Electronic Health Record Dataset Research Articles

Identifying factors associated with persistent opioid use after total joint arthroplasty: A retrospective review.

To identify predictors of persistent opioid use in opioid-naïve individuals undergoing total joint arthroplasty. Retrospective cohort study. Maine Health System. Opioid-naïve patients who underwent at least one total joint arthroplasty (knee, hip, or shoulder) between 2015 and 2020. Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression was used to create a predictive model for persistent opioid use after surgery from a US Electronic Health Record dataset in the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) format. 75% of the data was used to build the LASSO model using 10-fold cross-validation and 25% of the data was used to determine the optimal probability threshold for predicting the binary outcome. Out of 6432 patients, 12.3% (792) were identified as having persistent opioid use across combined total joint arthroplasties defined as at least one opioid prescription between 90 days and one year after surgery. Patients with persistent opioid use were more likely to be current smokers (OR 1.65), use antidepressants (OR 1.76), or have a diagnosis of post-traumatic stress disorder (OR 2.07), or a substance related disorder (OR 1.69). Other factors associated with persistent opioid use included back pain (OR 1.43), dementia (OR 1.65), and BMI over 40 (OR 2.50). The probability of persistent opioid use was not associated with age, sex, or ethnicity. This predictive model for persistent opioid use after total joint arthroplasty shows promise as an evidence-based, validated, and standardized tool for identifying high-risk patients before surgery in order to target strategies and interventions to reduce the reliance on opioids for post-operative pain control.

Safety of abdominal paracentesis in hospitalised patients receiving uninterrupted therapeutic or prophylactic anticoagulants.

Abdominal paracentesis is a frequently performed procedure in hospitalised patients with ascites. Concurrently, most hospitalised adult patients receive pharmacologic anticoagulation, either for therapeutic purposes or prophylactically to prevent venous thromboembolism. Despite this, minimal evidence exists to guide management of anticoagulant therapy pre- and post-paracentesis. The authors aimed to investigate the safety of abdominal paracentesis in hospitalised patients receiving therapeutic or prophylactic anticoagulation, including in patients for whom these medications were withheld periprocedurally. TriNetX, an electronic health record data set, was queried to identify patients between the ages of 18 and 80 years who received an abdominal paracentesis while hospitalised at the authors' institution between September 2017 and June 2022. Patients receiving prophylactic anticoagulation (137), therapeutic anticoagulation (74) and no anticoagulation because of coagulopathy or thrombocytopenia (15) were compared. Rates of withholding anticoagulation, performing service, pre- and post-paracentesis haemoglobin, bleeding complications, thrombotic complications and need for red blood cell transfusion were analysed. Procedure-related bleeding complications occurred in two (1.4%) patients in the prophylactic group and 0 (0%) patients in the therapeutic group (P = 0.54). No thrombotic complications occurred. Rates of red blood cell transfusions post-paracentesis were similar between groups. Analysis of secondary end-points identified significant differences in rates of withholding anticoagulation and mean change in haemoglobin between performing services. Performance of abdominal paracentesis in patients receiving therapeutic or prophylactic anticoagulation appears to be safe regardless of whether anticoagulation was interrupted periprocedurally, with low rates of bleeding complications, thrombotic complications or need for red blood cell transfusions post-paracentesis.

Quantitatively assessing the impact of the quality of SNOMED CT subtype hierarchy on cohort queries.

SNOMED CT provides a standardized terminology for clinical concepts, allowing cohort queries over heterogeneous clinical data including Electronic Health Records (EHRs). While it is intuitive that missing and inaccurate subtype (or is-a) relations in SNOMED CT reduce the recall and precision of cohort queries, the extent of these impacts has not been formally assessed. This study fills this gap by developing quantitative metrics to measure these impacts and performing statistical analysis on their significance. We used the Optum de-identified COVID-19 Electronic Health Record dataset. We defined micro-averaged and macro-averaged recall and precision metrics to assess the impact of missing and inaccurate is-a relations on cohort queries. Both practical and simulated analyses were performed. Practical analyses involved 407 missing and 48 inaccurate is-a relations confirmed by domain experts, with statistical testing using Wilcoxon signed-rank tests. Simulated analyses used two random sets of 400 is-a relations to simulate missing and inaccurate is-a relations. Wilcoxon signed-rank tests from both practical and simulated analyses (P-values < .001) showed that missing is-a relations significantly reduced the micro- and macro-averaged recall, and inaccurate is-a relations significantly reduced the micro- and macro-averaged precision. The introduced impact metrics can assist SNOMED CT maintainers in prioritizing critical hierarchical defects for quality enhancement. These metrics are generally applicable for assessing the quality impact of a terminology's subtype hierarchy on its cohort query applications. Our results indicate a significant impact of missing and inaccurate is-a relations in SNOMED CT on the recall and precision of cohort queries. Our work highlights the importance of high-quality terminology hierarchy for cohort queries over EHR data and provides valuable insights for prioritizing quality improvements of SNOMED CT's hierarchy.

Automated Electronic Health Record Data Extraction and Curation Using ExtractEHR.

Although the potential transformative effect of electronic health record (EHR) data on clinical research in adult patient populations has been very extensively discussed, the effect on pediatric oncology research has been limited. Multiple factors contribute to this more limited effect, including the paucity of pediatric cancer cases in commercial EHR-derived cancer data sets and phenotypic case identification challenges in pediatric federated EHR data. The ExtractEHR software package was initially developed as a tool to improve clinical trial adverse event reporting but has expanded its use cases to include the development of multisite EHR data sets and the support of cancer cohorts. ExtractEHR enables customized, automated data extraction from the EHR that, when implemented across multiple hospitals, can create pediatric cancer EHR data sets to address a very wide range of research questions in pediatric oncology. After ExtractEHR data acquisition, EHR data can be cleaned and graded using CleanEHR and GradeEHR, companion software packages. ExtractEHR has been installed at four leading pediatric institutions: Children's Healthcare of Atlanta, Children's Hospital of Philadelphia, Texas Children's Hospital, and Seattle Children's Hospital. ExtractEHR has supported multiple use cases, including five clinical epidemiology studies, multicenter clinical trials, and cancer cohort assembly. Work is ongoing to develop Fast Health care Interoperability Resources ExtractEHR and implement other sustainability and scalability enhancements.

Who transitions to bipolar disorder? A comparison of major depressive disorder, anxiety, and ADHD

BackgroundDiagnostic delays in Bipolar Disorder (BD) are common and can contribute to worse outcomes. While most studies focus on depression as a primary precursor, both anxiety and attention deficit disorders are also frequent initial diagnoses. In the current study, we utilized a large, diverse electronic health record (EHR) dataset to quantify the rates and correlates of conversion to BD from these major precursor diagnoses. MethodsOur study analyzed a comprehensive ten-year EHR dataset from Johns Hopkins Medicine, a diverse urban medical center, to assess and compare the rates and correlates of conversion to BD from Major Depressive Disorder (MDD), anxiety disorders, and ADHD. Risk factors for transition were assessed as time-varying variables in proportional hazards models. ResultsOf the 21,341 patients initially included, 1232 later transitioned to a diagnosis of BD. Adjusted-one-year conversion rates for patients with MDD, anxiety disorders, and ADHD were 4.2 %, 3.4 %, and 4.0 %, respectively, with ten-year rates at 11.4 %, 9.4 %, and 10.9 %, respectively. Age (19–29 years), treatment setting (emergency and inpatient), and psychotropic medications were associated with conversion to BD across all precursor diagnoses. Severe and psychotic forms of MDD were among the strongest risk factors for transitioning to BD. Although transition rates were lower in children, risk factors were similar, with the exception of ADHD, which showed a stronger association with BD conversion in adults. ConclusionsThe highest risk of transitioning to BD was observed in patients initially diagnosed with MDD, though significant risk was also noted in those with initial diagnoses of anxiety disorders and adult ADHD.

Targeted Development and Validation of Clinical Prediction Models in Secondary Care Settings: Opportunities and Challenges for Electronic Health Record Data.

Before deploying a clinical prediction model (CPM) in clinical practice, its performance needs to be demonstrated in the population of intended use. This is also called "targeted validation." Many CPMs developed in tertiary settings may be most useful in secondary care, where the patient case mix is broad and practitioners need to triage patients efficiently. However, since structured or rich datasets of sufficient quality from secondary to assess the performance of a CPM are scarce, a validation gap exists that hampers the implementation of CPMs in secondary care settings. In this viewpoint, we highlight the importance of targeted validation and the use of CPMs in secondary care settings and discuss the potential and challenges of using electronic health record (EHR) data to overcome the existing validation gap. The introduction of software applications for text mining of EHRs allows the generation of structured "big" datasets, but the imperfection of EHRs as a research database requires careful validation of data quality. When using EHR data for the development and validation of CPMs, in addition to widely accepted checklists, we propose considering three additional practical steps: (1) involve a local EHR expert (clinician or nurse) in the data extraction process, (2) perform validity checks on the generated datasets, and (3) provide metadata on how variables were constructed from EHRs. These steps help to generate EHR datasets that are statistically powerful, of sufficient quality and replicable, and enable targeted development and validation of CPMs in secondary care settings. This approach can fill a major gap in prediction modeling research and appropriately advance CPMs into clinical practice.

Conceptualizing bias in EHR data: A case study in performance disparities by demographic subgroups for a pediatric obesity incidence classifier.

Electronic Health Records (EHRs) are increasingly used to develop machine learning models in predictive medicine. There has been limited research on utilizing machine learning methods to predict childhood obesity and related disparities in classifier performance among vulnerable patient subpopulations. In this work, classification models are developed to recognize pediatric obesity using temporal condition patterns obtained from patient EHR data in a U.S. study population. We trained four machine learning algorithms (Logistic Regression, Random Forest, Gradient Boosted Trees, and Neural Networks) to classify cases and controls as obesity positive or negative, and optimized hyperparameter settings through a bootstrapping methodology. To assess the classifiers for bias, we studied model performance by population subgroups then used permutation analysis to identify the most predictive features for each model and the demographic characteristics of patients with these features. Mean AUC-ROC values were consistent across classifiers, ranging from 0.72-0.80. Some evidence of bias was identified, although this was through the models performing better for minority subgroups (African Americans and patients enrolled in Medicaid). Permutation analysis revealed that patients from vulnerable population subgroups were over-represented among patients with the most predictive diagnostic patterns. We hypothesize that our models performed better on under-represented groups because the features more strongly associated with obesity were more commonly observed among minority patients. These findings highlight the complex ways that bias may arise in machine learning models and can be incorporated into future research to develop a thorough analytical approach to identify and mitigate bias that may arise from features and within EHR datasets when developing more equitable models.

Implications of mappings between International Classification of Diseases clinical diagnosis codes and Human Phenotype Ontology terms.

Integrating electronic health record (EHR) data with other resources is essential in rare disease research due to low disease prevalence. Such integration is dependent on the alignment of ontologies used for data annotation. The international classification of diseases (ICD) is used to annotate clinical diagnoses, while the human phenotype ontology (HPO) is used to annotate phenotypes. Although these ontologies overlap in the biomedical entities they describe, the extent to which they are interoperable is unknown. We investigate how well aligned these ontologies are and whether such alignments facilitate EHR data integration. We conducted an empirical analysis of the coverage of mappings between ICD and HPO. We interpret this mapping coverage as a proxy for how easily clinical data can be integrated with research ontologies such as HPO. We quantify how exhaustively ICD codes are mapped to HPO by analyzing mappings in the unified medical language system (UMLS) Metathesaurus. We analyze the proportion of ICD codes mapped to HPO within a real-world EHR dataset. Our analysis revealed that only 2.2% of ICD codes have direct mappings to HPO in UMLS. Within our EHR dataset, less than 50% of ICD codes have mappings to HPO terms. ICD codes that are used frequently in EHR data tend to have mappings to HPO; ICD codes that represent rarer medical conditions are seldom mapped. We find that interoperability between ICD and HPO via UMLS is limited. While other mapping sources could be incorporated, there are no established conventions for what resources should be used to complement UMLS.

B-137 Sepsis risk prediction using machine learning (ML) and routine blood markers up to 1 week before emergency admission

Abstract Background Sepsis is a major global health concern, responsible for approximately 50 million cases and 11 million deaths annually, representing 20% of global mortality. It is the leading cause of hospital readmissions and deaths, with associated costs exceeding $38 billion per year. Alarmingly, 80% of sepsis cases occur outside the hospital, and 70% of patients visit an outpatient facility within a week of admission. Early diagnosis and appropriate treatment could prevent 80% of sepsis-related deaths. Our study demonstrates that machine learning models hold significant promise in identifying sepsis risks using routine blood tests up to 1 week before admission. Methods For our study, we utilized a dataset of over 25,000 patient records, including both sepsis and non-sepsis cases, sourced from MIMIC-IV (version 2.2), a comprehensive electronic health record dataset from Beth Israel Deaconess Medical Center, Boston, MA. We employed a gradient boosted model with 300 trees, a maximal depth of 30 layers, and gain ratio as the criterion for attribute selection. The model's performance was evaluated using 10-fold cross-validation, demonstrating optimal performance. Input parameters included age, gender, and results of routine blood markers such as complete blood counts, differential counts, comprehensive metabolic panels, and lipid panels recorded up to 1 week before sepsis diagnosis. Results Our model achieved remarkable predictive accuracy for sepsis risk, with an area under the receiver operating characteristic curve (AUC) of 0.99 and accuracy of 0.99. Notably, the model achieved 1.0 sensitivity, 0.99 specificity, 0.99 positive predictive value, 1.0 negative predictive value, 0.99 F measure, and 0.99 Youden index. Calcium, protein, liver enzymes, hematocrit, white blood cells, and cholesterol were identified as key contributors to sepsis risk prediction. Importantly, prediction accuracy remained consistent up to 1 week prior to diagnosis. Conclusions In conclusion, dysregulated intracellular calcium handling during sepsis is associated with an intensified inflammatory response, cellular death, and subsequent organ dysfunction. Low serum protein levels, particularly albumin, are linked to poor outcomes in sepsis. Recent studies have identified liver dysfunction as an early event in sepsis, underscoring its significance in the disease process. Elevated bilirubin levels are associated with an increased risk of mortality. Low hematocrit (HCT) levels are linked to higher 30-day mortality and serve as an independent prognostic biomarker in sepsis. Monocytes play a significant role in sepsis pathophysiology, undergoing various changes including preserved phagocytic activity, increased reactive oxygen species and nitric oxide generation, and decreased production of inflammatory cytokines such as IL-6 and TNF-alpha. Septic patients often exhibit lymphopenia, affecting CD4 and CD8 T cells, B cells, and natural killer (NK) cells. Hypocholesterolemia is common in sepsis and serves as an important early prognostic factor, with low plasma concentrations associated with poor outcomes. Since every hour of delayed treatment increases the risk of death by 4%-9%, changes in routine blood markers detected by machine learning models can provide early indications of impending sepsis up to 1 week before diagnosis.

Risk Factors for Empiric Treatment Failure in US Female Outpatients with Uncomplicated Urinary Tract Infection: an Observational Study.

Treatment failure (TF) in uncomplicated urinary tract infection (uUTI) increases disease burden and risk of antimicrobial resistance. Identification of risk factors for TF could inform empiric treatment decisions and reduce suboptimal outcomes. To evaluate the incidence of TF to empirically prescribed oral antibiotics and identify risk factors for TF in females with uUTI in the United States (US). This retrospective cohort study used Optum's de-identified Electronic Health Record dataset (January 2017-September 2022). Eligible female patients aged ≥ 12years had ≥ 1 diagnosis of urinary tract infection (UTI) in an outpatient ambulatory/emergency department (ED) setting, ≥ 1 empiric oral antibiotic prescription, and no evidence of complicated UTI (cUTI). TF was defined as having a new/repeat oral antibiotic prescription, IV antibiotic administration or acute UTI diagnosis ≤ 28days following initial empiric oral antibiotic prescription​. Risk factors of TF were selected using LASSO and reported using adjusted risk ratios (aRR) and 95% CIs. Of 376,004 patients with uUTI, 62,873 (16.7%) experienced TF. Incidence of TF was highest in patients with history of antibiotic TF (33.9%) or fosfomycin prescription (30.1%). Significant risk factors of TF included ≥ 3 prior antibiotic prescriptions (aRR [95% CI]: 1.60 [1.56-1.64]); fosfomycin prescription (1.60 [1.38-1.86]); uUTI diagnosis in ED (1.49 [1.46-1.52]), Southern US residence (1.37 [1.35-1.40]), age ≥ 75years (1.35 [1.29-1.41]), recurrent UTI (1.12 [1.10-1.14]) and obesity (1.06 [1.04-1.08]). Incidence of TF to empirically prescribed oral antibiotics for uUTI is considerable. Prior infections requiring antibiotic prescription and location of care are key risk factors for TF in female outpatients with uUTI. Knowledge of these TF risk factors can inform shared-decision making and supplement existing guidance on uUTI treatment.

PheW2P2V: a phenome-wide prediction framework with weighted patient representations using electronic health records.

Electronic health records (EHRs) provide opportunities for the development of computable predictive tools. Conventional machine learning methods and deep learning methods have been widely used for this task, with the approach of usually designing one tool for one clinical outcome. Here we developed PheW2P2V, a Phenome-Wide prediction framework using Weighted Patient Vectors. PheW2P2V conducts tailored predictions for phenome-wide phenotypes using numeric representations of patients' past medical records weighted based on their similarities with individual phenotypes. PheW2P2V defines clinical disease phenotypes using Phecode mapping based on International Classification of Disease codes, which reduces redundancy and case-control misclassification in real-life EHR datasets. Through upweighting medical records of patients that are more relevant to a phenotype of interest in calculating patient vectors, PheW2P2V achieves tailored incidence risk prediction of a phenotype. The calculation of weighted patient vectors is computationally efficient, and the weighting mechanism ensures tailored predictions across the phenome. We evaluated prediction performance of PheW2P2V and baseline methods with simulation studies and clinical applications using the MIMIC-III database. Across 942 phenome-wide predictions using the MIMIC-III database, PheW2P2V has median area under the receiver operating characteristic curve (AUC-ROC) 0.74 (baseline methods have values ≤0.72), median max F1-score 0.20 (baseline methods have values ≤0.19), and median area under the precision-recall curve (AUC-PR) 0.10 (baseline methods have values ≤0.10). PheW2P2V can predict phenotypes efficiently by using medical concept embeddings and upweighting relevant past medical histories. By leveraging both labeled and unlabeled data, PheW2P2V reduces overfitting and improves predictions for rare phenotypes, making it a useful screening tool for early diagnosis of high-risk conditions, though further research is needed to assess the transferability of embeddings across different databases. PheW2P2V is fast, flexible, and has superior prediction performance for many clinical disease phenotypes across the phenome of the MIMIC-III database compared to that of several popular baseline methods.

Criticality of Nursing Care for Patients With Alzheimer's Disease in the ICU: Insights From MIMIC III Dataset.

Alzheimer's disease (AD) patients admitted to intensive care units (ICUs) exhibit varying survival outcomes due to the unique challenges in managing AD patients. Stratifying patient mortality risk and understanding the criticality of nursing care are important to improve the clinical outcomes of AD patients. This study aimed to leverage machine learning (ML) and electronic health records (EHRs) only consisting of demographics, disease history, and routine lab tests, with a focus on nursing care, to facilitate the optimization of nursing practices for AD patients. We utilized Medical Information Mart for Intensive Care III, an open-source EHR dataset, and AD patients were identified based on the International Classification of Diseases, Ninth Revision codes. From a cohort of 453 patients, a total of 60 features, encompassing demographics, laboratory tests, disease history, and number of nursing events, were extracted. ML models, including XGBoost, random forest, logistic regression, and multi-layer perceptron, were trained to predict the 30-day mortality risk. In addition, the influence of nursing care was analyzed in terms of feature importance using values calculated from both the inherent XGBoost module and the SHapley Additive exPlanations (SHAP) library. XGBoost emerged as the lead model with a high accuracy of 0.730, area under the curve (AUC) of 0.750, sensitivity of 0.688, and specificity of 0.740. Feature importance analyses using inherent XGBoost module or SHAP both indicated the number of nursing care within 14 days post-admission as an important denominator for 30-day mortality risk. When nursing care events were excluded as a feature, stratifying patient mortality risk was also possible but the model's AUC of receiver operating characteristic curve was reduced to 0.68. Nursing care plays a pivotal role in the survival outcomes of AD patients in ICUs. ML models can be effectively employed to predict mortality risks and underscore the importance of specific features, including nursing care, in patient outcomes. Early identification of high-risk AD patients can aid in prioritizing intensive nursing care, potentially improving survival rates.

Integration of estimated regional gene expression with neuroimaging and clinical phenotypes at biobank scale.

An understanding of human brain individuality requires the integration of data on brain organization across people and brain regions, molecular and systems scales, as well as healthy and clinical states. Here, we help advance this understanding by leveraging methods from computational genomics to integrate large-scale genomic, transcriptomic, neuroimaging, and electronic-health record data sets. We estimated genetically regulated gene expression (gr-expression) of 18,647 genes, across 10 cortical and subcortical regions of 45,549 people from the UK Biobank. First, we showed that patterns of estimated gr-expression reflect known genetic-ancestry relationships, regional identities, as well as inter-regional correlation structure of directly assayed gene expression. Second, we performed transcriptome-wide association studies (TWAS) to discover 1,065 associations between individual variation in gr-expression and gray-matter volumes across people and brain regions. We benchmarked these associations against results from genome-wide association studies (GWAS) of the same sample and found hundreds of novel associations relative to these GWAS. Third, we integrated our results with clinical associations of gr-expression from the Vanderbilt Biobank. This integration allowed us to link genes, via gr-expression, to neuroimaging and clinical phenotypes. Fourth, we identified associations of polygenic gr-expression with structural and functional MRI phenotypes in the Human Connectome Project (HCP), a small neuroimaging-genomic data set with high-quality functional imaging data. Finally, we showed that estimates of gr-expression and magnitudes of TWAS were generally replicable and that the p-values of TWAS were replicable in large samples. Collectively, our results provide a powerful new resource for integrating gr-expression with population genetics of brain organization and disease.

Risk Assessment and Diagnosis Code Prediction from Electronic Health Records (EHR) using deep learning

Knowledge in the medical field is often expressed by distinct and subjective norms. Much research and development efforts have focused on using deep learning algorithms to predict the likelihood of illnesses from EHR in recent years. When it comes to risk prediction, deep learning-based techniques outperform more conventional machine learning models. But nothing in the literature fully accounts for what doctors already know, such as the connections between illnesses and their risk factors. This research examines the use of Multi-layer Perceptron models for the categorisation of diagnoses in electronic health records. The raw data and a modified version of the EHR dataset are used to train two MLPs with distinct topologies. For comparative purposes, a Random Forest is used as a baseline. To phenotype patients using their electronic health records, we provide a deep learning method. Predictive modelling of chronic illnesses is the particular scenario used to verify the suggested model on a real-world EHR data warehouse. Many deep learning applications on EHRs have been effective, and there is still a lot of potential to be realised. It has been discovered that deep learning models can learn from the limited EHR dataset, but not to a level where they outperform the baseline model.

Demers cartogram with rivers

Cartograms serve as representations of geographical and abstract data, employing a value-by-area mapping technique. As a variant of the Dorling cartogram, the Demers cartogram utilizes squares instead of circles to represent regions. This alternative approach allows for a more intuitive comparison of regions, utilizing screen space more efficiently. However, a drawback of the Dorling cartogram and its variants lies in the potential displacement of regions from their original positions, ultimately compromising legibility, readability, and accuracy. To tackle this limitation, we propose a novel hybrid cartogram layout algorithm that incorporates topological elements, such as rivers, into Demers cartograms. The presence of rivers significantly impacts both the layout and visual appearance of the cartograms. Through a user study conducted on an Electronic Health Records (EHR) dataset, we evaluate the efficacy of the proposed hybrid layout algorithm. The obtained results illustrate that this approach successfully retains key aspects of the original cartogram while enhancing legibility, readability, and overall accuracy.

Reliable prediction of childhood obesity using only routinely collected EHRs may be possible

BackgroundEarly identification of children at high risk of obesity can provide clinicians with the information needed to provide targeted lifestyle counseling to high-risk children at a critical time to change the disease course. ObjectivesThis study aimed to develop predictive models of childhood obesity, applying advanced machine learning methods to a large unaugmented electronic health record (EHR) dataset. This work improves on other studies that have (i) relied on data not routinely available in EHRs (like prenatal data), (ii) focused on single-age predictions, or (iii) not been rigorously validated. MethodsA customized sequential deep-learning model to predict the development of obesity was built, using EHR data from 36,191 diverse children aged 0–10 years. The model was evaluated using extensive discrimination, calibration, and utility analysis; and was validated temporally, geographically, and across various subgroups. ResultsOur results are mostly better or comparable to similar studies. Specifically, the model achieved an AUROC above 0.8 in all cases (with most cases around 0.9) for predicting obesity within the next 3 years for children 2–7 years of age. Validation results show the model's robustness and top predictors match important risk factors of obesity. ConclusionsOur model can predict the risk of obesity for young children at multiple time points using only routinely collected EHR data, greatly facilitating its integration into clinical care. Our model can be used as an objective screening tool to provide clinicians with insights into a patient's risk for developing obesity so that early lifestyle counseling can be provided to prevent future obesity in young children.

US Real-World Effectiveness Study of Nirmatrelvir/Ritonavir in Preventing Hospitalization of High-Risk COVID-19 Patients

PurposeWe describe nirmatrelvir/ritonavir (NMV/r) effectiveness in preventing hospitalization among COVID-19 patients at high risk of severe disease. MethodsAn ongoing US population-based observational cohort study with retrospective and prospective collection of national electronic healthcare data collected from the US Optum® deidentified COVID-19 Electronic Health Record dataset during December 22, 2021−July 20, 2022. Participants were ≥12 years old; had a positive SARS-CoV-2 test, COVID-19 diagnosis, or NMV/r prescription; and were at high risk of severe COVID-19 based on demographic/clinical characteristics. Potential confounders between groups were balanced using propensity score matching. Immortal time bias was addressed. Hospitalization rates within 30 days from COVID-19 diagnosis were evaluated. Sensitivity analyses included 15-day hospitalization, chart review to investigate incidental hospitalization effects, and exclusion of patients identified as having COVID-19 based on NMV/r prescription alone. Outcomes were also evaluated by race, age, and COVID-19 vaccine status. FindingsOverall, 12,440 and 234,123 patients were included in the NMV/r and non-NMV/r groups, respectively. After propensity score matching, baseline characteristics were well balanced across groups (NMV/r, n = 12,439; non-NMV/r, n = 36,490). Incidence of hospitalization (95% CI) within 30 days was 0.90% (0.74%−1.08%) for the NMV/r group and 5.91% (5.67%−6.16%) for the non-NMV/r group, with relative risk (95% CI) of 0.15 (0.13−0.18; 85% risk reduction). NMV/r was comparably effective in Black patients (relative risk, 0.19 [0.10−0.34]; 81% risk reduction). Sensitivity analyses supported the main outcomes. ImplicationsReal-world NMV/r effectiveness against hospitalization during Omicron predominance among COVID-19 patients at high risk of severe disease supports demonstrated clinical trial efficacy. Black patients underutilized NMV/r despite high effectiveness.

Preliminary Evaluation of Fine-Tuning the OpenDeLD Deidentification Pipeline Across Multi-Center Corpora.

Automatic deidentification of Electronic Health Records (EHR) is a crucial step in secondary usage for biomedical research. This study introduces evaluation of an intricate hybrid deidentification strategy to enhance patient privacy in secondary usage of EHR. Specifically, this study focuses on assessing automatic deidentification using OpenDeID pipeline across diverse corpora for safeguarding sensitive information within EHR datasets by incorporating diverse corpora. Three distinct corpora were utilized: the OpenDeID v2 corpus containing pathology reports from Australian hospitals, the 2014 i2b2/UTHealth deidentification corpus with clinical narratives from the USA, and the 2016 CEGS N-GRID identification corpus comprising psychiatric notes. The OpenDeID pipeline employs a hybrid approach based on deep learning and contextual rules. Pre-processing steps involved harmonizing and addressing encoding and format issues. Precision, Recall, F-measure metrics were used to assess the performance. The evaluation metrics demonstrated the superior performance of the Discharge Summary BioBERT model. Trained on three corpora with a total of 4,038 reports, the best performing model exhibited robust deidentification capabilities when applied to EHR. It achieved impressive micro-averaged F1-scores of 0.9248 and 0.9692 for strict and relaxed settings, respectively. These results offer valuable insights into the model's efficacy and its potential role in safeguarding patient privacy in secondary usage of EHR.

Graph and text multi-modal representation learning with momentum distillation on Electronic Health Records

The emergence and widespread adoption of electronic health records (EHR) in modern healthcare systems has generated a large amount of data with the potential to significantly improve patient outcomes. However, extracting meaningful insights from EHR is challenging due to their extensive volume and inherent complexity. Leveraging diverse modalities of data enhances the exploration of patient data representation. Nevertheless, current research on multi-modal relationships between concealed graph structures within medical codes and unstructured text overlooks inherent disparities and inconsistencies. To address the significant gap, this study introduces a pretraining approach named graph and text multi-modal representation learning with momentum distillation on Electronic Health Records (GTMMRL). This approach tackles the challenge of noisy and unreliable labels using a large open-source EHR dataset, MIMIC-III, for pretraining. Five carefully selected proxy tasks are employed, each guiding the student model’s learning through pseudo-targets generated by the teacher model, which is an exponential moving average of the student model, thereby mitigating overfitting tendencies. We evaluate the performance of our model in five downstream tasks, each addressing real-world challenges in EHR data analysis. Results showcase that our method achieves state-of-the-art performance in various downstream tasks, highlighting the model’s critical role in enhancing the quality of insights extracted from complex EHR data. The code is available at https://github.com/deepEMR/GTMMRL.

Leveraging Electronic Health Records to Predict the Risk of Acute Kidney Injury after Allogeneic Hematopoietic Cell Transplantation.

The objective of this study is to assess the electronic health records (EHRs), which are potential risk factors for acute kidney injury (AKI) after allogenic hematopoietic cell transplantation (allo-HCT), and to propose a basic dataset and score for the calculation of HCT-acute kidney injury risk (HCT-AKIR). We undertook a retrospective analysis of the EHRs of 312 patients. Pre- and post-transplant factors were assessed, recognizing the following structured entries: laboratory data, encounters, medication, imaging studies, diagnoses, and procedures. Composite variables were used to create patient groups by combining two or more multivariate significant risk factors for AKI. The EHRs dataset and HCT-AKIR score were created based on the multivariate analysis of the composite variables. A multivariate analysis showed that previous CKD and once-impaired pre-transplant kidney function, sepsis, imaging procedures with contrast media, and cumulative length of intensive care unit stay after transplantation were significant risk factors. A new unit-weighted composite score based on the combination of significant risk factors contained in common EHR resources was proposed. Using our novel HCT-AKIR score calculated from the basic EHR dataset could be an easy way to increase awareness of post-transplant AKI and provide risk stratification.