %) of the deep or organ space surgical site infections are discovered after discharge, leading to a readmission. There has thus been a push to develop risk prediction models for targeted preventive interventions for PSCs. OBJECTIVE: We experiment several Gated Recurrent Unit with Decay (GRU-D) based deep learning architectures with various feature sampling schemes in predicting the risk of colorectal PSCs and compare with atemporal logistic regression models (logit). METHOD: We used electronic health record (EHR) data of 3,535 colorectal surgical patients involved in the national surgical quality improvement program (NSQIP) between 2006 and 2018. Single layer, stacked layer, and multimodal GRU-D models with sigmoid activation were used to develop risk prediction models. Area Under the Receiver Operating Characteristic curve (AUROC) was calculated by comparing predicted probability of developing complications versus truly observed PSCs (NSQIP adjudicated) within 30 days after surgery. We set up cross-validation and an independent held-out dataset for testing model performance consistency. RESULTS AND CONCLUSION: The primary contribution of our study is the formulation of a novel real-time PSC risk prediction task using GRU-D with demonstrated clinical utility. GRU-D outperforms the logit model in predicting wound and organ space infection and shows improved performance as additional data points become available. Logit model outperforms GRU-D before surgery for superficial infection and bleeding. For the same sampling scheme, there is no obvious advantage of complex architectures (stacked, multimodal) over single layer GRU-D. Obtaining more data points closer to the occurrence of PSCs is more important than using a more frequent sampling scheme in training GRU-D models. The fourth predicted risk quartile by single layer GRU-D contains 63 %, 59 %, and 66 % organ space infection cases, at 4 h before, 72 h after, and 168 h after surgery, respectively, suggesting its potential application as a bedside risk assessment tool.

Agreement between neuroimages and reports for natural language processing-based detection of silent brain infarcts and white matter disease.

BACKGROUND: There are numerous barriers to identifying patients with silent brain infarcts (SBIs) and white matter disease (WMD) in routine clinical care. A natural language processing (NLP) algorithm may identify patients from neuroimaging reports, but it is unclear if these reports contain reliable information on these findings. METHODS: Four radiology residents reviewed 1000 neuroimaging reports (RI) of patients age > 50 years without clinical histories of stroke, TIA, or dementia for the presence, acuity, and location of SBIs, and the presence and severity of WMD. Four neuroradiologists directly reviewed a subsample of 182 images (DR). An NLP algorithm was developed to identify findings in reports. We assessed interrater reliability for DR and RI, and agreement between these two and with NLP. RESULTS: For DR, interrater reliability was moderate for the presence of SBIs (k = 0.58, 95 % CI 0.46-0.69) and WMD (k = 0.49, 95 % CI 0.35-0.63), and moderate to substantial for characteristics of SBI and WMD. Agreement between DR and RI was substantial for the presence of SBIs and WMD, and fair to substantial for characteristics of SBIs and WMD. Agreement between NLP and DR was substantial for the presence of SBIs (k = 0.64, 95 % CI 0.53-0.76) and moderate (k = 0.52, 95 % CI 0.39-0.65) for the presence of WMD. CONCLUSIONS: Neuroimaging reports in routine care capture the presence of SBIs and WMD. An NLP can identify these findings (comparable to direct imaging review) and can likely be used for cohort identification.

An aberration detection-based approach for sentinel syndromic surveillance of COVID-19 and other novel influenza-like illnesses.

Coronavirus Disease 2019 has emerged as a significant global concern, triggering harsh public health restrictions in a successful bid to curb its exponential growth. As discussion shifts towards relaxation of these restrictions, there is significant concern of second-wave resurgence. The key to managing these outbreaks is early detection and intervention, and yet there is a significant lag time associated with usage of laboratory confirmed cases for surveillance purposes. To address this, syndromic surveillance can be considered to provide a timelier alternative for first-line screening. Existing syndromic surveillance solutions are however typically focused around a known disease and have limited capability to distinguish between outbreaks of individual diseases sharing similar syndromes. This poses a challenge for surveillance of COVID-19 as its active periods tend to overlap temporally with other influenza-like illnesses. In this study we explore performing sentinel syndromic surveillance for COVID-19 and other influenza-like illnesses using a deep learning-based approach. Our methods are based on aberration detection utilizing autoencoders that leverages symptom prevalence distributions to distinguish outbreaks of two ongoing diseases that share similar syndromes, even if they occur concurrently. We first demonstrate that this approach works for detection of outbreaks of influenza, which has known temporal boundaries. We then demonstrate that the autoencoder can be trained to not alert on known and well-managed influenza-like illnesses such as the common cold and influenza. Finally, we applied our approach to 2019-2020 data in the context of a COVID-19 syndromic surveillance task to demonstrate how implementation of such a system could have provided early warning of an outbreak of a novel influenza-like illness that did not match the symptom prevalence profile of influenza and other known influenza-like illnesses.

Delirium occurrence and association with outcomes in hospitalized COVID-19 patients.

Delirium is reported to be one of the manifestations of coronavirus infectious disease 2019 (COVID-19) infection. COVID-19 hospitalized patients are at a higher risk of delirium. Pathophysiology behind the association of delirium and COVID-19 is uncertain. We analyzed the association of delirium occurrence with outcomes in hospitalized COVID-19 patients, across all age groups, at Mayo Clinic hospitals.A retrospective study of all hospitalized COVID-19 patients at Mayo Clinic between March 1, 2020 and December 31, 2020 was performed. Occurrence of delirium and outcomes of mortality, length of stay, readmission, and 30-day mortality after hospital discharge were measured. Chi-square test, student t-test, survival analysis, and logistic regression analysis were performed to measure and compare outcomes of delirium group adjusted for age, sex, Charlson comorbidity score, and COVID-19 severity with no-delirium group.A total of 4351 COVID-19 patients were included in the study. Delirium occurrence in the overall study population was noted to be 22.4%. The highest occurrence of delirium was also noted in patients with critical COVID-19 illness severity. A statistically significant OR 4.35 (3.27-5.83) for in-hospital mortality and an OR 4.54 (3.25-6.38) for 30-day mortality after discharge in the delirium group were noted. Increased hospital length of stay, 30-day readmission, and need for skilled nursing facility on discharge were noted in the delirium group. Delirium in hospitalized COVID-19 patients is a marker for increased mortality and morbidity. In this group, outcomes appear to be much worse when patients are older and have a critical severity of COVID-19 illness.

Natural Language Processing and Machine Learning for Identifying Incident Stroke From Electronic Health Records: Algorithm Development and Validation.

BACKGROUND: Stroke is an important clinical outcome in cardiovascular research. However, the ascertainment of incident stroke is typically accomplished via time-consuming manual chart abstraction. Current phenotyping efforts using electronic health records for stroke focus on case ascertainment rather than incident disease, which requires knowledge of the temporal sequence of events. OBJECTIVE: The aim of this study was to develop a machine learning-based phenotyping algorithm for incident stroke ascertainment based on diagnosis codes, procedure codes, and clinical concepts extracted from clinical notes using natural language processing. METHODS: The algorithm was trained and validated using an existing epidemiology cohort consisting of 4914 patients with atrial fibrillation (AF) with manually curated incident stroke events. Various combinations of feature sets and machine learning classifiers were compared. Using a heuristic rule based on the composition of concepts and codes, we further detected the stroke subtype (ischemic stroke/transient ischemic attack or hemorrhagic stroke) of each identified stroke. The algorithm was further validated using a cohort (n=150) stratified sampled from a population in Olmsted County, Minnesota (N=74,314). RESULTS: Among the 4914 patients with AF, 740 had validated incident stroke events. The best-performing stroke phenotyping algorithm used clinical concepts, diagnosis codes, and procedure codes as features in a random forest classifier. Among patients with stroke codes in the general population sample, the best-performing model achieved a positive predictive value of 86% (43/50; 95% CI 0.74-0.93) and a negative predictive value of 96% (96/100). For subtype identification, we achieved an accuracy of 83% in the AF cohort and 80% in the general population sample. CONCLUSIONS: We developed and validated a machine learning-based algorithm that performed well for identifying incident stroke and for determining type of stroke. The algorithm also performed well on a sample from a general population, further demonstrating its generalizability and potential for adoption by other institutions.

Automated Detection of Periprosthetic Joint Infections and Data Elements Using Natural Language Processing.

BACKGROUND: Periprosthetic joint infection (PJI) data elements are contained in both structured and unstructured documents in electronic health records and require manual data collection. The goal of this study is to develop a natural language processing (NLP) algorithm to replicate manual chart review for PJI data elements. METHODS: PJI was identified among all total joint arthroplasty (TJA) procedures performed at a single academic institution between 2000 and 2017. Data elements that comprise the Musculoskeletal Infection Society (MSIS) criteria were manually extracted and used as the gold standard for validation. A training sample of 1208 TJA surgeries (170 PJI cases) was randomly selected to develop the prototype NLP algorithms and an additional 1179 surgeries (150 PJI cases) were randomly selected as the test sample. The algorithms were applied to all consultation notes, operative notes, pathology reports, and microbiology reports to predict the correct status of PJI based on MSIS criteria. RESULTS: The algorithm, which identified patients with PJI based on MSIS criteria, achieved an f1-score (harmonic mean of precision and recall) of 0.911. Algorithm performance in extracting the presence of sinus tract, purulence, pathologic documentation of inflammation, and growth of cultured organisms from the involved TJA achieved f1-scores that ranged from 0.771 to 0.982, sensitivity that ranged from 0.730 to 1.000, and specificity that ranged from 0.947 to 1.000. CONCLUSION: NLP-enabled algorithms have the potential to automate data collection for PJI diagnostic elements, which could directly improve patient care and augment cohort surveillance and research efforts. Further validation is needed in other hospital settings. LEVEL OF EVIDENCE: Level III, Diagnostic.

Use of Natural Language Processing Algorithms to Identify Common Data Elements in Operative Notes for Knee Arthroplasty.

BACKGROUND: Natural language processing (NLP) methods have the capability to process clinical free text in electronic health records, decreasing the need for costly manual chart review, and improving data quality. We developed rule-based NLP algorithms to automatically extract surgery specific data elements from knee arthroplasty operative notes. METHODS: Within a cohort of 20,000 knee arthroplasty operative notes from 2000 to 2017 at a large tertiary institution, we randomly selected independent pairs of training and test sets to develop and evaluate NLP algorithms to detect five major data elements. The size of the training and test datasets were similar and ranged between 420 to 1592 surgeries. Expert rules using keywords in operative notes were used to implement NLP algorithms capturing: (1) category of surgery (total knee arthroplasty, unicompartmental knee arthroplasty, patellofemoral arthroplasty), (2) laterality of surgery, (3) constraint type, (4) presence of patellar resurfacing, and (5) implant model (catalog numbers). We used institutional registry data as our gold standard to evaluate the NLP algorithms. RESULTS: NLP algorithms to detect the category of surgery, laterality, constraint, and patellar resurfacing achieved 98.3%, 99.5%, 99.2%, and 99.4% accuracy on test datasets, respectively. The implant model algorithm achieved an F1-score (harmonic mean of precision and recall) of 99.9%. CONCLUSIONS: NLP algorithms are a promising alternative to costly manual chart review to automate the extraction of embedded information within knee arthroplasty operative notes. Further validation in other hospital settings will enhance widespread implementation and efficiency in data capture for research and clinical purposes. LEVEL OF EVIDENCE: Level III.

Clinical concept extraction: A methodology review.

BACKGROUND: Concept extraction, a subdomain of natural language processing (NLP) with a focus on extracting concepts of interest, has been adopted to computationally extract clinical information from text for a wide range of applications ranging from clinical decision support to care quality improvement. OBJECTIVES: In this literature review, we provide a methodology review of clinical concept extraction, aiming to catalog development processes, available methods and tools, and specific considerations when developing clinical concept extraction applications. METHODS: Based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, a literature search was conducted for retrieving EHR-based information extraction articles written in English and published from January 2009 through June 2019 from Ovid MEDLINE In-Process & Other Non-Indexed Citations, Ovid MEDLINE, Ovid EMBASE, Scopus, Web of Science, and the ACM Digital Library. RESULTS: A total of 6,686 publications were retrieved. After title and abstract screening, 228 publications were selected. The methods used for developing clinical concept extraction applications were discussed in this review.

Assessment of the impact of EHR heterogeneity for clinical research through a case study of silent brain infarction.

BACKGROUND: The rapid adoption of electronic health records (EHRs) holds great promise for advancing medicine through practice-based knowledge discovery. However, the validity of EHR-based clinical research is questionable due to poor research reproducibility caused by the heterogeneity and complexity of healthcare institutions and EHR systems, the cross-disciplinary nature of the research team, and the lack of standard processes and best practices for conducting EHR-based clinical research. METHOD: We developed a data abstraction framework to standardize the process for multi-site EHR-based clinical studies aiming to enhance research reproducibility. The framework was implemented for a multi-site EHR-based research project, the ESPRESSO project, with the goal to identify individuals with silent brain infarctions (SBI) at Tufts Medical Center (TMC) and Mayo Clinic. The heterogeneity of healthcare institutions, EHR systems, documentation, and process variation in case identification was assessed quantitatively and qualitatively. RESULT: We discovered a significant variation in the patient populations, neuroimaging reporting, EHR systems, and abstraction processes across the two sites. The prevalence of SBI for patients over age 50 for TMC and Mayo is 7.4 and 12.5% respectively. There is a variation regarding neuroimaging reporting where TMC are lengthy, standardized and descriptive while Mayo's reports are short and definitive with more textual variations. Furthermore, differences in the EHR system, technology infrastructure, and data collection process were identified. CONCLUSION: The implementation of the framework identified the institutional and process variations and the heterogeneity of EHRs across the sites participating in the case study. The experiment demonstrates the necessity to have a standardized process for data abstraction when conducting EHR-based clinical studies.

Use of Natural Language Processing Tools to Identify and Classify Periprosthetic Femur Fractures.

BACKGROUND: Manual chart review is labor-intensive and requires specialized knowledge possessed by highly trained medical professionals. The cost and infrastructure challenges required to implement this is prohibitive for most hospitals. Natural language processing (NLP) tools are distinctive in their ability to extract critical information from unstructured text in the electronic health records. As a simple proof-of-concept for the potential application of NLP technology in total hip arthroplasty (THA), we examined its ability to identify periprosthetic femur fractures (PPFFx) followed by more complex Vancouver classification. METHODS: PPFFx were identified among all THAs performed at a single academic institution between 1998 and 2016. A randomly selected training cohort (1538 THAs with 89 PPFFx cases) was used to develop the prototype NLP algorithm and an additional randomly selected cohort (2982 THAs with 84 PPFFx cases) was used to further validate the algorithm. Keywords to identify, and subsequently classify, Vancouver type PPFFx about THA were defined. The gold standard was confirmed by experienced orthopedic surgeons using chart and radiographic review. The algorithm was applied to consult and operative notes to evaluate language used by surgeons as a means to predict the correct pathology in the absence of a listed, precise diagnosis. Given the variability inherent to fracture descriptions by different surgeons, an iterative process was used to improve the algorithm during the training phase following error identification. Validation statistics were calculated using manual chart review as the gold standard. RESULTS: In distinguishing PPFFx, the NLP algorithm demonstrated 100% sensitivity and 99.8% specificity. Among 84 PPFFx test cases, the algorithm demonstrated 78.6% sensitivity and 94.8% specificity in determining the correct Vancouver classification. CONCLUSION: NLP-enabled algorithms are a promising alternative to manual chart review for identifying THA outcomes. NLP algorithms applied to surgeon notes demonstrated excellent accuracy in delineating PPFFx, but accuracy was low for Vancouver classification subtype. This proof-of-concept study supports the use of NLP technology to extract THA-specific data elements from the unstructured text in electronic health records in an expeditious and cost-effective manner. LEVEL OF EVIDENCE: Level III.

An Integrated Voice Recognition and Natural Language Processing Platform to Automatically Extract Thoracolumbar Injury Classification Score Features From Radiology Reports.

BACKGROUND: Many predictive models for estimating clinical outcomes after spine surgery have been reported in the literature. However, implementation of predictive scores in practice is limited by the time-intensive nature of manually abstracting relevant predictors. In this study, we designed natural language processing (NLP) algorithms to automate data abstraction for the thoracolumbar injury classification score (TLICS). METHODS: We retrieved the radiology reports of all Mayo Clinic patients with an International Classification of Diseases, 9th or 10th revision, code corresponding to a fracture of the thoracolumbar spine between January 2005 and October 2020. Annotated data were used to train an N-gram NLP model using machine learning methods, including random forest, stepwise linear discriminant analysis, k-nearest neighbors, and penalized logistic regression models. RESULTS: A total of 1085 spine radiology reports were included in our analysis. Our dataset included 483 compression, 401 burst, 103 translational/rotational, and 98 distraction fractures. A total of 103 reports had documented an injury of the posterior ligamentous complex. The overall accuracy of the random forest model for fracture morphology feature detection was 76.96% versus 65.90% in the stepwise linear discriminant analysis, 50.69% in the k-nearest neighbors, and 62.67% in the penalized logistic regression. The overall accuracy to detect posterior ligamentous complex integrity was highest in the random forest model at 83.41%. Our random forest model was implemented in the backend of a web application in which users can dictate reports and have TLICS features automatically extracted. CONCLUSIONS: We have developed a machine learning NLP model for extracting TLICS features from radiology reports, which we deployed in a web application that can be integrated into clinical practice.

Automatic uncovering of patient primary concerns in portal messages using a fusion framework of pretrained language models.

OBJECTIVES: The surge in patient portal messages (PPMs) with increasing needs and workloads for efficient PPM triage in healthcare settings has spurred the exploration of AI-driven solutions to streamline the healthcare workflow processes, ensuring timely responses to patients to satisfy their healthcare needs. However, there has been less focus on isolating and understanding patient primary concerns in PPMs-a practice which holds the potential to yield more nuanced insights and enhances the quality of healthcare delivery and patient-centered care. MATERIALS AND METHODS: We propose a fusion framework to leverage pretrained language models (LMs) with different language advantages via a Convolution Neural Network for precise identification of patient primary concerns via multi-class classification. We examined 3 traditional machine learning models, 9 BERT-based language models, 6 fusion models, and 2 ensemble models. RESULTS: The outcomes of our experimentation underscore the superior performance achieved by BERT-based models in comparison to traditional machine learning models. Remarkably, our fusion model emerges as the top-performing solution, delivering a notably improved accuracy score of 77.67 ± 2.74% and an F1 score of 74.37 ± 3.70% in macro-average. DISCUSSION: This study highlights the feasibility and effectiveness of multi-class classification for patient primary concern detection and the proposed fusion framework for enhancing primary concern detection. CONCLUSIONS: The use of multi-class classification enhanced by a fusion of multiple pretrained LMs not only improves the accuracy and efficiency of patient primary concern identification in PPMs but also aids in managing the rising volume of PPMs in healthcare, ensuring critical patient communications are addressed promptly and accurately.

Considerations for Quality Control Monitoring of Machine Learning Models in Clinical Practice.

Integrating machine learning (ML) models into clinical practice presents a challenge of maintaining their efficacy over time. While existing literature offers valuable strategies for detecting declining model performance, there is a need to document the broader challenges and solutions associated with the real-world development and integration of model monitoring solutions. This work details the development and use of a platform for monitoring the performance of a production-level ML model operating in Mayo Clinic. In this paper, we aimed to provide a series of considerations and guidelines necessary for integrating such a platform into a team's technical infrastructure and workflow. We have documented our experiences with this integration process, discussed the broader challenges encountered with real-world implementation and maintenance, and included the source code for the platform. Our monitoring platform was built as an R shiny application, developed and implemented over the course of 6 months. The platform has been used and maintained for 2 years and is still in use as of July 2023. The considerations necessary for the implementation of the monitoring platform center around 4 pillars: feasibility (what resources can be used for platform development?); design (through what statistics or models will the model be monitored, and how will these results be efficiently displayed to the end user?); implementation (how will this platform be built, and where will it exist within the IT ecosystem?); and policy (based on monitoring feedback, when and what actions will be taken to fix problems, and how will these problems be translated to clinical staff?). While much of the literature surrounding ML performance monitoring emphasizes methodological approaches for capturing changes in performance, there remains a battery of other challenges and considerations that must be addressed for successful real-world implementation.