Objectives Echocardiography and cardiac catheterization reports capture important clinical assessment information of cardiac function and disease severity. This study explores using open-source transformer-based language models (LMs) that are run locally within an institutional environment as a privacy-preserving alternative to external API-based large LM to systematically extract clinical data from unstructured echocardiography and cardiac catheterization reports, aiming to improve data accessibility for research and patient care. Materials and Methods Two transformer-based LMs, BioclinicalBERT and BART-Large-CNN, were fine-tuned in a secure local environment using a question-answering approach. The dataset included 3286 echocardiography and 1884 cardiac catheterization reports from Kaiser Permanente Southern California's electronic health records, annotated for 25 and 47 predefined categories, respectively. Three hundred reports from each type were randomly selected and used for validation, with the remainder for training. Model performance was assessed using accuracy, precision, recall, and F1-score at 2 probability thresholds. The effect of training set size on model performance was also evaluated. Results Both models achieved consistent and high accuracy, precision, and recall (all >90%) across the 5 seed runs for both report types. For echocardiography, BioclinicalBERT reached mean accuracy of 95.7%, precision of 97.6%, recall of 97.4%, and F1-score of 0.98 at the probability threshold of 0.1; BART-Large-CNN had similar results. For cardiac catheterization, BART-Large-CNN slightly outperformed BioclinicalBERT with mean accuracy 94.9% vs 94.3%; precision 96.7% vs 96.3%; recall 96.1% vs 95.7%, and F1-score 0.96 vs 0.96 at the probability threshold of 0.1. Most individual categories showed strong performance, though a few (eg, prosthetic mitral valve, right atrial pressure) had lower scores. Performance improved with more training data, but plateauing around 1000 reports. Discussion and conclusion Fine-tuned transformer-based LMs can effectively extract structured data from unstructured cardiac reports, supporting automated information extraction to enhance research and clinical applications.

BACKGROUND:Aortic stenosis (AS) is a progressive disease with substantial variability in its rate of progression. Current surveillance guidelines may not adequately identify individuals at highest risk for rapid haemodynamic deterioration. This study aims to assess AS progression rates and identify factors associated with rapid progression using real-world, longitudinal data. METHODS:Retrospective cohort study conducted within Kaiser Permanente Southern California, an integrated healthcare system. Adults aged ≥18 years with ≥2 transthoracic echocardiograms (TTEs) performed ≥180 days apart between 2011 and 2021 were included. Multivariable logistic regression was used to identify predictors of rapid progression, defined as an annualised increase in Vmax ≥0.3 m/s/year. The final cohort included 70 850 patients contributing 111 189 TTE pairs. RESULTS:Rapid progression occurred in 7.8% of patients with aortic sclerosis, 16.4% with mild AS and 29.8% with moderate AS. The mean annual increase in Vmax was 0.1 m/s/year; mean gradient increased by 1.6 mm Hg/year, and AVA declined by 0.1 cm²/year. Among patients with aortic sclerosis, rapid progression was significantly associated with older age, anaemia, liver disease, renal failure, pulmonary disease, peripheral vascular disease, cancer and obesity. In mild AS, anaemia and obesity remained significant, but echocardiographic parameters, including higher gradient and smaller AVA, were stronger predictors. In moderate AS, valve-related parameters predominated, and clinical characteristics were not significantly associated with progression. CONCLUSIONS AND RELEVANCE:AS progression varies by baseline severity and underlying risk profile. Incorporating clinical and echocardiographic risk factors into surveillance strategies may improve identification of patients most likely to benefit from closer monitoring or earlier intervention.

Background Incident atrial fibrillation (AF) is common among adults with kidney failure treated with maintenance dialysis and is associated with poor clinical outcomes. Limited data exist informing treatment of AF among patients on dialysis. We aimed to describe the use of rate‐control and antiarrhythmic medications for AF among patients on dialysis and associations of these medication use strategies with stroke and all‐cause death. Methods We evaluated patients on dialysis with incident AF between 2010 and 2017 in the Kaiser Permanente Northern and Southern California integrated health care delivery systems. We characterized time‐updated incident receipt of rate‐control (β blockers, calcium channel blockers, and digoxin) and antiarrhythmic medications from pharmacy databases. We evaluated associations of these therapies with the composite outcome of ischemic stroke and all‐cause death using Cox regression, adjusting for potential confounders. Results Of 2100 patients, 44.0% were newly prescribed rate‐control medications, 4.6% were prescribed antiarrhythmic medications, 8.9% were prescribed both, and 42.9% were prescribed neither within 12 months of newly diagnosed AF. During a median 1.66 (interquartile range, 0.45–3.39) years, we observed 1406 composite events (stroke and death). Time‐updated use of antiarrhythmics alone (adjusted hazard ratio [HR], 0.74 [95% CI, 0.57–0.96]) or with rate‐control (adjusted HR, 0.72 [95% CI, 0.58–0.90]) was associated with lower stroke or death risk versus neither medication. Use of rate‐control medications alone was not significantly associated with the composite outcome. Conclusions Among patients on dialysis with incident AF, use of antiarrhythmic medications may be associated with lower risk of stroke and death. Future randomized trials are needed to determine the efficacy and safety of antiarrhythmic medications in this high‐risk population.

STUDY OBJECTIVE:The evaluation for suspected acute coronary syndrome is a common and high-risk presentation in emergency departments (EDs). After exclusion of acute myocardial infarction (MI), patients often undergo early (within 72 hours) noninvasive cardiac testing. We evaluated the association between early noninvasive testing and death/acute MI in ED patients suspected of acute coronary syndrome. METHODS:We used a retrospective cohort study design within the adult ED patient population (from October 2015 to December 2020) in whom MI was ruled out, belonging to a large integrated health care delivery system. Using data on history (H), electrocardiogram (E), age (A), risk factors (R), and troponin (T), we computed the HEART risk score. We stratified the cohort into low (score 0 to 3), intermediate (score 4 to 6), and high (score ≥7) risk and followed them up to 1-year after ED discharge. The association between noninvasive testing within 3 days of the ED visit and composite risk of death/acute MI within 1-year of discharge was evaluated by propensity score analysis. RESULTS:The cohort included 174,917 patients (61% low risk [age 53; women 58%; noninvasive testing 5%], 36% intermediate risk [age 71; women 52%; noninvasive testing 18%], and 3% high risk [age 74, women 45%; noninvasive testing 23%]). The risk reduction in death/acute MI associated with early noninvasive testing was -1.54% (-1.95% to -1.12%) number needed to treat (NNT)=65; -4.93% (-5.66% to -4.20%) NNT=20, and -8.98% (95% confidence interval -11.32% to -6.64%) NNT=11; and, in the low, intermediate, and high-risk respectively. CONCLUSION:Early noninvasive testing was associated with reduced risk of 1-year death or acute MI across all risk groups.

BACKGROUND:Early noninvasive cardiac testing (NIT) is often performed in the initial workup of patients who present to the emergency department (ED) with suspected acute coronary syndrome (ACS). Our study objective was to calculate the cost-effectiveness of adopting early NIT for risk stratification to avoid future nonfatal acute myocardial infarction (MI) or death. METHODS:To obtain the incremental difference in cost and clinical outcomes, we first conducted a multicenter retrospective cohort study within the member population of the Kaiser Permanente Southern California integrated health care delivery system. We then adapted existing cost effectiveness models to generate long-term costs and quality-adjusted life-years (QALYs) gained by NIT. RESULTS:The cohort included 89,387 patients (mean age 57 years, 58% female) and 19% received early NIT. Total cost was higher by $2357 (95% confidence interval [CI] $77 to $4821) for early NIT compared to no early NIT and was mainly due to the increased cost of the index ED visit. Early NIT was associated with lower composite risk of death/nonfatal MI (absolute risk difference -3.7%, 95% CI -4.4% to -3.01%) during a 1-year follow-up. From a payor's perspective, early NIT was cost-effective at $5268/QALYs. CONCLUSIONS:In patients with suspected ACS evaluated in the ED, incorporation of early NIT was associated with an overall increase in cost of health care that was driven by increased cost of the initial ED visit. However, due to the significant clinical benefits, early NIT was cost-effective in the low- and intermediate-risk patients while it is a dominant strategy in high-risk patients saving cost and QALYs.

Importance:The Predicting Risk of Cardiovascular Disease Events (PREVENT) equations are an updated model developed to improve on the Pooled Cohort Equation (PCE) for estimating 10-year atherosclerotic cardiovascular disease (ASCVD) risk. These equations facilitate patient-clinician discussions on initiating statin therapy and are used to estimate risk without treatment. However, statin exposure during follow-up was not fully accounted for in the development of these equations. Objective:To assess the performance of the PCE and PREVENT equations in estimating ASCVD, accounting for statin exposure during follow-up. Design, Setting, and Participants:This retrospective cohort study included adults from an integrated health care system with 10-year follow-up data. Adults without diabetes or ASCVD were identified in 2013 and followed-up through December 31, 2023, with analyses performed in January 2025. Main Outcomes and Measures:The primary outcome was incident ASCVD. Estimated risks from PCE and PREVENT equations were compared with observed risks, with discrimination assessed via C statistics. The performance of these equations was evaluated in patient populations stratified by statin exposure during follow-up. Results:Among 193 885 adults (median [IQR] age, 55 [48-63] years; 113 400 [58.5%] women), 6528 experienced an ASCVD event. The C statistic was 0.725 (95% CI, 0.719-0.731) for PCE and 0.723 (95% CI, 0.716-0.729) for PREVENT. In the overall population, regardless of statin exposure, the observed 10-year ASCVD risk was lower than estimated by PCE: 3.6% for individuals with estimated risk of 5% to less than 7.5%, 4.5% for those with estimated risk of 7.5% to less than 10%, and 8.0% for those with estimated risk of 10% or greater. The observed risk more closely aligned with the estimated risk from PREVENT: 5.2% for individuals with estimated risk of 5% to 7.5%, 8.1% for those with estimated risk 7.5% to less than 10%, and 11.6% for those with estimated risk of 10% or greater. In contrast, among patients not exposed to statin therapy during follow-up, PREVENT underestimated risk: observed risk was 8.2% for individuals with estimated risk of 5% to less than 7.5%, and 13.5% for those with estimated risk of 7.5% to less than 10%, while PCE-estimated risk more closely approximated the observed risk. Conclusions and Relevance:In this retrospective cohort study, the PREVENT model underestimated risk in patients not treated with statins, whereas the PCE estimates more closely reflected what a patient's risk would be without statin therapy.

Background and Aims:Accurate assessment of aortic stenosis (AS) requires integration of both structural and functional information characterized by visual traits as well as quantitation of gradients. Existing artificial intelligence (AI) models utilize solely either structural or functional information. Methods:We developed EchoNet-AS, an open-source end-to-end integrated approach combining video based convolutional neural networks to assess valve motion as well as segmentation models to automate the measurement of aortic valve peak velocity to classify AS severity. Results:EchoNet-AS was trained on 210,193 images from 16,076 studies from Kaiser Permanente Northern California (KPNC) and validated on 1,589 held-out test studies and a temporally distinct cohort of 19,206 studies. The final model was also externally validated on 2,415 studies from Stanford Healthcare (SHC) and 9,038 studies from Cedars-Sinai Medical Center (CSMC). Combining assessments from multiple echocardiographic videos and Doppler measurements, EchoNet-AS achieved excellent discrimination of severe AS with AUC 0.964 [95% CI: 0.952 - 0.973] in the KPNC held-out cohort and 0.985 [0.981 - 0.988] in the temporally distinct cohort, which was superior to models using single views or only Doppler measurements. The performance was consistently robust in distinct external cohorts with an AUC 0.985 [0.975 - 0.992] at SHC and 0.989 [0.986 - 0.992] at CSMC. Conclusions:EchoNet-AS synthesizes information from both B-mode videos and Doppler images to accurately assess AS severity. Its strong performance generalizes robustly to external validation cohorts and shows potential as an automated clinical decision support tool.

BACKGROUND:Rosuvastatin and atorvastatin are the two primary high-intensity statins used for cardiovascular risk reduction. However, concerns have been raised regarding the renal safety profile of rosuvastatin. This study compared risks of hematuria, proteinuria, and cardiovascular events between rosuvastatin- and atorvastatin-treated patients. METHODS:This is a retrospective cohort study of atorvastatin and rosuvastatin users from 1 January 2018 to 31 December 2021. Inverse probability of treatment weighting was used to mitigate confounding. Poisson regression models were used to estimate adjusted relative risks. RESULTS:Of 136,680 patients (median age 61 years, 42.6% female), 46,292 were treated with atorvastatin and 90,388 rosuvastatin. During follow-up, there were 619 cases of proteinuria, 1843 cases of hematuria, and 19 cases of end-stage renal disease. Compared with atorvastatin, rosuvastatin was associated with a nonsignificant increase in the risk of hematuria (adjusted risk ratio [aRR] 1.1, 95% confidence interval [CI] 0.98-1.24) and proteinuria (aRR 1.02, 95% CI 0.83-1.25). Risk of cardiovascular events was significantly lower with rosuvastatin (aRR 0.77, 95% CI 0.70-0.83). Rosuvastatin was associated with a lower risk of myocardial infarction (aRR 0.75, 95% CI 0.63-0.91) and stroke (aRR 0.76, 95% CI 0.62-0.94) but not heart failure (aRR 0.82, 95% CI 0.63-1.06). Significant cardiovascular risk reduction was observed with rosuvastatin in the first 6 months. CONCLUSIONS AND RELEVANCE:Compared with atorvastatin, rosuvastatin was associated with a significantly lower risk of cardiovascular events and a nonsignificant increase in hematuria and proteinuria. Overall, the cardiovascular benefits of rosuvastatin appear to outweigh the minor renal risks.

PURPOSE:Contemporary epidemiologic research on acute myocardial infarction (AMI) using electronic health records (EHR) relies on International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes, but limited studies have been conducted to validate these codes in the United States. Therefore, this study aimed to validate AMI events identified by ICD-10-CM diagnosis codes. METHODS:The study was conducted as part of a hepatitis B vaccine safety study. Suspected cases of AMI were identified using ICD-10-CM codes (I21.* or I22.*) in any diagnosis position from August 7, 2018 to November 30, 2020. Cases were adjudicated independently by two cardiologists, with a third resolving disagreements. Positive predictive value (PPV) was calculated as the percentage of suspected cases that were confirmed as definite or probable AMI on review, and exact binomial 95% confidence intervals (CI) were estimated. RESULTS:Of 202 potential AMI events identified among 69 625 individuals, 162 (80.2% [95% CI: 74.0%-85.5%]) were confirmed. Encounters with AMI coded as the principal discharge diagnosis code were more likely to be confirmed (86.8% [80.5%-91.6%]) than those with AMI in another diagnosis position (55.8% [39.9%-70.9%]), while patients with a history of congestive heart failure and peripheral vascular disease had lower PPV compared to those without (83.2% [76.7%-88.6%] and 82.9% [76.4%-88.3%], respectively). CONCLUSION:We found that over 80% of AMI cases identified with ICD-10-CM codes were confirmed upon cardiologist adjudication. Cases not coded in the principal diagnosis position were much less likely to be confirmed, and care should be taken when using them in EHR-based research.

Background An increasing proportion of visits are now delivered via a virtual platform. Virtual visits are limited by the lack of important components of cardiovascular assessment such as physician examination and electrocardiogram. Objectives The purpose of this study was to evaluate the quality of care delivered by virtual visits compared to office-based visits among adults who sought care for three common cardiac-related symptoms: dyspnea, dizziness, or palpitations. Methods Retrospective cohort study of 992,526 outpatient visits between January 1, 2017, and December 31, 2021, within an integrated health system, including 356,159 visits for dyspnea, 412,913 for dizziness, and 223,454 for palpitations. We compared the differences in patient characteristics associated with telemedicine visits versus in-office visits, evaluated the referral rates for noninvasive cardiac testing, and examined the association between virtual visits and 30-day clinical outcomes. Results Among 992,526 visits, 71.5% were office visits, 25.8% telephone visits, and 2.7% video visits. Median age was 59 (IQR: 43-72) years, and 63.1% were women. Patient characteristics associated with increased likelihood of virtual visits included younger age, female sex, being non-Hispanic Black, and being from lower-income households. No association was observed between visit types and 30-day cardiovascular hospitalization for patients with dizziness or palpitations. However, for patients with dyspnea, evaluation via virtual visits was associated with a higher risk of 30-day hospitalization for heart failure (aOR: 1.25; 95% CI: 1.16-1.36 for telephone visits; aOR: 1.45; 95% CI: 1.17-1.80 for video visits). Compared to office-based visits, patients with dyspnea were less likely to be referred for echocardiogram with telephone (aOR: 0.73; 95% CI: 0.72-0.75) or video visits (aOR: 0.92; 95% CI: 0.87-0.98). Conclusions Virtual visits may be appropriate for some clinical concerns but not all. Optimal alignment of clinical conditions with appropriate care modalities is an important component of a successful telemedicine strategy.

Background: Palpitations represent a common complaint in primary care clinics. Although usually benign, palpitations are occasionally a manifestation of cardiac arrhythmias. Aims: This study aimed to investigate whether there are gender differences in the cardiac testing pattern and clinical outcomes of patients evaluated in outpatient clinics for palpitations. Methods: This is a retrospective observational study that included adult men and women who presented to an outpatient primary care or cardiology office in an integrated health system in California with a chief complaint of palpitations. Cardiac testing pattern was captured using electronic health records. The primary endpoint was hospitalization for arrhythmia at one year. The secondary endpoint was all-cause mortality at one year. Logistic regression models were constructed to evaluate the association between female gender and the outcomes. Results: Between 2017 and 2021, 89,680 patients were evaluated for palpitations, among whom 61,064 (68.1%) were women. A high proportion of women were Hispanic. Women were more likely to be obese and less likely to have hypertension, diabetes, atrial fibrillation, heart failure, or a history of myocardial infarction. A slightly higher proportion of women were started on beta-blockers (12.8% women vs. 12.2% men, p=0.004). Women were more likely to be referred for cardiac rhythm monitoring (19.8% women vs. 18.8% men, p <0.001). At one year, women had a lower rate of hospitalization for arrhythmias (0.5% in men versus 0.3% in women, adjusted OR 0.73, 95% CI 0.58-0.91). All-cause mortality was also lower for women at one year (adjusted OR 0.55, 95% CI 0.48-0.62). Conclusion: Among patients with palpitations, women were more likely than men to be treated with beta-blockers and referred for cardiac rhythm monitoring. Women had a better clinical prognosis, with a lower risk of hospitalization for arrhythmias and death at one year.

Background Valvular heart disease (VHD) is a leading cause of cardiovascular morbidity and mortality that poses a substantial health care and economic burden on health care systems. Administrative diagnostic codes for ascertaining VHD diagnosis are incomplete. Objective This study aimed to develop a natural language processing (NLP) algorithm to identify patients with aortic, mitral, tricuspid, and pulmonic valve stenosis and regurgitation from transthoracic echocardiography (TTE) reports within a large integrated health care system. Methods We used reports from echocardiograms performed in the Kaiser Permanente Southern California (KPSC) health care system between January 1, 2011, and December 31, 2022. Related terms/phrases of aortic, mitral, tricuspid, and pulmonic stenosis and regurgitation and their severities were compiled from the literature and enriched with input from clinicians. An NLP algorithm was iteratively developed and fine-trained via multiple rounds of chart review, followed by adjudication. The developed algorithm was applied to 200 annotated echocardiography reports to assess its performance and then the study echocardiography reports. Results A total of 1,225,270 TTE reports were extracted from KPSC electronic health records during the study period. In these reports, valve lesions identified included 111,300 (9.08%) aortic stenosis, 20,246 (1.65%) mitral stenosis, 397 (0.03%) tricuspid stenosis, 2585 (0.21%) pulmonic stenosis, 345,115 (28.17%) aortic regurgitation, 802,103 (65.46%) mitral regurgitation, 903,965 (73.78%) tricuspid regurgitation, and 286,903 (23.42%) pulmonic regurgitation. Among the valves, 50,507 (4.12%), 22,656 (1.85%), 1685 (0.14%), and 1767 (0.14%) were identified as prosthetic aortic valves, mitral valves, tricuspid valves, and pulmonic valves, respectively. Mild and moderate were the most common severity levels of heart valve stenosis, while trace and mild were the most common severity levels of regurgitation. Males had a higher frequency of aortic stenosis and all 4 valvular regurgitations, while females had more mitral, tricuspid, and pulmonic stenosis. Non-Hispanic Whites had the highest frequency of all 4 valvular stenosis and regurgitations. The distribution of valvular stenosis and regurgitation severity was similar across race/ethnicity groups. Frequencies of aortic stenosis, mitral stenosis, and regurgitation of all 4 heart valves increased with age. In TTE reports with stenosis detected, younger patients were more likely to have mild aortic stenosis, while older patients were more likely to have severe aortic stenosis. However, mitral stenosis was opposite (milder in older patients and more severe in younger patients). In TTE reports with regurgitation detected, younger patients had a higher frequency of severe/very severe aortic regurgitation. In comparison, older patients had higher frequencies of mild aortic regurgitation and severe mitral/tricuspid regurgitation. Validation of the NLP algorithm against the 200 annotated TTE reports showed excellent precision, recall, and F1-scores. Conclusions The proposed computerized algorithm could effectively identify heart valve stenosis and regurgitation, as well as the severity of valvular involvement, with significant implications for pharmacoepidemiological studies and outcomes research.