Electronic Health Record Dataset Research Articles

Outcomes of patients with COPD switching from multiple-inhaler to once-daily single-inhaler triple therapy in a real-world primary care setting in England: a retrospective pre-post cohort study

BackgroundCompared with multiple-inhaler triple therapy (MITT), single-inhaler triple therapy (SITT) with fluticasone furoate/umeclidinium/vilanterol (FF/UMEC/VI) demonstrated improved lung function and meaningful improvements in chronic obstructive pulmonary disease (COPD) Assessment Test score....

Generalizability of a Musculoskeletal Therapist Electronic Health Record for Modelling Outcomes to Work-Related Musculoskeletal Disorders.

Electronic Health Records (EHRs) can contain vast amounts of clinical information that could be reused in modelling outcomes of work-related musculoskeletal disorders (WMSDs). Determining the generalizability of an EHR dataset is an important step in determining the appropriateness of its reuse. The study aims to describe the EHR dataset used by occupational musculoskeletal therapists and determine whether the EHR dataset is generalizable to the Australian workers' population and injury characteristics seen in workers' compensation claims. Variables were considered if they were associated with outcomes of WMSDs and variables data were available. Completeness and external validity assessment analysed frequency distributions, percentage of records and confidence intervals. There were 48,434 patient care plans across 10 industries from 2014 to 2021. The EHR collects information related to clinical interventions, health and psychosocial factors, job demands, work accommodations as well as workplace culture, which have all been shown to be valuable variables in determining outcomes to WMSDs. Distributions of age, duration of employment, gender and region of birth were mostly similar to the Australian workforce. Upper limb WMSDs were higher in the EHR compared to workers' compensation claims and diagnoses were similar. The study shows the EHR has strong potential to be used for further research into WMSDs as it has a similar population to the Australian workforce, manufacturing industry and workers' compensation claims. It contains many variables that may be relevant in modelling outcomes to WMSDs that are not typically available in existing datasets.

Patient Characteristics, Management, and Outcomes in a Novel Cohort of Primary Hyperparathyroidism

Abstract Context Primary hyperparathyroidism (PHPT) increases the risk of bone loss, debilitating fractures, kidney stones, impaired renal function, and neurocognitive symptoms. Studies describing the natural history of PHPT have been limited to small samples, single-institutions, or specific populations. Objective We assessed the natural history of PHPT through a large, diverse national cohort from an electronic health record (EHR) dataset representing over 100 million patients. Methods The TriNetX database was queried for adult patients with PHPT. We extracted demographics, comorbidities, and longitudinal biochemistries. Primary outcomes included major osteoporotic fracture (MOF) and chronic kidney disease (CKD). Outcomes were stratified by treatment strategy (surgical parathyroidectomy (PTX) vs non-surgical) and age. Results Among 49,119 patients with PHPT, 23.6% were treated surgically at a median of 2.0 years post-diagnosis. At diagnosis, median age was 72 years, 73.2% were female, and median calcium level was 10.9 mg/dL. Black and older patients underwent PTX less frequently than white and younger patients. MOF 10-year incidence was 4.5% (PTX) and 6.7% (non-surgical), with median 2.7-year delay with PTX compared to non-surgical. PTX-associated MOF absolute risk reduction (ARR) was 0.54% (age < 65) and 2.42% (age ≥ 65). CKD 10-year incidence was 22.1% (PTX) and 33.5% (non-surgical), with median 2.8-year delay with PTX. PTX-associated CKD ARR was 17.1% (age < 65) and 8.28% (age ≥ 65). Conclusions We report one of the largest, representative, population-based natural histories of PHPT with different management strategies. A minority of patients underwent PTX, especially in older age. Patients managed surgically had lower incidence of fracture and CKD, and older patients experienced differential benefit.

Dynamic Risk Stratification of worsening heart failure using a Deep learning enabled Implanted Ambulatory Single lead ECG

Abstract Background and Aims Implantable loop recorders (ILRs) provide continuous single-lead ambulatory electrocardiogram (aECG) monitoring. It is unknown whether these aECGs could be used to identify worsening heart failure. Methods and results We linked ILR aECG from Medtronic device database to the LVEF measurements in Optum® de-identified electronic health record dataset. We trained an AI algorithm (aECG-CNN) on a dataset of 35,741 aECGs from 2247 patients to identify left ventricular ejection fraction (LVEF) ≤ 40% and assessed its performance using the area under the receiver operating characteristic curve (AUROC). aECG-CNN was then used to identify patients with increasing risk of heart-failure hospitalization in a real-world cohort of 909 patients with prior heart failure diagnosis. This dataset provided 12,467 follow up monthly evaluations, with 201 heart failure hospitalizations. For every month, time series features from these predictions were used to categorize patients into high and low risk groups and predict heart failure hospitalization in the next month. The risk of heart-failure hospitalization in the next 30 days was significantly higher in the cohort that aECG-CNN identified as high risk (hazard ratio 1·89; 95% confidence interval 1·28-2·79; p = 0·001) compared to low risk, even after adjusting patient demographics. (Hazard ratio 1·88, 1.27 to 2·79 p = 0·002). Conclusion An AI algorithm trained to detect LVEF ≤40% using ILR aECGs can also readily identify patients at increased risk for HF hospitalizations by monitoring changes in the probability of heart failure over 30 days.

Artificial intelligence-assisted automated heart failure detection and classification from electronic health records.

Electronic health records (EHR) linked to Digital Imaging and Communications in Medicine (DICOM), biological specimens, and deep learning (DL) algorithms could potentially improve patient care through automated case detection and surveillance. We hypothesized that by applying keyword searches to routinely stored EHR, in conjunction with AI-powered automated reading of DICOM echocardiography images and analysing biomarkers from routinely stored plasma samples, we were able to identify heart failure (HF) patients. We used EHR data between 1993 and 2021 from Tayside and Fife (~20% of the Scottish population). We implemented a keyword search strategy complemented by filtering based on International Classification of Diseases (ICD) codes and prescription data to EHR data set. We then applied DL for the automated interpretation of echocardiographic DICOM images. These methods were then integrated with the analysis of routinely stored plasma samples to identify and categorize patients into HF with reduced ejection fraction (HFrEF), HF with preserved ejection fraction (HFpEF), and controls without HF. The final diagnosis was verified through a manual review of medical records, measured natriuretic peptides in stored blood samples, and by comparing clinical outcomes among groups. In our study, we selected the patient cohort through an algorithmic workflow. This process started with 60850 EHR data and resulted in a final cohort of 578 patients, divided into 186 controls, 236 with HFpEF, and 156 with HFrEF, after excluding individuals with mismatched data or significant valvular heart disease. The analysis of baseline characteristics revealed that compared with controls, patients with HFrEF and HFpEF were generally older, had higher BMI, and showed a greater prevalence of co-morbidities such as diabetes, COPD, and CKD. Echocardiographic analysis, enhanced by DL, provided high coverage, and detailed insights into cardiac function, showing significant differences in parameters such as left ventricular diameter, ejection fraction, and myocardial strain among the groups. Clinical outcomes highlighted a higher risk of hospitalization and mortality for HF patients compared with controls, with particularly elevated risk ratios for both HFrEF and HFpEF groups. The concordance between the algorithmic selection of patients and manual validation demonstrated high accuracy, supporting the effectiveness of our approach in identifying and classifying HF subtypes, which could significantly impact future HF diagnosis and management strategies. Our study highlights the feasibility of combining keyword searches in EHR, DL automated echocardiographic interpretation, and biobank resources to identify HF subtypes.

Electronic health record signatures identify undiagnosed patients with common variable immunodeficiency disease.

Human inborn errors of immunity include rare disorders entailing functional and quantitative antibody deficiencies due to impaired B cells called the common variable immunodeficiency (CVID) phenotype. Patients with CVID face delayed diagnoses and treatments for 5 to 15 years after symptom onset because the disorders are rare (prevalence of ~1/25,000), and there is extensive heterogeneity in CVID phenotypes, ranging from infections to autoimmunity to inflammatory conditions, overlapping with other more common disorders. The prolonged diagnostic odyssey drives excessive system-wide costs before diagnosis. Because there is no single causal mechanism, there are no genetic tests to definitively diagnose CVID. Here, we present PheNet, a machine learning algorithm that identifies patients with CVID from their electronic health records (EHRs). PheNet learns phenotypic patterns from verified CVID cases and uses this knowledge to rank patients by likelihood of having CVID. PheNet could have diagnosed more than half of our patients with CVID 1 or more years earlier than they had been diagnosed. When applied to a large EHR dataset, followed by blinded chart review of the top 100 patients ranked by PheNet, we found that 74% were highly probable to have CVID. We externally validated PheNet using >6 million records from disparate medical systems in California and Tennessee. As artificial intelligence and machine learning make their way into health care, we show that algorithms such as PheNet can offer clinical benefits by expediting the diagnosis of rare diseases.

Use of FOLFOXIRI Plus Bevacizumab and Subsequent Therapies in Metastatic Colorectal Cancer: An Age-Stratified Analysis

BackgroundTreatment recommendations for metastatic colorectal cancer (mCRC) do not differ by age group; nevertheless, aggressive multi-agent chemotherapy comprising FOLFOXIRI+bevacizumab (triplet+bev) is routinely administered in younger patients. This study analyzed real-world data on index triplet+bev use and subsequent systemic therapies. Materials and MethodsThis retrospective, observational cohort study was conducted in patients aged ≥18 years with mCRC, who were initiated on triplet+bev. Data were derived from the Optum® de-identified electronic health record dataset. ResultsOf 36,056 patients, 14%, 36%, and 50% were aged 18–49, 50–64, and ≥65 years, respectively. During the study period (2010–2021), triplet+bev use increased in patients aged 18–49 years (1–4%) but remained at approximately 3% and 1% in patients aged 50–64 and ≥65 years, respectively. Patient demographics and clinical characteristics varied slightly; of patients receiving triplet+bev (n = 921) versus non-triplet+bev (n = 35,132) most were male (57% vs. 52%), resided in the Midwest (54% vs. 49%) and Northeast (18% vs. 14%) US regions, and had secondary malignancies (86% vs. 73%). Following triplet+bev, most patients received subsequent therapies (including continued triplet component therapies; 97%) or subsequent ‘new’ therapies (therapies that did not include any agents comprising triplet+bev; 57%), most frequently EGFR inhibitors (28%) and regorafenib (21%), with a similar trend among all age groups. ConclusionsOverall, this study shows that younger patients with mCRC are more likely to receive first-line triplet+bev. These results also reveal that non-chemotherapy options are often used beyond first-line triplet chemotherapy for patients with mCRC. MicroAbstractWe conducted an analysis of electronic health records to assess FOLFOXIRI+bevacizumab (triplet+bev) use and subsequent therapies in adult patients with metastatic colorectal cancer. Results showed increasing triplet+bev use over time in patients aged <65 years and administration of ‘new’ subsequent therapies (not including any agents comprising triplet+bev) in over half of the study population, consistent with available evidence.

OntoMedRec: Logically-pretrained model-agnostic ontology encoders for medication recommendation

Recommending medications with electronic health records (EHRs) is a challenging task for data-driven clinical decision support systems. Most existing models learnt representations for medical concepts based on EHRs and make recommendations with the learnt representations. However, most medications appear in EHR datasets for limited times (the frequency distribution of medications follows power law distribution), resulting in insufficient learning of their representations of the medications. Medical ontologies are the hierarchical classification systems for medical terms where similar terms will be in the same class on a certain level. In this paper, we propose OntoMedRec, the logically-pretrained and model-agnostic medical Ontology Encoders for Medication Recommendation that addresses data sparsity problem with medical ontologies. We conduct comprehensive experiments on real-world EHR datasets to evaluate the effectiveness of OntoMedRec by integrating it into various existing downstream medication recommendation models. The result shows the integration of OntoMedRec improves the performance of various models in both the entire EHR datasets and the admissions with few-shot medications. We provide the GitHub repository for the source code. (https://github.com/WaicongTam/OntoMedRec)

Self-Supervised Representation Learning on Electronic Health Records with Graph Kernel Infomax

Learning Electronic Health Records (EHRs) representation is a preeminent yet under-discovered research topic. It benefits various clinical decision support applications, e.g., medication outcome prediction or patient similarity search. Current approaches focus on task-specific label supervision on vectorized sequential EHR, which is not applicable to large-scale unsupervised scenarios. Recently, contrastive learning has shown great success in self-supervised representation learning problems. However, complex temporality often degrades the performance. We propose Graph Kernel Infomax, a self-supervised graph kernel learning approach on the graphical representation of EHR, to overcome the previous problems. Unlike the state-of-the-art, we do not change the graph structure to construct augmented views. Instead, we use Kernel Subspace Augmentation to embed nodes into two geometrically different manifold views. The entire framework is trained by contrasting nodes and graph representations on those two manifold views through the commonly used contrastive objectives. Empirically, using publicly available benchmark EHR datasets, our approach yields performance on clinical downstream tasks that exceeds the state-of-the-art. Theoretically, the variation in distance metrics naturally creates different views as data augmentation without changing graph structures. Practically, our method is non-ad hoc and confirms superior performance on commonly used graph benchmark datasets.

MixEHR-SurG: A joint proportional hazard and guided topic model for inferring mortality-associated topics from electronic health records

Survival models can help medical practitioners to evaluate the prognostic importance of clinical variables to patient outcomes such as mortality or hospital readmission and subsequently design personalized treatment regimes. Electronic Health Records (EHRs) hold the promise for large-scale survival analysis based on systematically recorded clinical features for each patient. However, existing survival models either do not scale to high dimensional and multi-modal EHR data or are difficult to interpret. In this study, we present a supervised topic model called MixEHR-SurG to simultaneously integrate heterogeneous EHR data and model survival hazard. Our contributions are three-folds: (1) integrating EHR topic inference with Cox proportional hazards likelihood; (2) integrating patient-specific topic hyperparameters using the PheCode concepts such that each topic can be identified with exactly one PheCode-associated phenotype; (3) multi-modal survival topic inference. This leads to a highly interpretable survival topic model that can infer PheCode-specific phenotype topics associated with patient mortality. We evaluated MixEHR-SurG using a simulated dataset and two real-world EHR datasets: the Quebec Congenital Heart Disease (CHD) data consisting of 8211 subjects with 75,187 outpatient claim records of 1767 unique ICD codes; the MIMIC-III consisting of 1458 subjects with multi-modal EHR records. Compared to the baselines, MixEHR-SurG achieved a superior dynamic AUROC for mortality prediction, with a mean AUROC score of 0.89 in the simulation dataset and a mean AUROC of 0.645 on the CHD dataset. Qualitatively, MixEHR-SurG associates severe cardiac conditions with high mortality risk among the CHD patients after the first heart failure hospitalization and critical brain injuries with increased mortality among the MIMIC-III patients after their ICU discharge. Together, the integration of the Cox proportional hazards model and EHR topic inference in MixEHR-SurG not only leads to competitive mortality prediction but also meaningful phenotype topics for in-depth survival analysis. The software is available at GitHub: https://github.com/li-lab-mcgill/MixEHR-SurG.

AKA-SafeMed: A safe medication recommendation based on attention mechanism and knowledge augmentation

Medication recommendation (MR) focuses on generating a medication combination without adverse drug-drug interactions (DDIs) based on electronic health records (EHRs) in making prescriptions. However, how to capture the temporality in historical visits and concurrence in clinical events are two challenges that determine the outcomes of an MR model. To achieve effective and safe MR with imbalanced EHR data, we propose a medication recommendation model based on the attention mechanism and knowledge augmentation strategy (AKA-SafeMed). Specifically, in the patient representation module, bidirectional long short-term memory (BiLSTM) models are deployed to encode the patient's sequential visits, and a self-attention mechanism is employed to learn the weights of diagnosis and procedure vectors from a patient's historical visits; in the medication generation module, the EHR and DDI graphs are leveraged to fully explore the relationships between pairwise drugs as a knowledge augmentation strategy. Finally, extensive experiments are conducted to evaluate the performance of AKA-SafeMed on public EHR datasets. The experimental results demonstrate that the AKA-SafeMed model achieves superior performance in MR tasks compared with the state-of-the-art baseline models.

Breast cancer (BC) and severe COVID-19 (C-19) outcomes: a matched analysis.

Patients with cancer receiving anticancer treatment have a higher risk of severe COVID-19 (C-19) outcomes. We examine the association between breast cancer (BC), recent treatment (systemic therapy, surgery, radiation), and C-19 outcomes. Retrospective matched cohort study using the Optum® de-identified C-19 Electronic Health Record dataset (2007-2022). Patients with C-19 were categorized into: no cancer, BC with recent treatment, and BC without recent treatment and matched based on age, C-19 diagnosis date, and comorbidity score. We evaluated 30-day mortality, mechanical ventilation, intensive care unit (ICU) stay, and hospitalization. A composite outcome including all outcomes was analyzed. Multivariable logistic regression models were used. 2200 matched triplets (1:1:10) of patients with BC recently treated, BC not recently treated, and no cancer were included. Rates of adverse outcomes improved in 2021 compared to 2020. Compared to patients without cancer, those with BC recently treated had a similar risk of adverse outcomes, while patients with BC not recently treated had a lower risk of ICU stay and hospitalization. Using the composite variable, BC recently treated had similar outcomes (OR = 1.02; 95%CI 0.93-1.11) to patients without cancer, while BC patients not recently treated had better outcomes (OR = 0.66; 95%CI 0.59-0.74). Among patients with BC, chemotherapy within 3months was associated with a higher risk of hospitalization (OR = 2.30; 95%CI 1.76-2.99) and composite outcome (OR = 2.11; 95%CI 1.64-2.72). Patients with BC have a similar risk of adverse C-19 outcomes compared to patients without cancer. Among patients with BC, recent chemotherapy was associated with a higher risk of hospitalization.

Ambulatory Risk Stratification for Worsening Heart Failure in Patients with Reduced and Preserved Ejection Fraction Using Diagnostic Parameters Available in Implantable Cardiac Monitors.

Ambulatory risk stratification for worsening heart failure (HF) using diagnostics measured by insertable cardiac monitors (ICM) may depend on the left ventricular ejection fraction (LVEF). We evaluated risk stratification performance in patients with reduced versus preserved LVEF. ICM patients with a history of HF events (HFEs) were included from the Optum® de-identified Electronic Health Record dataset merged with ICM device-collected data during 2007-2021. ICM measures nighttime heart rate (NHR), heart rate variability (HRV), atrial fibrillation (AF) burden, rate during AF, and activity duration (ACT) daily. Each diagnostic was categorized into high, medium, or low risk using previously defined features. HFEs were HF-related inpatient, observation unit, or emergency department stays with IV diuresis administration. Patients were divided into two cohorts: LVEF ≤ 40% and LVEF > 40%. A marginal Cox proportional hazards model compared HFEs for different risk groups. A total of 1020 ICM patients with 18,383 follow-up months and 301 months with HFEs (1.6%) were included. Monthly evaluations with a high risk were 2.3, 4.2, 5.0, and 4.5 times (p < 0.001 for all) more likely to have HFEs in the next 30 days compared to those with a low risk for AF, ACT, NHR, and HRV, respectively. HFE rates were higher for patients with LVEF > 40% compared to LVEF ≤ 40% (2.0% vs. 1.3%), and the relative risk between high-risk and low-risk for each diagnostic parameter was higher for patients with LVEF ≤ 40%. Diagnostics measured by ICM identified patients at risk for impending HFEs. Patients with preserved LVEF showed a higher absolute risk, and the relative risk between risk groups was higher in patients with reduced LVEF.

Development of Predictive Models to Inform a Novel Risk Categorization Framework for Antibiotic Resistance in E. coli-Causing Uncomplicated Urinary Tract Infection.

In clinical practice, challenges in identifying patients with uncomplicated urinary tract infections (uUTIs) at risk of antibiotic non-susceptibility may lead to inappropriate prescribing and contribute to antibiotic resistance. We developed predictive models to quantify risk of non-susceptibility to four commonly prescribed antibiotic classes for uUTI, identify predictors of non-susceptibility to each class, and construct a corresponding risk categorization framework for non-susceptibility. Eligible females aged ≥12 years with E. coli-caused uUTI were identified from Optum's de-identified Electronic Health Record dataset (10/1/2015‒2/29/2020). Four predictive models were developed to predict non-susceptibility to each antibiotic class and a risk categorization framework was developed to classify patients' isolates as low, moderate, and high risk of non-susceptibility to each antibiotic class. Predictive models were developed among 87487 patients. Key predictors of having a non-susceptible isolate to ≥3 antibiotic classes included number of previous UTI episodes, prior β-lactam non-susceptibility, prior fluoroquinolone treatment, census bureau region, and race. The risk categorization framework classified 8.1%, 14.4%, 17.4%, and 6.3% of patients as having isolates at high risk of non-susceptibility to nitrofurantoin, trimethoprim-sulfamethoxazole, β-lactams, and fluoroquinolones, respectively. Across classes, the proportion of patients categorized as having high-risk isolates was 3-12 folds higher among patients with non-susceptible isolates versus susceptible isolates. Our predictive models highlight factors that increase risk of non-susceptibility to antibiotics for uUTIs, while the risk categorization framework contextualizes risk of non-susceptibility to these treatments. Our findings provide valuable insight to clinicians treating uUTIs and may help inform empiric prescribing in this population.

Measuring the Impact of Primary Care Team Composition on Patient Activation Utilizing Electronic Health Record Big Data Analytics.

Team-based care has been linked to key outcomes associated with the Quadruple Aim and a key driver of high-value patient-centered care. Use of the electronic health record (EHR) and machine learning have significant potential to overcome previous barriers to studying the impact of teams, including delays in accessing data to improve teamwork and optimize patient outcomes. This study utilized a large EHR dataset (n=316,542) from an urban health system to explore the relationship between team composition and patient activation, a key driver of patient engagement. Teams were operationalized using consensus definitions of teamwork from the literature. Patient activation was measured using the Patient Activation Measure (PAM). Results from multilevel regression analyses were compared to machine learning analyses using multinomial logistic regression to calculate propensity scores for the effect of team composition on PAM scores. Under the machine learning approach, a causal inference model with generalized overlap weighting was used to calculate the average treatment effect of teamwork. Seventeen different team types were observed in the data from the analyzed sample (n=12,448). Team sizes ranged from 2 to 5 members. After controlling for confounding variables in both analyses, more diverse, multidisciplinary teams (team size of 4 or more) were observed to have improved patient activation scores. This is the first study to explore the relationship between team composition and patient activation using the EHR and big data analytics. Implications for further research using EHR data and machine learning to study teams and other patient-centered care are promising and could be used to advance team science.

Prediction of cardiovascular and renal risk among patients with apparent treatment-resistant hypertension in the United States using machine learning methods.

Apparent treatment-resistant hypertension (aTRH), defined as blood pressure (BP) that remains uncontrolled despite unconfirmed concurrent treatment with threeantihypertensives, is associated with an increased risk of developing cardiovascular and renal complications compared with controlled hypertension. We aimed to identify the characteristics of aTRH patients with an elevated risk of major adverse cardiovascular events plus (MACE+; defined as stroke, myocardial infarction, or heart failure hospitalization) and end stage renal disease (ESRD). This retrospective cohort study included aTRH patients (BP ≥140/90mmHg and taking ≥3antihypertensives) from the United States-based Optum® de-identified Electronic Health Record dataset and used machine learning models to identify risk factors of MACE+ or ESRD. Patients had claims for ≥3antihypertensive classes within 30 days between January 1, 2015andJune 30, 2021, and two office BP measures recorded 1-90 days apart within 30 days to 11 months after the index regimen date. Of a total 18797 070 patients identified with any hypertension, 71100 patients had aTRH. During the study period (mean 25.5 months), 4944 (7.0%) patients had a MACE+ and 2403 (3.4%) developed ESRD. In total, 22risk factors were included in the MACE+ model and 16 in the ESRD model, and most were significantly associated with study outcomes. The risk factors with the largest impact on MACE+ risk were congestive heart failure, stages 4 and 5 chronic kidney disease (CKD), age ≥80years, and living in the Southern region of the United States. The risk factors with the largest impact on ESRD risk, other than pre-existing CKD, were anemia, congestive heart failure, and type 2 diabetes. The overall study cohort had a 5-year predicted MACE+ risk of 13.4%; this risk was increased in those in the top 50% and 25% high-risk groups (21.2% and 29.5%, respectively). The overall study cohort had a predicted 5-year risk of ESRD of 6.8%, which was increased in the top 50% and 25% high-risk groups (10.9% and 17.1%, respectively). We conclude that risk models developed in our study can reliably identify patients with aTRH at risk of MACE+ and ESRD based on information available in electronic health records; such models may be used to identify aTRH patients at high risk of adverse outcomes who may benefit from novel treatment interventions.

Analyzing the Impact of Personalization on Fairness in Federated Learning for Healthcare.

As machine learning (ML) usage becomes more popular in the healthcare sector, there are also increasing concerns about potential biases and risks such as privacy. One countermeasure is to use federated learning (FL) to support collaborative learning without the need for patient data sharing across different organizations. However, the inherent heterogeneity of data distributions among participating FL parties poses challenges for exploring group fairness in FL. While personalization within FL can handle performance degradation caused by data heterogeneity, its influence on group fairness is not fully investigated. Therefore, the primary focus of this study is to rigorously assess the impact of personalized FL on group fairness in the healthcare domain, offering a comprehensive understanding of how personalized FL affects group fairness in clinical outcomes. We conduct an empirical analysis using two prominent real-world Electronic Health Records (EHR) datasets, namely eICU and MIMIC-IV. Our methodology involves a thorough comparison between personalized FL and two baselines: standalone training, where models are developed independently without FL collaboration, and standard FL, which aims to learn a global model via the FedAvg algorithm. We adopt Ditto as our personalized FL approach, which enables each client in FL to develop its own personalized model through multi-task learning. Our assessment is achieved through a series of evaluations, comparing the predictive performance (i.e., AUROC and AUPRC) and fairness gaps (i.e., EOPP, EOD, and DP) of these methods. Personalized FL demonstrates superior predictive accuracy and fairness over standalone training across both datasets. Nevertheless, in comparison with standard FL, personalized FL shows improved predictive accuracy but does not consistently offer better fairness outcomes. For instance, in the 24-h in-hospital mortality prediction task, personalized FL achieves an average EOD of 27.4% across racial groups in the eICU dataset and 47.8% in MIMIC-IV. In comparison, standard FL records a better EOD of 26.2% for eICU and 42.0% for MIMIC-IV, while standalone training yields significantly worse EOD of 69.4% and 54.7% on these datasets, respectively. Our analysis reveals that personalized FL has the potential to enhance fairness in comparison to standalone training, yet it does not consistently ensure fairness improvements compared to standard FL. Our findings also show that while personalization can improve fairness for more biased hospitals (i.e., hospitals having larger fairness gaps in standalone training), it can exacerbate fairness issues for less biased ones. These insights suggest that the integration of personalized FL with additional strategic designs could be key to simultaneously boosting prediction accuracy and reducing fairness disparities. The findings and opportunities outlined in this paper can inform the research agenda for future studies, to overcome the limitations and further advance health equity research.

Machine Learning Methods Using Artificial Intelligence Deployed on Electronic Health Record Data for Identification and Referral of At-Risk Patients From Primary Care Physicians to Eye Care Specialists: Retrospective, Case-Controlled Study

Background Identification and referral of at-risk patients from primary care practitioners (PCPs) to eye care professionals remain a challenge. Approximately 1.9 million Americans suffer from vision loss as a result of undiagnosed or untreated ophthalmic conditions. In ophthalmology, artificial intelligence (AI) is used to predict glaucoma progression, recognize diabetic retinopathy (DR), and classify ocular tumors; however, AI has not yet been used to triage primary care patients for ophthalmology referral. Objective This study aimed to build and compare machine learning (ML) methods, applicable to electronic health records (EHRs) of PCPs, capable of triaging patients for referral to eye care specialists. Methods Accessing the Optum deidentified EHR data set, 743,039 patients with 5 leading vision conditions (age-related macular degeneration [AMD], visually significant cataract, DR, glaucoma, or ocular surface disease [OSD]) were exact-matched on age and gender to 743,039 controls without eye conditions. Between 142 and 182 non-ophthalmic parameters per patient were input into 5 ML methods: generalized linear model, L1-regularized logistic regression, random forest, Extreme Gradient Boosting (XGBoost), and J48 decision tree. Model performance was compared for each pathology to select the most predictive algorithm. The area under the curve (AUC) was assessed for all algorithms for each outcome. Results XGBoost demonstrated the best performance, showing, respectively, a prediction accuracy and an AUC of 78.6% (95% CI 78.3%-78.9%) and 0.878 for visually significant cataract, 77.4% (95% CI 76.7%-78.1%) and 0.858 for exudative AMD, 79.2% (95% CI 78.8%-79.6%) and 0.879 for nonexudative AMD, 72.2% (95% CI 69.9%-74.5%) and 0.803 for OSD requiring medication, 70.8% (95% CI 70.5%-71.1%) and 0.785 for glaucoma, 85.0% (95% CI 84.2%-85.8%) and 0.924 for type 1 nonproliferative diabetic retinopathy (NPDR), 82.2% (95% CI 80.4%-84.0%) and 0.911 for type 1 proliferative diabetic retinopathy (PDR), 81.3% (95% CI 81.0%-81.6%) and 0.891 for type 2 NPDR, and 82.1% (95% CI 81.3%-82.9%) and 0.900 for type 2 PDR. Conclusions The 5 ML methods deployed were able to successfully identify patients with elevated odds ratios (ORs), thus capable of patient triage, for ocular pathology ranging from 2.4 (95% CI 2.4-2.5) for glaucoma to 5.7 (95% CI 5.0-6.4) for type 1 NPDR, with an average OR of 3.9. The application of these models could enable PCPs to better identify and triage patients at risk for treatable ophthalmic pathology. Early identification of patients with unrecognized sight-threatening conditions may lead to earlier treatment and a reduced economic burden. More importantly, such triage may improve patients’ lives.

Factors Associated With Hospital Admission in the Last Month: A Retrospective Single Center Analysis

ContextDriven by concerns about care quality, patient experience, and national metrics, health systems are increasingly focusing on identifying risk factors for patients who are hospitalized in the last month of life. ObjectiveTo evaluate patient factors associated with hospital admission in the last month (30 days). MethodsWe analyzed a retrospective cohort of 8488 patients with a primary care visit in a tertiary health system in the last year of life using a linked electronic health record and decedent dataset. We examined healthcare utilization (primary care, emergency, hospital, intensive care unit encounters) and end-of-life related outcomes (palliative care consultation, do-not-resuscitate orders, advance care planning documentation, hospice at hospital discharge, death in health system). Multivariable logistic regressions identified patient factors associated with admission in the last month. ResultsAbout 2202 (25.9%) patients had a hospital admission in the last month. Among the 1282 (15.1%) who died in a health system facility, most (1103/1282, 86.0%) were admitted to the hospital in the last month. Among patients with a hospital admission and discharged in the last month, 60.9% (686/1126) were discharged on hospice. Compared to those without these diseases, metastatic cancer, liver disease, or heart failure had the highest odds of admission in the last month (adjusted OR 2.36 95%CI 2.05–2.72; 2.28, 95%CI 1.98–2.62; and 2.17 95%CI 1.93–2.45 respectively). ConclusionsAs patients with heart or liver disease or metastatic cancer had the highest odds of admission in the last month, collaborative interventions between primary, palliative, and specialty care may improve quality of care at the end of life.

Biomarker Assessment of a High-Risk, Data-Driven Pediatric Sepsis Phenotype Characterized by Persistent Hypoxemia, Encephalopathy, and Shock.

Identification of children with sepsis-associated multiple organ dysfunction syndrome (MODS) at risk for poor outcomes remains a challenge. We sought to the determine reproducibility of the data-driven "persistent hypoxemia, encephalopathy, and shock" (PHES) phenotype and determine its association with inflammatory and endothelial biomarkers, as well as biomarker-based pediatric risk strata. We retrained and validated a random forest classifier using organ dysfunction subscores in the 2012-2018 electronic health record (EHR) dataset used to derive the PHES phenotype. We used this classifier to assign phenotype membership in a test set consisting of prospectively (2003-2023) enrolled pediatric septic shock patients. We compared profiles of the PERSEVERE family of biomarkers among those with and without the PHES phenotype and determined the association with established biomarker-based mortality and MODS risk strata. Twenty-five PICUs across the United States. EHR data from 15,246 critically ill patients with sepsis-associated MODS split into derivation and validation sets and 1,270 pediatric septic shock patients in the test set of whom 615 had complete biomarker data. None. The area under the receiver operator characteristic curve of the modified classifier to predict PHES phenotype membership was 0.91 (95% CI, 0.90-0.92) in the EHR validation set. In the test set, PHES phenotype membership was associated with both increased adjusted odds of complicated course (adjusted odds ratio [aOR] 4.1; 95% CI, 3.2-5.4) and 28-day mortality (aOR of 4.8; 95% CI, 3.11-7.25) after controlling for age, severity of illness, and immunocompromised status. Patients belonging to the PHES phenotype were characterized by greater degree of systemic inflammation and endothelial activation, and were more likely to be stratified as high risk based on PERSEVERE biomarkers predictive of death and persistent MODS. The PHES trajectory-based phenotype is reproducible, independently associated with poor clinical outcomes, and overlapped with higher risk strata based on prospectively validated biomarker approaches.