Blinded, randomized trial of sonographer versus AI cardiac function assessment.

Artificial intelligence (AI) has been developed for echocardiography1-3, although it has not yet been tested with blinding and randomization. Here we designed a blinded, randomized non-inferiority clinical trial (ClinicalTrials.gov ID: NCT05140642; no outside funding) of AI versus sonographer initial assessment of left ventricular ejection fraction (LVEF) to evaluate the impact of AI in the interpretation workflow. The primary end point was the change in the LVEF between initial AI or sonographer assessment and final cardiologist assessment, evaluated by the proportion of studies with substantial change (more than 5% change). From 3,769 echocardiographic studies screened, 274 studies were excluded owing to poor image quality. The proportion of studies substantially changed was 16.8% in the AI group and 27.2% in the sonographer group (difference of -10.4%, 95% confidence interval: -13.2% to -7.7%, P < 0.001 for non-inferiority, P < 0.001 for superiority). The mean absolute difference between final cardiologist assessment and independent previous cardiologist assessment was 6.29% in the AI group and 7.23% in the sonographer group (difference of -0.96%, 95% confidence interval: -1.34% to -0.54%, P < 0.001 for superiority). The AI-guided workflow saved time for both sonographers and cardiologists, and cardiologists were not able to distinguish between the initial assessments by AI versus the sonographer (blinding index of 0.088). For patients undergoing echocardiographic quantification of cardiac function, initial assessment of LVEF by AI was non-inferior to assessment by sonographers.

Video-based AI for beat-to-beat assessment of cardiac function.

Accurate assessment of cardiac function is crucial for the diagnosis of cardiovascular disease1, screening for cardiotoxicity2 and decisions regarding the clinical management of patients with a critical illness3. However, human assessment of cardiac function focuses on a limited sampling of cardiac cycles and has considerable inter-observer variability despite years of training4,5. Here, to overcome this challenge, we present a video-based deep learning algorithm-EchoNet-Dynamic-that surpasses the performance of human experts in the critical tasks of segmenting the left ventricle, estimating ejection fraction and assessing cardiomyopathy. Trained on echocardiogram videos, our model accurately segments the left ventricle with a Dice similarity coefficient of 0.92, predicts ejection fraction with a mean absolute error of 4.1% and reliably classifies heart failure with reduced ejection fraction (area under the curve of 0.97). In an external dataset from another healthcare system, EchoNet-Dynamic predicts the ejection fraction with a mean absolute error of 6.0% and classifies heart failure with reduced ejection fraction with an area under the curve of 0.96. Prospective evaluation with repeated human measurements confirms that the model has variance that is comparable to or less than that of human experts. By leveraging information across multiple cardiac cycles, our model can rapidly identify subtle changes in ejection fraction, is more reproducible than human evaluation and lays the foundation for precise diagnosis of cardiovascular disease in real time. As a resource to promote further innovation, we also make publicly available a large dataset of 10,030 annotated echocardiogram videos.

Detection of Left Ventricular Systolic Dysfunction From Electrocardiographic Images.

BACKGROUND: Left ventricular (LV) systolic dysfunction is associated with a >8-fold increased risk of heart failure and a 2-fold risk of premature death. The use of ECG signals in screening for LV systolic dysfunction is limited by their availability to clinicians. We developed a novel deep learning-based approach that can use ECG images for the screening of LV systolic dysfunction. METHODS: Using 12-lead ECGs plotted in multiple different formats, and corresponding echocardiographic data recorded within 15 days from the Yale New Haven Hospital between 2015 and 2021, we developed a convolutional neural network algorithm to detect an LV ejection fraction <40%. The model was validated within clinical settings at Yale New Haven Hospital and externally on ECG images from Cedars Sinai Medical Center in Los Angeles, CA; Lake Regional Hospital in Osage Beach, MO; Memorial Hermann Southeast Hospital in Houston, TX; and Methodist Cardiology Clinic of San Antonio, TX. In addition, it was validated in the prospective Brazilian Longitudinal Study of Adult Health. Gradient-weighted class activation mapping was used to localize class-discriminating signals on ECG images. RESULTS: Overall, 385 601 ECGs with paired echocardiograms were used for model development. The model demonstrated high discrimination across various ECG image formats and calibrations in internal validation (area under receiving operation characteristics [AUROCs], 0.91; area under precision-recall curve [AUPRC], 0.55); and external sets of ECG images from Cedars Sinai (AUROC, 0.90 and AUPRC, 0.53), outpatient Yale New Haven Hospital clinics (AUROC, 0.94 and AUPRC, 0.77), Lake Regional Hospital (AUROC, 0.90 and AUPRC, 0.88), Memorial Hermann Southeast Hospital (AUROC, 0.91 and AUPRC 0.88), Methodist Cardiology Clinic (AUROC, 0.90 and AUPRC, 0.74), and Brazilian Longitudinal Study of Adult Health cohort (AUROC, 0.95 and AUPRC, 0.45). An ECG suggestive of LV systolic dysfunction portended >27-fold higher odds of LV systolic dysfunction on transthoracic echocardiogram (odds ratio, 27.5 [95% CI, 22.3-33.9] in the held-out set). Class-discriminative patterns localized to the anterior and anteroseptal leads (V2 and V3), corresponding to the left ventricle regardless of the ECG layout. A positive ECG screen in individuals with an LV ejection fraction ≥40% at the time of initial assessment was associated with a 3.9-fold increased risk of developing incident LV systolic dysfunction in the future (hazard ratio, 3.9 [95% CI, 3.3-4.7]; median follow-up, 3.2 years). CONCLUSIONS: We developed and externally validated a deep learning model that identifies LV systolic dysfunction from ECG images. This approach represents an automated and accessible screening strategy for LV systolic dysfunction, particularly in low-resource settings.

Severe aortic stenosis detection by deep learning applied to echocardiography.

BACKGROUND AND AIMS: Early diagnosis of aortic stenosis (AS) is critical to prevent morbidity and mortality but requires skilled examination with Doppler imaging. This study reports the development and validation of a novel deep learning model that relies on two-dimensional (2D) parasternal long axis videos from transthoracic echocardiography without Doppler imaging to identify severe AS, suitable for point-of-care ultrasonography. METHODS AND RESULTS: In a training set of 5257 studies (17 570 videos) from 2016 to 2020 [Yale-New Haven Hospital (YNHH), Connecticut], an ensemble of three-dimensional convolutional neural networks was developed to detect severe AS, leveraging self-supervised contrastive pretraining for label-efficient model development. This deep learning model was validated in a temporally distinct set of 2040 consecutive studies from 2021 from YNHH as well as two geographically distinct cohorts of 4226 and 3072 studies, from California and other hospitals in New England, respectively. The deep learning model achieved an area under the receiver operating characteristic curve (AUROC) of 0.978 (95% CI: 0.966, 0.988) for detecting severe AS in the temporally distinct test set, maintaining its diagnostic performance in geographically distinct cohorts [0.952 AUROC (95% CI: 0.941, 0.963) in California and 0.942 AUROC (95% CI: 0.909, 0.966) in New England]. The model was interpretable with saliency maps identifying the aortic valve, mitral annulus, and left atrium as the predictive regions. Among non-severe AS cases, predicted probabilities were associated with worse quantitative metrics of AS suggesting an association with various stages of AS severity. CONCLUSION: This study developed and externally validated an automated approach for severe AS detection using single-view 2D echocardiography, with potential utility for point-of-care screening.

Pseudo-safety in a cohort of patients with COVID-19 discharged home from the emergency department.

INTRODUCTION: EDs are often the first line of contact with individuals infected with COVID-19 and play a key role in triage. However, there is currently little specific guidance for deciding when patients with COVID-19 require hospitalisation and when they may be safely observed as an outpatient. METHODS: In this retrospective study, we characterised all patients with COVID-19 discharged home from EDs in our US multisite healthcare system from March 2020 to August 2020, focusing on individuals who returned within 2 weeks and required hospital admission. We restricted analyses to first-encounter data that do not depend on laboratory or imaging diagnostics in order to inform point-of-care assessments in resource-limited environments. Vitals and comorbidities were extracted from the electronic health record. We performed ordinal logistic regression analyses to identify predictors of inpatient admission, intensive care and intubation. RESULTS: Of n=923 patients who were COVID-19 positive discharged from the ED, n=107 (11.6%) returned within 2 weeks and were admitted. In a multivariable-adjusted model including n=788 patients with complete risk factor information, history of hypertension increased odds of hospitalisation and severe illness by 1.92-fold (95% CI 1.07 to 3.41), diabetes by 2.20-fold (1.18 to 4.02), chronic lung disease by 2.21-fold (1.22 to 3.92) and fever by 2.89-fold (1.71 to 4.82). Having at least two of these risk factors increased the odds of future hospitalisation by 6.68-fold (3.54 to 12.70). Patients with hypertension, diabetes, chronic lung disease or fever had significantly longer hospital stays (median 5.92 days, 3.08-10.95 vs 3.21, 1.10-5.75, p<0.01) with numerically higher but not significantly different rates of intensive care unit admission (27.02% vs 14.30%, p=0.27) and intubation (12.16% vs 7.14%, p=0.71). DISCUSSION: Patients infected with COVID-19 may appear clinically safe for home convalescence. However, those with hypertension, diabetes, chronic lung disease and fever may in fact be only 'pseudo-safe' and are most at risk for subsequent hospitalisation with more severe illness and longer hospital stays.

Vision-language foundation model for echocardiogram interpretation.

The development of robust artificial intelligence models for echocardiography has been limited by the availability of annotated clinical data. Here, to address this challenge and improve the performance of cardiac imaging models, we developed EchoCLIP, a vision-language foundation model for echocardiography, that learns the relationship between cardiac ultrasound images and the interpretations of expert cardiologists across a wide range of patients and indications for imaging. After training on 1,032,975 cardiac ultrasound videos and corresponding expert text, EchoCLIP performs well on a diverse range of benchmarks for cardiac image interpretation, despite not having been explicitly trained for individual interpretation tasks. EchoCLIP can assess cardiac function (mean absolute error of 7.1% when predicting left ventricular ejection fraction in an external validation dataset) and identify implanted intracardiac devices (area under the curve (AUC) of 0.84, 0.92 and 0.97 for pacemakers, percutaneous mitral valve repair and artificial aortic valves, respectively). We also developed a long-context variant (EchoCLIP-R) using a custom tokenizer based on common echocardiography concepts. EchoCLIP-R accurately identified unique patients across multiple videos (AUC of 0.86), identified clinical transitions such as heart transplants (AUC of 0.79) and cardiac surgery (AUC 0.77) and enabled robust image-to-text search (mean cross-modal retrieval rank in the top 1% of candidate text reports). These capabilities represent a substantial step toward understanding and applying foundation models in cardiovascular imaging for preliminary interpretation of echocardiographic findings.

Deep learning evaluation of echocardiograms to identify occult atrial fibrillation.

Atrial fibrillation (AF) often escapes detection, given its frequent paroxysmal and asymptomatic presentation. Deep learning of transthoracic echocardiograms (TTEs), which have structural information, could help identify occult AF. We created a two-stage deep learning algorithm using a video-based convolutional neural network model that (1) distinguished whether TTEs were in sinus rhythm or AF and then (2) predicted which of the TTEs in sinus rhythm were in patients who had experienced AF within 90 days. Our model, trained on 111,319 TTE videos, distinguished TTEs in AF from those in sinus rhythm with high accuracy in a held-out test cohort (AUC 0.96 (0.95-0.96), AUPRC 0.91 (0.90-0.92)). Among TTEs in sinus rhythm, the model predicted the presence of concurrent paroxysmal AF (AUC 0.74 (0.71-0.77), AUPRC 0.19 (0.16-0.23)). Model discrimination remained similar in an external cohort of 10,203 TTEs (AUC of 0.69 (0.67-0.70), AUPRC 0.34 (0.31-0.36)). Performance held across patients who were women (AUC 0.76 (0.72-0.81)), older than 65 years (0.73 (0.69-0.76)), or had a CHA2DS2VASc ≥2 (0.73 (0.79-0.77)). The model performed better than using clinical risk factors (AUC 0.64 (0.62-0.67)), TTE measurements (0.64 (0.62-0.67)), left atrial size (0.63 (0.62-0.64)), or CHA2DS2VASc (0.61 (0.60-0.62)). An ensemble model in a cohort subset combining the TTE model with an electrocardiogram (ECGs) deep learning model performed better than using the ECG model alone (AUC 0.81 vs. 0.79, p = 0.01). Deep learning using TTEs can predict patients with active or occult AF and could be used for opportunistic AF screening that could lead to earlier treatment.

Deep Learning-Derived Myocardial Strain.

BACKGROUND: Echocardiographic strain measurements require extensive operator experience and have significant intervendor variability. Creating an automated, open-source, vendor-agnostic method to retrospectively measure global longitudinal strain (GLS) from standard echocardiography B-mode images would greatly improve post hoc research applications and may streamline patient analyses. OBJECTIVES: This study was seeking to develop an automated deep learning strain (DLS) analysis pipeline and validate its performance across multiple applications and populations. METHODS: Interobserver/-vendor variation of traditional GLS, and simulated effects of variation in contour on speckle-tracking measurements were assessed. The DLS pipeline was designed to take semantic segmentation results from EchoNet-Dynamic and derive longitudinal strain by calculating change in the length of the left ventricular endocardial contour. DLS was evaluated for agreement with GLS on a large external dataset and applied across a range of conditions that result in cardiac hypertrophy. RESULTS: In patients scanned by 2 sonographers using 2 vendors, GLS had an intraclass correlation of 0.29 (95% CI: -0.01 to 0.53, P = 0.03) between vendor measurements and 0.63 (95% CI: 0.48-0.74, P < 0.001) between sonographers. With minor changes in initial input contour, step-wise pixel shifts resulted in a mean absolute error of 3.48% and proportional strain difference of 13.52% by a 6-pixel shift. In external validation, DLS maintained moderate agreement with 2-dimensional GLS (intraclass correlation coefficient [ICC]: 0.56, P = 0.002) with a bias of -3.31% (limits of agreement: -11.65% to 5.02%). The DLS method showed differences (P < 0.0001) between populations with cardiac hypertrophy and had moderate agreement in a patient population of advanced cardiac amyloidosis: ICC was 0.64 (95% CI: 0.53-0.72), P < 0.001, with a bias of 0.57%, limits of agreement of -4.87% to 6.01% vs 2-dimensional GLS. CONCLUSIONS: The open-source DLS provides lower variation than human measurements and similar quantitative results. The method is rapid, consistent, vendor-agnostic, publicly released, and applicable across a wide range of imaging qualities.

A Multimodality Video-Based AI Biomarker For Aortic Stenosis Development And Progression.

Importance: Aortic stenosis (AS) is a major public health challenge with a growing therapeutic landscape, but current biomarkers do not inform personalized screening and follow-up. Objective: A video-based artificial intelligence (AI) biomarker (Digital AS Severity index [DASSi]) can detect severe AS using single-view long-axis echocardiography without Doppler. Here, we deploy DASSi to patients with no or mild/moderate AS at baseline to identify AS development and progression. Design Setting and Participants: We defined two cohorts of patients without severe AS undergoing echocardiography in the Yale-New Haven Health System (YNHHS) (2015-2021, 4.1[IQR:2.4-5.4] follow-up years) and Cedars-Sinai Medical Center (CSMC) (2018-2019, 3.4[IQR:2.8-3.9] follow-up years). We further developed a novel computational pipeline for the cross-modality translation of DASSi into cardiac magnetic resonance (CMR) imaging in the UK Biobank (2.5[IQR:1.6-3.9] follow-up years). Analyses were performed between August 2023-February 2024. Exposure: DASSi (range: 0-1) derived from AI applied to echocardiography and CMR videos. Main Outcomes and Measures: Annualized change in peak aortic valve velocity (AV-Vmax) and late (>6 months) aortic valve replacement (AVR). Results: A total of 12,599 participants were included in the echocardiographic study (YNHHS: n=8,798, median age of 71 [IQR (interquartile range):60-80] years, 4250 [48.3%] women, and CSMC: n=3,801, 67 [IQR:54-78] years, 1685 [44.3%] women). Higher baseline DASSi was associated with faster progression in AV-Vmax (per 0.1 DASSi increments: YNHHS: +0.033 m/s/year [95%CI:0.028-0.038], n=5,483, and CSMC: +0.082 m/s/year [0.053-0.111], n=1,292), with levels ≥ vs <0.2 linked to a 4-to-5-fold higher AVR risk (715 events in YNHHS; adj.HR 4.97 [95%CI: 2.71-5.82], 56 events in CSMC: 4.04 [0.92-17.7]), independent of age, sex, ethnicity/race, ejection fraction and AV-Vmax. This was reproduced across 45,474 participants (median age 65 [IQR:59-71] years, 23,559 [51.8%] women) undergoing CMR in the UK Biobank (adj.HR 11.4 [95%CI:2.56-50.60] for DASSi ≥vs<0.2). Saliency maps and phenome-wide association studies supported links with traditional cardiovascular risk factors and diastolic dysfunction. Conclusions and Relevance: In this cohort study of patients without severe AS undergoing echocardiography or CMR imaging, a new AI-based video biomarker is independently associated with AS development and progression, enabling opportunistic risk stratification across cardiovascular imaging modalities as well as potential application on handheld devices.

A Multimodal Video-Based AI Biomarker for Aortic Stenosis Development and Progression.

Importance: Aortic stenosis (AS) is a major public health challenge with a growing therapeutic landscape, but current biomarkers do not inform personalized screening and follow-up. A video-based artificial intelligence (AI) biomarker (Digital AS Severity index [DASSi]) can detect severe AS using single-view long-axis echocardiography without Doppler characterization. Objective: To deploy DASSi to patients with no AS or with mild or moderate AS at baseline to identify AS development and progression. Design, Setting, and Participants: This is a cohort study that examined 2 cohorts of patients without severe AS undergoing echocardiography in the Yale New Haven Health System (YNHHS; 2015-2021) and Cedars-Sinai Medical Center (CSMC; 2018-2019). A novel computational pipeline for the cross-modal translation of DASSi into cardiac magnetic resonance (CMR) imaging was further developed in the UK Biobank. Analyses were performed between August 2023 and February 2024. Exposure: DASSi (range, 0-1) derived from AI applied to echocardiography and CMR videos. Main Outcomes and Measures: Annualized change in peak aortic valve velocity (AV-Vmax) and late (>6 months) aortic valve replacement (AVR). Results: A total of 12 599 participants were included in the echocardiographic study (YNHHS: n = 8798; median [IQR] age, 71 [60-80] years; 4250 [48.3%] women; median [IQR] follow-up, 4.1 [2.4-5.4] years; and CSMC: n = 3801; median [IQR] age, 67 [54-78] years; 1685 [44.3%] women; median [IQR] follow-up, 3.4 [2.8-3.9] years). Higher baseline DASSi was associated with faster progression in AV-Vmax (per 0.1 DASSi increment: YNHHS, 0.033 m/s per year [95% CI, 0.028-0.038] among 5483 participants; CSMC, 0.082 m/s per year [95% CI, 0.053-0.111] among 1292 participants), with values of 0.2 or greater associated with a 4- to 5-fold higher AVR risk than values less than 0.2 (YNHHS: 715 events; adjusted hazard ratio [HR], 4.97 [95% CI, 2.71-5.82]; CSMC: 56 events; adjusted HR, 4.04 [95% CI, 0.92-17.70]), independent of age, sex, race, ethnicity, ejection fraction, and AV-Vmax. This was reproduced across 45 474 participants (median [IQR] age, 65 [59-71] years; 23 559 [51.8%] women; median [IQR] follow-up, 2.5 [1.6-3.9] years) undergoing CMR imaging in the UK Biobank (for participants with DASSi ≥0.2 vs those with DASSi <.02, adjusted HR, 11.38 [95% CI, 2.56-50.57]). Saliency maps and phenome-wide association studies supported associations with cardiac structure and function and traditional cardiovascular risk factors. Conclusions and Relevance: In this cohort study of patients without severe AS undergoing echocardiography or CMR imaging, a new AI-based video biomarker was independently associated with AS development and progression, enabling opportunistic risk stratification across cardiovascular imaging modalities as well as potential application on handheld devices.

Electrocardiographic deep learning for predicting post-procedural mortality: a model development and validation study.

BACKGROUND: Preoperative risk assessments used in clinical practice are insufficient in their ability to identify risk for postoperative mortality. Deep-learning analysis of electrocardiography can identify hidden risk markers that can help to prognosticate postoperative mortality. We aimed to develop a prognostic model that accurately predicts postoperative mortality in patients undergoing medical procedures and who had received preoperative electrocardiographic diagnostic testing. METHODS: In a derivation cohort of preoperative patients with available electrocardiograms (ECGs) from Cedars-Sinai Medical Center (Los Angeles, CA, USA) between Jan 1, 2015 and Dec 31, 2019, a deep-learning algorithm was developed to leverage waveform signals to discriminate postoperative mortality. We randomly split patients (8:1:1) into subsets for training, internal validation, and final algorithm test analyses. Model performance was assessed using area under the receiver operating characteristic curve (AUC) values in the hold-out test dataset and in two external hospital cohorts and compared with the established Revised Cardiac Risk Index (RCRI) score. The primary outcome was post-procedural mortality across three health-care systems. FINDINGS: 45 969 patients had a complete ECG waveform image available for at least one 12-lead ECG performed within the 30 days before the procedure date (59 975 inpatient procedures and 112 794 ECGs): 36 839 patients in the training dataset, 4549 in the internal validation dataset, and 4581 in the internal test dataset. In the held-out internal test cohort, the algorithm discriminates mortality with an AUC value of 0·83 (95% CI 0·79-0·87), surpassing the discrimination of the RCRI score with an AUC of 0·67 (0·61-0·72). The algorithm similarly discriminated risk for mortality in two independent US health-care systems, with AUCs of 0·79 (0·75-0·83) and 0·75 (0·74-0·76), respectively. Patients determined to be high risk by the deep-learning model had an unadjusted odds ratio (OR) of 8·83 (5·57-13·20) for postoperative mortality compared with an unadjusted OR of 2·08 (0·77-3·50) for postoperative mortality for RCRI scores of more than 2. The deep-learning algorithm performed similarly for patients undergoing cardiac surgery (AUC 0·85 [0·77-0·92]), non-cardiac surgery (AUC 0·83 [0·79-0·88]), and catheterisation or endoscopy suite procedures (AUC 0·76 [0·72-0·81]). INTERPRETATION: A deep-learning algorithm interpreting preoperative ECGs can improve discrimination of postoperative mortality. The deep-learning algorithm worked equally well for risk stratification of cardiac surgeries, non-cardiac surgeries, and catheterisation laboratory procedures, and was validated in three independent health-care systems. This algorithm can provide additional information to clinicians making the decision to perform medical procedures and stratify the risk of future complications. FUNDING: National Heart, Lung, and Blood Institute.

Community health education at student-run clinics leads to sustained improvement in patients' hepatitis B knowledge.

While student-run clinics are often important healthcare safety nets for underserved populations, their efficacy for improving patient health knowledge has not been thoroughly explored. From September 2011 to April 2012, we assessed patients' retention of hepatitis B virus (HBV) knowledge after receiving student-led education at two student-run HBV screening and vaccination clinics. Patient education was provided by trained first and second-year medical, nursing, and pharmacy students, aided by a script and interpreters. Patient knowledge of HBV was evaluated at three points: before education, after the initial visit, and at one-month follow-up. Student-led education produced improved knowledge of HBV transmission, prevention, and management, which was retained 1 month after education for 52 patients tracked through time. Mean scores on an HBV knowledge survey improved from 56.4 % (SD = 15.2 %) at baseline to 66.6 % (SD = 15.1 %) after education, and 68.3 % (SD = 15.2 %) after one month. There was a statistically significant difference between the first and second (paired T test, p < 0.001) and the first and third tests (paired T test, p < 0.001), but no difference between the second and third tests (paired T test, p = 0.45). Multivariate analysis demonstrated that retention was correlated with patient educational background but independent of patient age, gender, income, primary language and number of years lived in the United States. Our study suggests that trained health professional students can effectively impart health knowledge that is retained by patients for at least 1 month. These results warrant consideration of student-led educational sessions at SRCs as a promising community health education model.

High-Throughput Assessment of Real-World Medication Effects on QT Interval Prolongation: Observational Study.

BACKGROUND: Drug-induced prolongation of the corrected QT interval (QTc) increases the risk for Torsades de Pointes (TdP) and sudden cardiac death. Medication effects on the QTc have been studied in controlled settings but may not be well evaluated in real-world settings where medication effects may be modulated by patient demographics and comorbidities as well as the usage of other concomitant medications. OBJECTIVE: We demonstrate a new, high-throughput method leveraging electronic health records (EHRs) and the Surescripts pharmacy database to monitor real-world QTc-prolonging medication and potential interacting effects from demographics and comorbidities. METHODS: We included all outpatient electrocardiograms (ECGs) from September 2008 to December 2019 at a large academic medical system, which were in sinus rhythm with a heart rate of 40-100 beats per minute, QRS duration of <120 milliseconds, and QTc of 300-700 milliseconds, determined using the Bazett formula. We used prescription information from the Surescripts pharmacy database and EHR medication lists to classify whether a patient was on a medication during an ECG. Negative control ECGs were obtained from patients not currently on the medication but who had been or would be on that medication within 1 year. We calculated the difference in mean QTc between ECGs of patients who are on and those who are off a medication and made comparisons to known medication TdP risks per the CredibleMeds.org database. Using linear regression analysis, we studied the interaction of patient-level demographics or comorbidities on medication-related QTc prolongation. RESULTS: We analyzed the effects of 272 medications on 310,335 ECGs from 159,397 individuals. Medications associated with the greatest QTc prolongation were dofetilide (mean QTc difference 21.52, 95% CI 10.58-32.70 milliseconds), mexiletine (mean QTc difference 18.56, 95% CI 7.70-29.27 milliseconds), amiodarone (mean QTc difference 14.96, 95% CI 13.52-16.33 milliseconds), rifaximin (mean QTc difference 14.50, 95% CI 12.12-17.13 milliseconds), and sotalol (mean QTc difference 10.73, 95% CI 7.09-14.37 milliseconds). Several top QT prolonging medications such as rifaximin, lactulose, cinacalcet, and lenalidomide were not previously known but have plausible mechanistic explanations. Significant interactions were observed between demographics or comorbidities and QTc prolongation with many medications, such as coronary disease and amiodarone. CONCLUSIONS: We demonstrate a new, high-throughput technique for monitoring real-world effects of QTc-prolonging medications from readily accessible clinical data. Using this approach, we confirmed known medications for QTc prolongation and identified potential new associations and demographic or comorbidity interactions that could supplement findings in curated databases. Our single-center results would benefit from additional verification in future multisite studies that incorporate larger numbers of patients and ECGs along with more precise medication adherence and comorbidity data.

Deep Learning of Electrocardiograms in Sinus Rhythm From US Veterans to Predict Atrial Fibrillation.

Importance: Early detection of atrial fibrillation (AF) may help prevent adverse cardiovascular events such as stroke. Deep learning applied to electrocardiograms (ECGs) has been successfully used for early identification of several cardiovascular diseases. Objective: To determine whether deep learning models applied to outpatient ECGs in sinus rhythm can predict AF in a large and diverse patient population. Design, Setting, and Participants: This prognostic study was performed on ECGs acquired from January 1, 1987, to December 31, 2022, at 6 US Veterans Affairs (VA) hospital networks and 1 large non-VA academic medical center. Participants included all outpatients with 12-lead ECGs in sinus rhythm. Main Outcomes and Measures: A convolutional neural network using 12-lead ECGs from 2 US VA hospital networks was trained to predict the presence of AF within 31 days of sinus rhythm ECGs. The model was tested on ECGs held out from training at the 2 VA networks as well as 4 additional VA networks and 1 large non-VA academic medical center. Results: A total of 907 858 ECGs from patients across 6 VA sites were included in the analysis. These patients had a mean (SD) age of 62.4 (13.5) years, 6.4% were female, and 93.6% were male, with a mean (SD) CHA2DS2-VASc (congestive heart failure, hypertension, age, diabetes mellitus, prior stroke or transient ischemic attack or thromboembolism, vascular disease, age, sex category) score of 1.9 (1.6). A total of 0.2% were American Indian or Alaska Native, 2.7% were Asian, 10.7% were Black, 4.6% were Latinx, 0.7% were Native Hawaiian or Other Pacific Islander, 62.4% were White, 0.4% were of other race or ethnicity (which is not broken down into subcategories in the VA data set), and 18.4% were of unknown race or ethnicity. At the non-VA academic medical center (72 483 ECGs), the mean (SD) age was 59.5 (15.4) years and 52.5% were female, with a mean (SD) CHA2DS2-VASc score of 1.6 (1.4). A total of 0.1% were American Indian or Alaska Native, 7.9% were Asian, 9.4% were Black, 2.9% were Latinx, 0.03% were Native Hawaiian or Other Pacific Islander, 74.8% were White, 0.1% were of other race or ethnicity, and 4.7% were of unknown race or ethnicity. A deep learning model predicted the presence of AF within 31 days of a sinus rhythm ECG on held-out test ECGs at VA sites with an area under the receiver operating characteristic curve (AUROC) of 0.86 (95% CI, 0.85-0.86), accuracy of 0.78 (95% CI, 0.77-0.78), and F1 score of 0.30 (95% CI, 0.30-0.31). At the non-VA site, AUROC was 0.93 (95% CI, 0.93-0.94); accuracy, 0.87 (95% CI, 0.86-0.88); and F1 score, 0.46 (95% CI, 0.44-0.48). The model was well calibrated, with a Brier score of 0.02 across all sites. Among individuals deemed high risk by deep learning, the number needed to screen to detect a positive case of AF was 2.47 individuals for a testing sensitivity of 25% and 11.48 for 75%. Model performance was similar in patients who were Black, female, or younger than 65 years or who had CHA2DS2-VASc scores of 2 or greater. Conclusions and Relevance: Deep learning of outpatient sinus rhythm ECGs predicted AF within 31 days in populations with diverse demographics and comorbidities. Similar models could be used in future AF screening efforts to reduce adverse complications associated with this disease.

Implementation of an Electronic Health Records-Based Safe Contrast Limit for Preventing Contrast-Associated Acute Kidney Injury After Percutaneous Coronary Intervention.

BACKGROUND: Contrast-associated acute kidney injury (CA-AKI) after percutaneous coronary intervention is associated with increased mortality. We assessed the effectiveness of an electronic health records safe contrast limit tool in predicting CA-AKI risk and reducing contrast use and CA-AKI. METHODS: We created an alert displaying the safe contrast limit to cardiac catheterization laboratory staff prior to percutaneous coronary intervention. The alert used risk factors automatically extracted from the electronic health records. We included procedures from June 1, 2020 to October 1, 2021; the intervention went live February 10, 2021. Using difference-in-differences analysis, we evaluated changes in contrast volume and CA-AKI rates after contrast limit tool implementation compared to control hospitals. Cardiologists were surveyed prior to and 9 months after alert implementation on beliefs, practice patterns, and safe contrast estimates for example patients. RESULTS: At the one intervention site, there were 508 percutaneous coronary interventions before and 531 after tool deployment. At 15 control sites, there were 3550 and 3979 percutaneous coronary interventions, respectively. The contrast limit predicted CA-AKI with an accuracy of 64.1%, negative predictive value of 93.3%, and positive predictive value of 18.7%. After implementation, in high/modifiable risk patients (defined as having a calculated contrast limit <500ml) there was a small but significant -4.60 mL/month (95% CI, -8.24 to -1.00) change in average contrast use but no change in CA-AKI rates (odds ratio, 0.96 [95% CI, 0.84-1.10]). Low-risk patients had no change in contrast use (-0.50 mL/month [95% CI, -7.49 to 6.49]) or CA-AKI (odds ratio, 1.24 [95% CI, 0.79-1.93]). In assessing CA-AKI risk, clinicians heavily weighted age and diabetes but often did not consider anemia, cardiogenic shock, and heart failure. CONCLUSIONS: Clinicians often used a simplified assessment of CA-AKI risk that did not include important risk factors, leading to risk estimations inconsistent with established models. Despite clinician skepticism, an electronic health records-based contrast limit tool more accurately predicted CA-AKI risk and was associated with a small decrease in contrast use during percutaneous coronary intervention but no change in CA-AKI rates.

Prediction of Coronary Artery Calcium Using Deep Learning of Echocardiograms.

BACKGROUND: Coronary artery calcification (CAC), often assessed by computed tomography (CT), is a powerful marker of coronary artery disease that can guide preventive therapies. Computed tomographies, however, are not always accessible or serially obtainable. It remains unclear whether other widespread tests such as transthoracic echocardiograms (TTEs) can be used to predict CAC. METHODS: Using a data set of 2,881 TTE videos paired with coronary calcium CTs, we trained a video-based artificial intelligence convolutional neural network to predict CAC scores from parasternal long-axis views. We evaluated the model's ability to classify patients from a held-out sample as well as an external site sample into zero CAC and high CAC (CAC ≥ 400 Agatston units) groups by receiver operating characteristic and precision-recall curves. We also investigated whether such classifications prognosticated significant differences in 1-year mortality rates by the log-rank test of Kaplan-Meier curves. RESULTS: Transthoracic echocardiogram artificial intelligence models had high discriminatory abilities in predicting zero CAC (receiver operating characteristic area under the curve [AUC] = 0.81 [95% CI, 0.74-0.88], F1 score = 0.95) and high CAC (AUC = 0.74 [0.68-0.8], F1 score = 0.74). This performance was confirmed in an external test data set of 92 TTEs (AUC = 0.75 [0.65-0.85], F1 score = 0.77; and AUC = 0.85 [0.76-0.93], F1 score = 0.59, respectively). Risk stratification by TTE-predicted CAC performed similarly to CT CAC scores in prognosticating significant differences in 1-year survival in high-CAC patients (CT CAC ≥ 400 vs CT CAC < 400, P = .03; TTE-predicted CAC ≥ 400 vs TTE-predicted CAC < 400, P = .02). CONCLUSIONS: A video-based deep learning model successfully used TTE videos to predict zero CAC and high CAC with high accuracy. Transthoracic echocardiography-predicted CAC prognosticated differences in 1-year survival similar to CT CAC. Deep learning of TTEs holds promise for future adjunctive coronary artery disease risk stratification to guide preventive therapies.

Deep learning-based electrocardiographic screening for chronic kidney disease.

BACKGROUND: Undiagnosed chronic kidney disease (CKD) is a common and usually asymptomatic disorder that causes a high burden of morbidity and early mortality worldwide. We developed a deep learning model for CKD screening from routinely acquired ECGs. METHODS: We collected data from a primary cohort with 111,370 patients which had 247,655 ECGs between 2005 and 2019. Using this data, we developed, trained, validated, and tested a deep learning model to predict whether an ECG was taken within one year of the patient receiving a CKD diagnosis. The model was additionally validated using an external cohort from another healthcare system which had 312,145 patients with 896,620 ECGs between 2005 and 2018. RESULTS: Using 12-lead ECG waveforms, our deep learning algorithm achieves discrimination for CKD of any stage with an AUC of 0.767 (95% CI 0.760-0.773) in a held-out test set and an AUC of 0.709 (0.708-0.710) in the external cohort. Our 12-lead ECG-based model performance is consistent across the severity of CKD, with an AUC of 0.753 (0.735-0.770) for mild CKD, AUC of 0.759 (0.750-0.767) for moderate-severe CKD, and an AUC of 0.783 (0.773-0.793) for ESRD. In patients under 60 years old, our model achieves high performance in detecting any stage CKD with both 12-lead (AUC 0.843 [0.836-0.852]) and 1-lead ECG waveform (0.824 [0.815-0.832]). CONCLUSIONS: Our deep learning algorithm is able to detect CKD using ECG waveforms, with stronger performance in younger patients and more severe CKD stages. This ECG algorithm has the potential to augment screening for CKD.

Chronic kidney disease (CKD) is a common condition involving loss of kidney function over time and results in a substantial number of deaths. However, CKD often has no symptoms during its early stages. To detect CKD earlier, we developed a computational approach for CKD screening using routinely acquired electrocardiograms (ECGs), a cheap, rapid, non-invasive, and commonly obtained test of the heart's electrical activity. Our model achieved good accuracy in identifying any stage of CKD, with especially high accuracy in younger patients and more severe stages of CKD. Given the high global burden of undiagnosed CKD, novel and accessible CKD screening strategies have the potential to help prevent disease progression and reduce premature deaths related to CKD.

A deep learning-based electrocardiogram risk score for long term cardiovascular death and disease.

The electrocardiogram (ECG) is the most frequently performed cardiovascular diagnostic test, but it is unclear how much information resting ECGs contain about long term cardiovascular risk. Here we report that a deep convolutional neural network can accurately predict the long-term risk of cardiovascular mortality and disease based on a resting ECG alone. Using a large dataset of resting 12-lead ECGs collected at Stanford University Medical Center, we developed SEER, the Stanford Estimator of Electrocardiogram Risk. SEER predicts 5-year cardiovascular mortality with an area under the receiver operator characteristic curve (AUC) of 0.83 in a held-out test set at Stanford, and with AUCs of 0.78 and 0.83 respectively when independently evaluated at Cedars-Sinai Medical Center and Columbia University Irving Medical Center. SEER predicts 5-year atherosclerotic disease (ASCVD) with an AUC of 0.67, similar to the Pooled Cohort Equations for ASCVD Risk, while being only modestly correlated. When used in conjunction with the Pooled Cohort Equations, SEER accurately reclassified 16% of patients from low to moderate risk, uncovering a group with an actual average 9.9% 10-year ASCVD risk who would not have otherwise been indicated for statin therapy. SEER can also predict several other cardiovascular conditions such as heart failure and atrial fibrillation. Using only lead I of the ECG it predicts 5-year cardiovascular mortality with an AUC of 0.80. SEER, used alongside the Pooled Cohort Equations and other risk tools, can substantially improve cardiovascular risk stratification and aid in medical decision making.

Confounders mediate AI prediction of demographics in medical imaging.

Deep learning has been shown to accurately assess "hidden" phenotypes from medical imaging beyond traditional clinician interpretation. Using large echocardiography datasets from two healthcare systems, we test whether it is possible to predict age, race, and sex from cardiac ultrasound images using deep learning algorithms and assess the impact of varying confounding variables. Using a total of 433,469 videos from Cedars-Sinai Medical Center and 99,909 videos from Stanford Medical Center, we trained video-based convolutional neural networks to predict age, sex, and race. We found that deep learning models were able to identify age and sex, while unable to reliably predict race. Without considering confounding differences between categories, the AI model predicted sex with an AUC of 0.85 (95% CI 0.84-0.86), age with a mean absolute error of 9.12 years (95% CI 9.00-9.25), and race with AUCs ranging from 0.63 to 0.71. When predicting race, we show that tuning the proportion of confounding variables (age or sex) in the training data significantly impacts model AUC (ranging from 0.53 to 0.85), while sex and age prediction was not particularly impacted by adjusting race proportion in the training dataset AUC of 0.81-0.83 and 0.80-0.84, respectively. This suggests significant proportion of AI's performance on predicting race could come from confounding features being detected. Further work remains to identify the particular imaging features that associate with demographic information and to better understand the risks of demographic identification in medical AI as it pertains to potentially perpetuating bias and disparities.

High-Throughput Precision Phenotyping of Left Ventricular Hypertrophy With Cardiovascular Deep Learning.

IMPORTANCE: Early detection and characterization of increased left ventricular (LV) wall thickness can markedly impact patient care but is limited by under-recognition of hypertrophy, measurement error and variability, and difficulty differentiating causes of increased wall thickness, such as hypertrophy, cardiomyopathy, and cardiac amyloidosis. OBJECTIVE: To assess the accuracy of a deep learning workflow in quantifying ventricular hypertrophy and predicting the cause of increased LV wall thickness. DESIGN, SETTINGS, AND PARTICIPANTS: This cohort study included physician-curated cohorts from the Stanford Amyloid Center and Cedars-Sinai Medical Center (CSMC) Advanced Heart Disease Clinic for cardiac amyloidosis and the Stanford Center for Inherited Cardiovascular Disease and the CSMC Hypertrophic Cardiomyopathy Clinic for hypertrophic cardiomyopathy from January 1, 2008, to December 31, 2020. The deep learning algorithm was trained and tested on retrospectively obtained independent echocardiogram videos from Stanford Healthcare, CSMC, and the Unity Imaging Collaborative. MAIN OUTCOMES AND MEASURES: The main outcome was the accuracy of the deep learning algorithm in measuring left ventricular dimensions and identifying patients with increased LV wall thickness diagnosed with hypertrophic cardiomyopathy and cardiac amyloidosis. RESULTS: The study included 23 745 patients: 12 001 from Stanford Health Care (6509 [54.2%] female; mean [SD] age, 61.6 [17.4] years) and 1309 from CSMC (808 [61.7%] female; mean [SD] age, 62.8 [17.2] years) with parasternal long-axis videos and 8084 from Stanford Health Care (4201 [54.0%] female; mean [SD] age, 69.1 [16.8] years) and 2351 from CSMS (6509 [54.2%] female; mean [SD] age, 69.6 [14.7] years) with apical 4-chamber videos. The deep learning algorithm accurately measured intraventricular wall thickness (mean absolute error [MAE], 1.2 mm; 95% CI, 1.1-1.3 mm), LV diameter (MAE, 2.4 mm; 95% CI, 2.2-2.6 mm), and posterior wall thickness (MAE, 1.4 mm; 95% CI, 1.2-1.5 mm) and classified cardiac amyloidosis (area under the curve [AUC], 0.83) and hypertrophic cardiomyopathy (AUC, 0.98) separately from other causes of LV hypertrophy. In external data sets from independent domestic and international health care systems, the deep learning algorithm accurately quantified ventricular parameters (domestic: R2, 0.96; international: R2, 0.90). For the domestic data set, the MAE was 1.7 mm (95% CI, 1.6-1.8 mm) for intraventricular septum thickness, 3.8 mm (95% CI, 3.5-4.0 mm) for LV internal dimension, and 1.8 mm (95% CI, 1.7-2.0 mm) for LV posterior wall thickness. For the international data set, the MAE was 1.7 mm (95% CI, 1.5-2.0 mm) for intraventricular septum thickness, 2.9 mm (95% CI, 2.4-3.3 mm) for LV internal dimension, and 2.3 mm (95% CI, 1.9-2.7 mm) for LV posterior wall thickness. The deep learning algorithm accurately detected cardiac amyloidosis (AUC, 0.79) and hypertrophic cardiomyopathy (AUC, 0.89) in the domestic external validation site. CONCLUSIONS AND RELEVANCE: In this cohort study, the deep learning model accurately identified subtle changes in LV wall geometric measurements and the causes of hypertrophy. Unlike with human experts, the deep learning workflow is fully automated, allowing for reproducible, precise measurements, and may provide a foundation for precision diagnosis of cardiac hypertrophy.