Citations

Abstract Aims Spectrum bias can arise when a diagnostic test is derived from study populations with different disease spectra than the target population, resulting in poor generalizability. We used a real-world artificial intelligence (AI)-derived algorithm to detect severe aortic stenosis (AS) to experimentally assess the effect of spectrum bias on test performance. Methods and results All adult patients at the Mayo Clinic between 1 January 1989 and 30 September 2019 with transthoracic echocardiograms within 180 days after electrocardiogram (ECG) were identified. Two models were developed from two distinct patient cohorts: a whole-spectrum cohort comparing severe AS to any non-severe AS and an extreme-spectrum cohort comparing severe AS to no AS at all. Model performance was assessed. Overall, 258 607 patients had valid ECG and echocardiograms pairs. The area under the receiver operator curve was 0.87 and 0.91 for the whole-spectrum and extreme-spectrum models, respectively. Sensitivity and specificity for the whole-spectrum model was 80% and 81%, respectively, while for the extreme-spectrum model it was 84% and 84%, respectively. When applying the AI-ECG derived from the extreme-spectrum cohort to patients in the whole-spectrum cohort, the sensitivity, specificity, and area under the curve dropped to 83%, 73%, and 0.86, respectively. Conclusion While the algorithm performed robustly in identifying severe AS, this study shows that limiting datasets to clearly positive or negative labels leads to overestimation of test performance when testing an AI algorithm in the setting of classifying severe AS using ECG data. While the effect of the bias may be modest in this example, clinicians should be aware of the existence of such a bias in AI-derived algorithms.

AbstractIntroductionThe efficacy of cardiac resynchronization therapy (CRT) has been widely studied in the medical literature; however, about 30% of candidates fail to respond to this treatment strategy. Smart computational approaches based on clinical data can help expose hidden patterns useful for identifying CRT responders.MethodsWe retrospectively analyzed the electronic health records of 1664 patients who underwent CRT procedures from January 1, 2002 to December 31, 2017. An ensemble of ensemble (EoE) machine learning (ML) system composed of a supervised and an unsupervised ML layers was developed to generate a prediction model for CRT response.ResultsWe compared the performance of EoE against traditional ML methods and the state‐of‐the‐art convolutional neural network (CNN) model trained on raw electrocardiographic (ECG) waveforms. We observed that the models exhibited improvement in performance as more features were incrementally used for training. Using the most comprehensive set of predictors, the performance of the EoE model in terms of the area under the receiver operating characteristic curve and F1‐score were 0.76 and 0.73, respectively. Direct application of the CNN model on the raw ECG waveforms did not generate promising results.ConclusionThe proposed CRT risk calculator effectively discriminates which heart failure (HF) patient is likely to respond to CRT significantly better than using clinical guidelines and traditional ML methods, thus suggesting that the tool can enhanced care management of HF patients by helping to identify high‐risk patients.

Introduction: Although the vast majority of mitral valve prolapse (MVP) is benign, women with bileaflet MVP (biMVP), complex ventricular ectopy (VE), and abnormal T waves may comprise the recently described malignant biMVP syndrome. The mechanism of ventricular arrhythmia is unknown. To further characterize the arrhythmic substrate, we reviewed our center’s ablation experience in 6 biMVP patients with prior cardiac arrest and recurrent ICD shocks for drug refractory ventricular fibrillation (VF). Methods and Results: Six women with biMVP (median age 31.5 [range 24.2 - 58.7] years, EF 65 [45 - 67]%, all ≤moderate mitral regurgitation) experienced 6 (3 - 25) appropriate ICD shocks over 4.8 (2.8 - 10.7) years and underwent index ablation between 2/2007 - 10/2013. All had multiple VE morphologies (median 7 [3 - 24]) with variable coupling intervals but with a predominant VE trigger for the VF. A median 2 (1 - 4) VE foci were ablated. Sites of successful ablation of VF-triggering and other dominant VE were left ventricular papillary muscles [PM] (1 anterior, 1 posterior, 1 both), fascicles (1 anterior, 1 posterior), or both (1 both PM and posterior fascicle). Outflow tract VE was also present and targeted (1 left, 1 right)i. Two underwent repeat ablation (288 and 312 days) for recurrent complex VE without shocks, with different foci to the index ablation (1 posterior fascicle, 1 both fascicles). The VF-triggering VE in all patients was confirmed as originating from within the left fascicular system, which in 3/6 was at a papillary muscle. Acute procedural success was seen in all with no complications to date. A VF storm occurred within 24 hours of ablation in a single patient. At follow-up of a mean 662 (47 - 2099) days, 1 patient received a single shock (p=0.03 vs. preablation). Symptomatic VE was reduced in all; while 3/6 continue Class 1c antiarrhythmics and 5/6 have beta blockade. Conclusion: Malignant biMVP syndrome is characterized by fascicular and papillary muscle PVCs that trigger ventricular fibrillation, yet in all patients, the VE is multifocal. Ablation of at least one focus appears to improve symptoms and reduce shocks.

Introduction: The availability, national scope and perceived unbiased nature of administrative claims data makes it an attractive choice by health care policy makers for evaluating quality and outcomes. We calculated Ventricular tachycardia (VT) ablation complication rates from national administrative data and compared them with data from randomized trials and prospective studies. Methods: A PUBMED database search was done to identify existing studies of VT ablation. Studies were divided into clinical (randomized controlled trials (RCTs) and prospective studies). Pooled rates of vascular, cardiac, neurological complications and procedure related mortality were computed and compared to reported complication rates from administrative database studies Results: We identified two administrative database studies of VT ablation that included a total of 14,352 procedures and 18 clinical studies with a total of 1,705 procedure. Overall complication rate in administrative studies was higher than in clinical studies (9.39% vs. 7.97%). Rates of cardiac/pericardial (4.47% vs. 2.29%) and vascular access complications (6.9% vs. 3.0%) were higher in real world studies. Neurological complications (1.17% vs. 0.45%) and procedure related mortality (1.53% vs. 1.29%) were higher in clinical studies. Among clinical studies, RCTs had the lowest reported complication rate. Conclusion: Despite substantial improvement, “real-world” complication rates from administrative data remain higher than those reported in RCTs. The discrepancy may be attributable to differences in operator skills, selection bias and variation in data quality. Use of large databases provides an important analytical resource as we embark on improving national health care efficacy and safety in the current challenging environment.

Introduction: Long QT syndrome (LQTS) is a genetic disorder associated with abnormal ventricular repolarization and a predisposition for torsade de pointes and sudden cardiac death. Identification of markers to elucidate high risk patients is of major practical importance. Our aim was to use a novel, automated quantitative T wave analysis program to distinguish symptomatic from asymptomatic LQTS patients. Methods: We analyzed a genotyped cohort of 420 patients (22 ± 15.7years, 43% male) with either LQT1 (61%) or LQT2 (39%). ECG analysis was conducted using novel software that we created to automatically detect subtle changes in T wave morphology. Symptomatic patients were defined as those with a history of syncope or resuscitated cardiac arrest that was suspected of being LQTS triggered. Classification was performed using the top three T wave features in a linear discriminant classifier by 10x10 cross-validation. Results: Symptomatic LQT1 patients have a greater QTc duration in lead aVF (437 msec vs 416 msec, p <0.0001), a greater T wave center of gravity on the x axis (COGx) in lead V2 (0.304 vs 0.280, p <0.0001) and longer Tpeak-Tend/QT in lead II (0.208 vs 0.188, p < 0.0001) than asymptomatic patients with LQT1. Use of these 3 features enabled classification of symptomatic status in 72% of LQT1 cases, compared to 68% when QTc was used to classify alone. Symptomatic LQT2 patients had a longer QTc in Lead 1 (460 vs 418 msec, p < 0.0001), COGx in 4th segment of T wave was greater in lead V5 (0.399 vs 0.350, p<0.0001), and the Tpeak-Tend interval in V2 was longer (126 vs 99 msec, P =0.0004). Use of these 3 features enabled correct classification in 73% of LQT2 cases, compared to 67% when QTc was used alone. Conclusions: Application of an automated T wave analysis program was able to identify T wave features that distinguished the patients who presented as symptomatic LQTS, associated with syncope or resuscitated cardiac arrest. Use of these features could help assess arrhythmic risk and guide management strategies for patients with LQTS.

Introduction: As the US population ages and indications for cardiac device implantation expand, there is a concomitant increase in patients undergoing lead removal/extraction (LR/E).. We sought to examine trends in outcomes and complications associated with LR/E. Methods: We conducted a retrospective analysis using data from a large U.S. commercial insurance database (privately insured and Medicare Advantage patients) identifying all patients undergoing LR/E between 2004 and 2013 with at least one year continuous medical plan enrollment prior to LR/E (to identify baseline comorbidities). Using ICD9 codes and CPT codes, major complications (hemorrhage requiring transfusion, vascular injury, pericardial complications, respiratory complications, cerebrovascular complications, and requirement for cardiac surgery) associated with LR/E and 30-day readmission rates were evaluated. Results: Among 42,878 patients who underwent LR/E (median age of 62 ( 54, 72) years, 70% male); approximately 1 in 40 experienced a major adverse event (2.4%) during the study period. Respiratory complications were the most common complication and 311 patients required cardiac surgery. 30-day readmission occurred in 11.1% of the total population. The most common reasons for readmission were heart failure exacerbation (18.7%) and device related complications (13.2%). Conclusion: The observed complication rates after LE/R in clinical practice are similar to those seen in recent clinical trials and have been relatively stable over the past 10 years.