Artificial intelligence-enabled electrocardiogram (AI-ECG) does not predict atrial fibrillation following patent foramen ovale closure

Background

Atrial fibrillation (AF) is a known complication following patent foramen ovale (PFO) closure. AI-enabled ECG (AI-ECG) acquired during normal sinus rhythm has been shown to identify individuals with AF by noting high-risk ECG features invisible to the human eye. We sought to characterize the value of AI-ECG in predicting AF development following PFO closure and investigate key clinical and procedural characteristics possibly associated with post-procedural AF.

Methods

We performed a retrospective analysis of patients who underwent PFO closure at our hospital from January 2011 to December 2022. We recorded the probability (%) of AF using the Mayo Clinic AI-ECG dashboard from pre- and post-procedure ECGs. The cut-off point of ≥ 11 %, which was found to optimally balance sensitivity and specificity in the original derivation paper (the Youden index) was used to label an AI-ECG "positive" for AF. Pre-procedural transesophageal echocardiography (TEE) and pre- and post-procedure transcranial doppler (TCD) data was also recorded.

Results

Out of 93 patients, 49 (53 %) were male, mean age was 55 ± 15 years with mean post-procedure follow up of 29 ± 3 months. Indication for PFO closure in 69 (74 %) patients was for secondary prevention of transient ischemic attack (TIA) and/or stroke. Twenty patients (22 %) developed paroxysmal AF post-procedure, with the majority within the first month post-procedure (15 patients, 75 %). Patients who developed AF were not significantly more likely to have a positive post-procedure AI-ECG than those who did not develop AF (30 % AF vs 27 % no AF, p = 0.8).Based on the PFO-Associated Stroke Causal Likelihood (PASCAL) classification, patients who had PFO closure for secondary prevention of TIA and/or stroke in the "possible" group were significantly more likely to develop AF than patients in "probable" and "unlikely" groups (p = 0.034). AF-developing patients were more likely to have post-procedure implantable loop recorder (ILR) (55 % vs 9.6 %, p < 0.001), and longer duration of ILR monitoring (121 vs 92.5 weeks, p = 0.035). There were no significant differences in TCD and TEE characteristics, device type, or device size between those who developed AF vs those who did not.

Conclusions

In this small, retrospective study, AI-ECG did not accurately distinguish patients who developed AF post-PFO closure from those who did not. Although AI-ECG has emerged as a valuable tool for risk prediction of AF, extrapolation of its performance to procedural settings such as PFO closure requires further investigation.

Promises and Perils of Consumer Mobile Technologies in Cardiovascular Care: JACC Scientific Statement

Direct-to-consumer (D2C) wearables are becoming increasingly popular in cardiovascular health management because of their affordability and capability to capture diverse health data. Wearables may enable continuous health care provider-patient partnerships and reduce the volume of episodic clinic-based care (thereby reducing health care costs). However, challenges arise from the unregulated use of these devices, including questionable data reliability, potential misinterpretation of information, unintended psychological impacts, and an influx of clinically nonactionable data that may overburden the health care system. Further, these technologies could exacerbate, rather than mitigate, health disparities. Experience with wearables in atrial fibrillation underscores these challenges. The prevalent use of D2C wearables necessitates a collaborative approach among stakeholders to ensure effective integration into cardiovascular care. Wearables are heralding innovative disease screening, diagnosis, and management paradigms, expanding therapeutic avenues, and anchoring personalized medicine.

Assessing the use of observational methods and real-world data to emulate ongoing randomized controlled trials

BACKGROUND/AIMS

There has been growing interest in better understanding the potential of observational research methods in medical product evaluation and regulatory decision-making. Previously, we used linked claims and electronic health record data to emulate two ongoing randomized controlled trials, characterizing the populations and results of each randomized controlled trial prior to publication of its results. Here, our objective was to compare the populations and results from the emulated trials with those of the now-published randomized controlled trials.

METHODS

This study compared participants' demographic and clinical characteristics and study results between the emulated trials, which used structured data from OptumLabs Data Warehouse, and the published PRONOUNCE and GRADE trials. First, we examined the feasibility of implementing the baseline participant characteristics included in the published PRONOUNCE and GRADE trials' using real-world data and classified each variable as ascertainable, partially ascertainable, or not ascertainable. Second, we compared the emulated trials and published randomized controlled trials for baseline patient characteristics (concordance determined using standardized mean differences <0.20) and results of the primary and secondary endpoints (concordance determined by direction of effect estimates and statistical significance).

RESULTS

The PRONOUNCE trial enrolled 544 participants, and the emulated trial included 2226 propensity score-matched participants. In the PRONOUNCE trial publication, one of the 32 baseline participant characteristics was listed as an exclusion criterion on ClinicalTrials.gov but was ultimately not used. Among the remaining 31 characteristics, 9 (29.0%) were ascertainable, 11 (35.5%) were partially ascertainable, and 10 (32.2%) were not ascertainable using structured data from OptumLabs. For one additional variable, the PRONOUNCE trial did not provide sufficient detail to allow its ascertainment. Of the nine variables that were ascertainable, values in the emulated trial and published randomized controlled trial were discordant for 6 (66.7%). The primary endpoint of time from randomization to the first major adverse cardiovascular event and secondary endpoints of nonfatal myocardial infarction and stroke were concordant between the emulated trial and published randomized controlled trial. The GRADE trial enrolled 5047 participants, and the emulated trial included 7540 participants. In the GRADE trial publication, 8 of 34 (23.5%) baseline participant characteristics were ascertainable, 14 (41.2%) were partially ascertainable, and 11 (32.4%) were not ascertainable using structured data from OptumLabs. For one variable, the GRADE trial did not provide sufficient detail to allow for ascertainment. Of the eight variables that were ascertainable, values in the emulated trial and published randomized controlled trial were discordant for 4 (50.0%). The primary endpoint of time to hemoglobin A1c ≥7.0% was mostly concordant between the emulated trial and the published randomized controlled trial.

CONCLUSION

Despite challenges, observational methods and real-world data can be leveraged in certain important situations for a more timely evaluation of drug effectiveness and safety in more diverse and representative patient populations.

Artificial Intelligence ECG as a Novel Screening Tool to Detect a Newly Abnormal Left Ventricular Ejection Fraction After Anthracycline-Based Cancer Therapy

AIM

Cardiotoxicity is a serious side effect of anthracycline treatment, most commonly manifesting as a reduction in left ventricular ejection fraction (LVEF). Early recognition and treatment have been advocated, but robust, convenient and cost-effective alternatives to cardiac imaging are missing. Recent developments in artificial intelligence (AI) techniques applied to electrocardiograms (ECGs) may fill this gap, but no study so far has demonstrated its merit for the detection of an abnormal LVEF after anthracycline therapy.

METHODS

Single center consecutive cohort study of all breast cancer patients with ECG and transthoracic echocardiography (TTE) evaluation before and after (neo)adjuvant anthracycline chemotherapy. Patients with HER-2-directed therapy, metastatic disease, second primary malignancy or pre-existing cardiovascular disease were excluded from the analyses as were patients with LVEF decline for reasons other than anthracycline-induced cardiotoxicity. Primary readout was the diagnostic performance of AI-ECG by area under the curve (AUC) for LVEFs <50%.

RESULTS

Of 989 consecutive female breast cancer patients, 22 developed a decline in LVEF attributed to anthracycline therapy over a follow-up time of 9.83 ± 4.2 years. After exclusion of patients who did not have an ECGs within 90 days of a TTE, 20 cases and 683 controls remained. The AI-ECG model detected an LVEF <50% and ≤35% after anthracycline therapy with an AUC of 0.93 and 0.94, respectively.

CONCLUSIONS

These data support the use of AI-ECG for cardiotoxicity screening after anthracycline-based chemotherapy. This technology could serve as a gatekeeper to more costly cardiac imaging and could enable patients to monitor themselves over long periods of time.

Simplified rules-based tool to facilitate the application of up-to-date management recommendations in cardio-oncology

BACKGROUND

Millions of cancer survivors are at risk of cardiovascular diseases, a leading cause of morbidity and mortality. Tools to potentially facilitate implementation of cardiology guidelines, consensus recommendations, and scientific statements to prevent atherosclerotic cardiovascular disease (ASCVD) and other cardiovascular diseases are limited. Thus, inadequate utilization of cardiovascular medications and imaging is widespread, including significantly lower rates of statin use among cancer survivors for whom statin therapy is indicated.

METHODS

In this methodological study, we leveraged published guidelines documents to create a rules-based tool to include guidelines, expert consensus, and medical society scientific statements relevant to point of care cardiovascular disease prevention in the cardiovascular care of cancer survivors. Any overlap, redundancy, or ambiguous recommendations were identified and eliminated across all converted sources of knowledge. The integrity of the tool was assessed with use case examples and review of subsequent care suggestions.

RESULTS

An initial selection of 10 guidelines, expert consensus, and medical society scientific statements was made for this study. Then 7 were kept owing to overlap and revisions in society recommendations over recent years. Extensive formulae were employed to translate the recommendations of 7 selected guidelines into rules and proposed action measures. Patient suitability and care suggestions were assessed for several use case examples.

CONCLUSION

A simple rules-based application was designed to provide a potential format to deliver critical cardiovascular disease best-practice prevention recommendations at the point of care for cancer survivors. A version of this tool may potentially facilitate implementing these guidelines across clinics, payers, and health systems for preventing cardiovascular diseases in cancer survivors.

TRIAL REGISTRATION

ClinicalTrials.Gov Identifier: NCT05377320.

Team principles for successful interdisciplinary research teams

Interdisciplinary research teams can be extremely beneficial when addressing difficult clinical problems. The incorporation of conceptual and methodological strategies from a variety of research disciplines and health professions yields transformative results. In this setting, the long-term goal of team science is to improve patient care, with emphasis on population health outcomes. However, team principles necessary for effective research teams are rarely taught in health professional schools. To form successful interdisciplinary research teams in cardio-oncology and beyond, guiding principles and organizational recommendations are necessary. Cardiovascular disease results in annual direct costs of $220 billion (about $680 per person in the US) and is the leading cause of death for cancer survivors, including adult survivors of childhood cancers. Optimizing cardio-oncology research in interdisciplinary research teams has the potential to aid in the investigation of strategies for saving hundreds of thousands of lives each year in the United States and mitigating the annual cost of cardiovascular disease. Despite published reports on experiences developing research teams across organizations, specialties and settings, there is no single journal article that compiles principles for cardiology or cardio-oncology research teams. In this review, recurring threads linked to working as a team, as well as optimal methods, advantages, and problems that arise when managing teams are described in the context of career development and research. The worth and hurdles of a team approach, based on practical lessons learned from establishing our multidisciplinary research team and information gleaned from relevant specialties in the development of a successful team are presented.

Patient similarity and other artificial intelligence machine learning algorithms in clinical decision aid for shared decision-making in the Prevention of Cardiovascular Toxicity (PACT): a feasibility trial design

BACKGROUND

The many improvements in cancer therapies have led to an increased number of survivors, which comes with a greater risk of consequent/subsequent cardiovascular disease. Identifying effective management strategies that can mitigate this risk of cardiovascular complications is vital. Therefore, developing computer-driven and personalized clinical decision aid interventions that can provide early detection of patients at risk, stratify that risk, and recommend specific cardio-oncology management guidelines and expert consensus recommendations is critically important.

OBJECTIVES

To assess the feasibility, acceptability, and utility of the use of an artificial intelligence (AI)-powered clinical decision aid tool in shared decision making between the cancer survivor patient and the cardiologist regarding prevention of cardiovascular disease.

DESIGN

This is a single-center, double-arm, open-label, randomized interventional feasibility study. Our cardio-oncology cohort of > 4000 individuals from our Clinical Research Data Warehouse will be queried to identify at least 200 adult cancer survivors who meet the eligibility criteria. Study participants will be randomized into either the Clinical Decision Aid Group (where patients will use the clinical decision aid in addition to current practice) or the Control Group (current practice). The primary endpoint of this study is to assess for each patient encounter whether cardiovascular medications and imaging pursued were consistent with current medical society recommendations. Additionally, the perceptions of using the clinical decision tool will be evaluated based on patient and physician feedback through surveys and focus groups. This trial will determine whether a clinical decision aid tool improves cancer survivors' medication use and imaging surveillance recommendations aligned with current medical guidelines.

TRIAL REGISTRATION

ClinicalTrials.Gov Identifier: NCT05377320.

Transformation of computerized electrocardiogram data into novel means to differentiate wide complex tachycardias

Background

Accurate automated wide QRS complex tachycardia (WCT) discrimination between ventricular tachycardia (VT) and supraventricular wide complex tachycardia (SWCT) can be accomplished using novel calculations derived from computerized electrocardiogram (ECG) data from paired WCT and baseline ECGs.

Purpose

Our aim was to develop and trial novel WCT discrimination approaches for WCT patients with and without a corresponding baseline ECG. Central to this analysis was the creation and use of a novel parameter (i.e., percent monophasic time-voltage area [PMonoTVA] [%]) that may be derived from computerized ECG measurements present on the WCT ECG alone.

Methods

In a two-part study, we derived and tested WCT differentiation models comprised of novel and previously established parameters formulated from computerized data of paired WCT and baseline ECGs. In Part 1, novel and established parameters generated from WCT and baseline ECG data were used to derive, validate, and compare five different binary classification models: (i) logistic regression [LR], (ii) artificial neural network [ANN], (iii) Random Forests [RF], (iv) support vector machine [SVM], and (v) ensemble learning (EL). In Part 2, two unique LR models were derived, validated, and compared using parameters generated from computerized data of the (i) WCT ECG alone (i.e., Solo Model) and (ii) paired WCT and baseline ECGs (i.e., Paired Model).

Results

In Part 1, among 103 patients with VT or SWCT diagnoses established from corroborating electrophysiology studies or intra-cardiac device recordings, favorable diagnostic performance was achieved by all modeling technique subtypes: LR (area under the receiver operating characteristic curve [AUC] 0.95), ANN (AUC 0.91), RF (AUC 0.97), SVM (AUC 0.98), and EL (AUC 0.97). In Part 2, among 235 patients with a VT or SWCT diagnosis established with (Gold Standard cohort) or without (Non-Gold Standard cohort) a corroborating electrophysiology procedure or intra-cardiac device recording, favorable diagnostic performance was achieved by the Solo Model (AUC 0.86) and Paired Model (AUC 0.95) (Table).

Conclusion

Accurate WCT discrimination may be accomplished using novel parameters derived from computerized data of the WCT ECG alone and paired WCT and baseline ECGs.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Institute of Health

Emulating the GRADE trial using real world data: retrospective comparative effectiveness study

OBJECTIVE

To emulate the GRADE (Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study) trial using real world data before its publication. GRADE directly compared second line glucose lowering drugs for their ability to lower glycated hemoglobin A_1c (HbA_1c).

DESIGN

Observational study.

SETTING

OptumLabs® Data Warehouse (OLDW), a nationwide claims database in the US, 25 January 2010 to 30 June 2019.

PARTICIPANTS

Adults with type 2 diabetes and HbA_1c 6.8-8.5% while using metformin monotherapy, identified according to the GRADE trial specifications, who also used glimepiride, liraglutide, sitagliptin, or insulin glargine.

MAIN OUTCOME MEASURES

The primary outcome was time to HbA_1c ≥7.0%. Secondary outcomes were time to HbA_1c >7.5%, incident microvascular complications, incident macrovascular complications, adverse events, all cause hospital admissions, and all cause mortality. Propensity scores were estimated using the gradient boosting machine method, and inverse propensity score weighting was used to emulate randomization of the treatment groups, which were then compared using Cox proportional hazards regression.

RESULTS

8252 people were identified (19.7% of adults starting the study drugs in OLDW) who met eligibility criteria for the GRADE trial (glimepiride arm=4318, liraglutide arm=690, sitagliptin arm=2993, glargine arm=251). The glargine arm was excluded from analyses owing to small sample size. Median times to HbA_1c ≥7.0% were 442 days (95% confidence interval 394 to 480 days) for glimepiride, 764 (741 to not calculable) days for liraglutide, and 427 (380 to 483) days for sitagliptin. Liraglutide was associated with lower risk of reaching HbA_1c ≥7.0% compared with glimepiride (hazard ratio 0.57, 95% confidence interval 0.43 to 0.75) and sitagliptin (0.55, 0.41 to 0.73). Results were consistent for the secondary outcome of time to HbA_1c >7.5%. No significant differences were observed among treatment groups for the remaining secondary outcomes.

CONCLUSIONS

In this emulation of the GRADE trial, liraglutide was statistically significantly more effective at maintaining glycemic control than glimepiride or sitagliptin when added to metformin monotherapy. Generating timely evidence on medical treatments using real world data as a complement to prospective trials is of value.

Screening for cardiac allograft rejection among heart transplant recipients using an electrocardiogram-based deep learning model

Background

Current approaches utilizing non-invasive methods to screen for cardiac allograft rejection (gene expression profiling and cell free DNA) have yet to be broadly integrated into heart transplant management and have shown limited discrimination (AUCs of 0.68 and 0.77, respectively). Changes in the electrocardiogram (ECG) have been reported at the time of severe cardiac rejection, including low voltages and conduction abnormalities. It remains unknown if subtle ECG changes correlating with cardiac allograft rejection can be detected earlier using machine learning methods.

Purpose

We sought to develop an artificial intelligence (AI) model to detect cardiac allograft rejection based on the 12 lead ECG.

Methods

We identified all patients who underwent a heart transplant at 3 hospital sites within a single health system from Jan 1998 through Apr 2021 and extracted digital 12-lead ECG data as well as endomyocardial biopsy pathology results from the electronic medical record. We partitioned our data into a training (80%), validation (10%), and test set (10%) based on a group stratification sampling. Each patient was present in only one set and each set had a positivity rate of 2.6% with 6,074/758/758 ECGs belonging to 1,146/140/141 unique patients in each set respectively. Cardiac allograft rejection was defined as moderate or severe acute cellular rejection based on International Society for Heart and Lung Transplantation (ISHLT) guidelines. A convolutional neural network, using the 12-lead ECG data as input, was trained with hyperparameter optimization for regularization, learning rate adjustments, and class weights. Model performance metrics were based on the test data and estimated using the final model architecture.

Results

1,587 heart transplant recipients who had at least one endomyocardial biopsy were evaluated for inclusion. We limited our sample to ECGs performed within 30 days of the biopsy date (7,590 ECGs, representing 1,425 unique patients). Our study population had a median age of 55.8 years and 28.7% were female. The median number of ECG-biopsy pairs per patient was 5. The majority of endomyocardial biopsy results were classified as none or mild rejection (97.1%), and 2.9% had moderate/severe rejection. The ECG-based AI model detected cardiac allograft rejection with an area under the receiver operative curve (AUC) of 0.84 in the test set. The sensitivity, specificity, positive and negative predictive values were 95%, 52.6%. 5.2% and 99.7% respectively.

Conclusions

An AI-ECG model appears to outperform novel non-invasive laboratory tests (gene expression profiling and cell free DNA) for detecting cardiac allograft rejection and does not require a blood draw or the additional complexities surrounding sample processing. This model relies on a readily available and relatively inexpensive test, the ECG. In addition, AI predictions can be made available within a few minutes following ECG acquisition.

Funding Acknowledgement

Type of funding sources: Private hospital(s). Main funding source(s): Mayo Clinic

Relationship between cardiovascular symptoms, health status assessment and cardiomyopathy in the obstetric population

Background

Pregnancy related cardiomyopathy is a significant cause of maternal morbidity and mortality globally. A presumed overlap between normal pregnancy-associated symptoms and clinical symptoms of cardiomyopathy contributes to delays in diagnosis and increased risk of maternal mortality.

Purpose

We sought to evaluate the association between patient-reported cardiovascular symptoms and the presence of cardiomyopathy among pregnant and postpartum patients. We hypothesize that individual cardiovascular symptoms are unrelated to the presence of cardiomyopathy. We also evaluated the use of a novel adaptation of a validated health status questionnaire in relation to cardiomyopathy.

Methods

We enrolled 48 pregnant (&gt;13 weeks) and postpartum (up to 12 months) participants in a prospective study between October 2021 and February 2022. All study participants completed a baseline questionnaire, which included current cardiovascular symptoms, an assessment of health status using an adapted version of the Kansas City Cardiomyopathy Questionnaire (KCCQ-12), followed by a resting transthoracic echocardiogram on the same day. We defined cardiomyopathy as a left ventricular ejection fraction (LVEF) &lt;50% based on 2-D echocardiography. Fisher's exact and Wilcoxon rank-sum tests were employed to evaluate the association between reported cardiovascular symptoms, the adapted KCCQ-12 (KCC-A) score, and cardiomyopathy.

Results

At the time of enrollment, 67% were pregnant and 33% postpartum. Forty-eight percent identified as White, 31% as Black, 10% as Asian, and 10% as other race. The median age was 31 years (Q1: 28, Q3: 35) and 6% had an LVEF &lt;50%. We found no statistically significant association between four reported cardiovascular symptoms (shortness of breath, orthopnea, fast breathing, and episodes of “asthma” that did not improve with inhalers or other treatment) and cardiomyopathy or medial E/e' ratio. KCC-A scores were low in the study population overall (median 52; Q1:40, Q3: 61). We demonstrated a significantly lower KCC-A score among women with LVEF &lt;50% (median 24; Q1: 15, Q3: 44) compared to women with LVEF ≥50% (median 54; Q1: 44, Q3: 61) p=0.02.

Conclusions

We showed no significant association between individual cardiovascular symptoms and cardiomyopathy in an obstetric population. However, we demonstrate for the first time that an adapted KCCQ-12 questionnaire for health status assessment could potentially identify women with a high-likelihood of cardiomyopathy during the peripartum period who may benefit from additional evaluation including echocardiography. Larger studies are needed to validate this finding.

Funding Acknowledgement

Type of funding sources: Foundation. Main funding source(s): This study was funded by a research grant from the Miami Heart Research Institute, Florida Heart Foundation. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health

Enrolling people of color to evaluate a practice intervention: lessons from the shared decision-making for atrial fibrillation (SDM4AFib) trial

Background

Trial recruitment of Black, indigenous, and people of color (BIPOC) is key for interventions that interact with socioeconomic factors and cultural norms, preferences, and values. We report on our experience enrolling BIPOC participants into a multicenter trial of a shared decision-making intervention about anticoagulation to prevent strokes, in patients with atrial fibrillation (AF).

Methods

We enrolled patients with AF and their clinicians in 5 healthcare systems (three academic medical centers, an urban/suburban community medical center, and a safety-net inner-city medical center) located in three states (Minnesota, Alabama, and Mississippi) in the United States. Clinical encounters were randomized to usual care with or without a shared decision-making tool about anticoagulation.

Analysis

We analyzed BIPOC patient enrollment by site, categorized reasons for non-enrollment, and examined how enrollment of BIPOC patients was promoted across sites.

Results

Of 2247 patients assessed, 922 were enrolled of which 147 (16%) were BIPOC patients. Eligible Black participants were significantly less likely (p < .001) to enroll (102, 11%) than trial-eligible White participants (185, 15%). The enrollment rate of BIPOC patients varied by site. The inclusion and prioritization of clinical practices that care for more BIPOC patients contributed to a higher enrollment rate into the trial. Specific efforts to reach BIPOC clinic attendees and prioritize their enrollment had lower yield.

Conclusions

Best practices to optimize the enrollment of BIPOC participants into trials that examined complex and culturally sensitive interventions remain to be developed. This study suggests a high yield from enrolling BIPOC patients from practices that prioritize their care.

Trial registration

ClinicalTrials.gov (NCT02905032).

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-022-08399-z.

Establishing an interdisciplinary research team for cardio-oncology artificial intelligence informatics precision and health equity

Study objective

A multi-institutional interdisciplinary team was created to develop a research group focused on leveraging artificial intelligence and informatics for cardio-oncology patients. Cardio-oncology is an emerging medical field dedicated to prevention, screening, and management of adverse cardiovascular effects of cancer/ cancer therapies. Cardiovascular disease is a leading cause of death in cancer survivors. Cardiovascular risk in these patients is higher than in the general population. However, prediction and prevention of adverse cardiovascular events in individuals with a history of cancer/cancer treatment is challenging. Thus, establishing an interdisciplinary team to create cardiovascular risk stratification clinical decision aids for integration into electronic health records for oncology patients was considered crucial.

Design/setting/participants

Core team members from the Medical College of Wisconsin (MCW), University of Wisconsin-Milwaukee (UWM), and Milwaukee School of Engineering (MSOE), and additional members from Cleveland Clinic, Mayo Clinic, and other institutions have joined forces to apply high-performance computing in cardio-oncology.

Results

The team is comprised of clinicians and researchers from relevant complementary and synergistic fields relevant to this work. The team has built an epidemiological cohort of ~5000 cancer survivors that will serve as a database for interdisciplinary multi-institutional artificial intelligence projects.

Conclusion

Lessons learned from establishing this team, as well as initial findings from the epidemiology cohort, are presented. Barriers have been broken down to form a multi-institutional interdisciplinary team for health informatics research in cardio-oncology. A database of cancer survivors has been created collaboratively by the team and provides initial insight into cardiovascular outcomes and comorbidities in this population.

The effect of shared decision-making for stroke prevention on treatment adherence and safety outcomes in patients with atrial fibrillation: a randomized clinical trial

Background

Guidelines promote shared decision-making (SDM) for anticoagulation in patients with atrial fibrillation (AF). We recently showed that adding a within-encounter SDM tool to usual care (UC) increases patient involvement in decision making and clinician satisfaction, without affecting encounter length.

Purpose

We aimed to estimate the extent to which use of an SDM tool affected patient adherence to the decided care plan and clinical safety endpoints.

Methods

We conducted a multi-center, encounter-level randomized trial assessing the efficacy of UC with vs. without an SDM conversation tool for use during the clinical encounter (Anticoagulation Choice, AC) in patients with non-valvular AF considering starting or reviewing anticoagulation treatment. We conducted a chart and pharmacy review, blinded to randomization status, at 10 months post-enrollment to assess primary adherence (proportion of patients who were prescribed an anticoagulant who filled their first prescription) and secondary adherence (estimated using the proportion of days for which treatment was supplied and filled [PDC] for DOAC, and as time in therapeutic range (TTR) for warfarin). We also followed for any safety outcomes (stroke [stroke or transient ischemic attack], major bleeding, or death).

Results

We enrolled 922 evaluable patient encounters (AC=463, UC=459), of which 814 (88%) had pharmacy and clinical follow-up. We found no differences between arms in either primary (78% of patients in AC filled their first prescription vs. 81% in UC) or secondary adherence to anticoagulation (see Figure, PDCDOAC was 74.1% in AC vs. 71.6% in UC; TTRwarfarin was 66.6% in AC vs. 64.4% in UC). PDCDOAC was better (65%) in AC than in UC (55%) (OR 1.49, CIs 1.00, 2.22). Safety outcomes, mostly bleeds, occurred in 13% (AC) of and 14% (UC) of participants.

Conclusions

This is the largest reported randomized trial in AF comparing usual care with and without an SDM tool to promote SDM. Although patients were more actively involved in SDM, we found no significant differences between arms in primary or secondary adherence to anticoagulation or clinical safety outcomes.

Funding Acknowledgement

Type of funding sources: Other. Main funding source(s): The trial was funded by and conducted independently of the National Heart, Lung, and Blood Institute (NHLBI) of the U.S. National Institutes of Health (RO1 HL131535-01). The funding body had no influence on the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. Figure 1

Real-world Cardiovascular Outcomes Associated With Degarelix vs Leuprolide for Prostate Cancer Treatment

IMPORTANCE

With a growing interest in the use of real-world evidence for regulatory decision-making, it is important to understand whether real-world data can be used to emulate the results of randomized clinical trials.

OBJECTIVE

To use electronic health record and administrative claims data to emulate the ongoing PRONOUNCE trial (A Trial Comparing Cardiovascular Safety of Degarelix Versus Leuprolide in Patients With Advanced Prostate Cancer and Cardiovascular Disease).

DESIGN, SETTING, AND PARTICIPANTS

This retrospective, propensity-matched cohort study included adult men with a diagnosis of prostate cancer and cardiovascular disease who initiated either degarelix or leuprolide between December 24, 2008, and June 30, 2019. Participants were commercially insured individuals and Medicare Advantage beneficiaries included in a large US administrative claims database.

EXPOSURES

Degarelix or leuprolide.

MAIN OUTCOMES AND MEASURES

The primary end point was time to first occurrence of a major adverse cardiovascular event (MACE), defined as death due to any cause, myocardial infarction, or stroke, analogous to the PRONOUNCE trial. Secondary end points were time to death due to any cause, myocardial infarction, stroke, and angina. Cox proportional hazards regression was used to evaluate primary and secondary end points.

RESULTS

A total of 32 172 men initiated degarelix or leuprolide for prostate cancer; of them, 9490 (29.5%) had cardiovascular disease, and 7800 (24.2%) met the PRONOUNCE trial eligibility criteria and were included in this study. Overall, 165 participants (2.1%) were Asian, 1390 (17.8%) were Black, 663 (8.5%) were Hispanic, and 5258 (67.4%) were White. The mean (SD) age was 74.4 (7.4) years. Among 2226 propensity score-matched patients, no significant difference was observed in the risk of MACE for patients taking degarelix vs those taking leuprolide (10.18 vs 8.60 events per 100 person-years; hazard ratio [HR], 1.18; 95% CI, 0.86-1.61). Degarelix was associated with a higher risk of death from any cause (HR, 1.48; 95% CI, 1.01-2.18) but not of myocardial infarction (HR, 1.16; 95% CI, 0.60-2.25), stroke (HR, 0.92; 95% CI, 0.45-1.85), or angina (HR, 1.36; 95% CI, 0.43-4.27).

CONCLUSIONS AND RELEVANCE

In this emulation of a clinical trial of men with cardiovascular disease undergoing treatment for prostate cancer, degarelix was not associated with a lower risk of cardiovascular events than leuprolide. Comparison of these data with PRONOUNCE trial results, when published, will help enhance our understanding of the appropriate role of using real-world data to emulate clinical trials.

Using ensemble of ensemble machine learning methods to predict outcomes of cardiac resynchronization

INTRODUCTION

The 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.

METHODS

We 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.

RESULTS

We 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.

CONCLUSION

The 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.

Conversion of left atrial volume to diameter for automated estimation of sudden cardiac death risk in hypertrophic cardiomyopathy

Background

A subset of patients with hypertrophic cardiomyopathy (HCM) is at high risk of sudden cardiac death (SCD). Practice guidelines endorse use of a risk calculator, which requires entry of left atrial (LA) diameter. However, American Society of Echocardiography (ASE) guidelines recommend the use of LA volume index (LAVI) for routine quantification of LA size. The aims of this study were to (a) develop a model to estimate LA diameter from LAVI and (b) evaluate whether substitution of measured LA diameter by estimated LA diameter derived from LAVI reclassifies HCM‐SCD risk.

Methods

The study cohort was comprised of 500 randomly selected HCM patients who underwent transthoracic echocardiography (TTE). LA diameter and LAVI were measured offline using digital clips from TTE. Linear regression models were developed to estimate LA diameter from LAVI. A European Society of Cardiology endorsed equation estimated SCD risk, which was measured using LA diameter and estimated LA diameter derived from LAVI.

Results

The mean LAVI was 48.5 ± 18.8 mL/m². The derived LA diameter was 45.1 mm (SD: 5.5 mm), similar to the measured LA diameter (45.1 mm, SD: 7.1 mm). Median SCD risk at 5 years estimated by measured LA diameter was 2.22% (interquartile range (IQR): 1.39, 3.56), while median risk calculated by estimated LA diameter was 2.18% (IQR: 1.44, 3.52). 476/500 (95%) patients maintained the same risk classification regardless of whether the measured or estimated LA diameter was used.

Conclusions

Substitution of measured LA diameter by estimated LA diameter in the HCM‐SCD calculator did not reclassify risk.

External validation of an electrocardiography artificial intelligence-generated algorithm to detect left ventricular systolic function in a general cardiac clinic in Uganda

Background

Left ventricular systolic dysfunction (LVSD) is associated with increased morbidity and mortality. Although there are effective treatments for patients with LVSD to prevent mortality, heart failure and to improve symptoms, many patients remain undetected and untreated. We have recently derived a deep learning algorithm to detect LVSD using the electrocardiogram (ECG) which could have an important screening role, particularly in limited resources settings. We evaluated the accuracy of this algorithm for the first time in Africa in a sample of subjects attending a cardiology clinic.

Methods

We conducted a retrospective study in a general cardiac clinic in Uganda. Consecutive patients ≥18 years who had a digital ECG and echocardiogram done within two days of each other were included. We excluded patients with pacemakers or missing information regarding left ventricular ejection fraction (LVEF). Routine 10-second, twelve-lead surface rest ECG were performed using an Edan PC ECG Model SE-1515, Hamburg, Germany. The probability of LVSD was estimated with the Mayo Clinic artificial intelligence (AI) ECG algorithm. LVEF was calculated by the MMode (Teichholz method) using a Philips Ultrasound system, HD7XE, Bothel, Washington, USA. LVSD was defined as a LVEF≤35%. We assessed the overall diagnostic performance of the algorithm to identify LVSD in this population with the area under the receiver operating curve (AUC), and estimated sensitivity, specificity and accuracy using a pre-specified cut-off based on the probability for LVSD generated by the algorithm. We conducted secondary analyses using different LVEF cutoff values.

Results

We included 634 subjects, 32% (200) of whom had hypertension and 12% (77) clinical heart failure. Mean age was 57±18.8 years, 58% were women and the overall prevalence of LVSD was 4%. The AI-ECG had an AUC of 0.866 (see figure below), sensitivity 73.08%, specificity 91.10%, negative predictive value 98.75%, positive predictive value 26.03% and an accuracy of 90.96% using the original threshold. Using the optimal cutoff based on the AUCs, the sensitivity was 80.77% and specificity was 81.05% with a negative predictive value of 98.99%. The ROC for the detection of LVEF of 40% or below was 0.821.

Conclusion

The Mayo AI-ECG algorithm demonstrated good accuracy, sensitivity and specificity to detect LVSD in patients seen in a clinical setting in Uganda. This tool may facilitate the identification of people at a high risk for LVSD in settings with low resources.

ROC

Funding Acknowledgement

Type of funding source: None

Artificial intelligence-enabled detection of paroxysmal atrial fibrillation from normal sinus ECGs in patients with coronary microvascular dysfunction

Background

Artificial Intelligence (AI) algorithms enabled the detection of patients with paroxysmal atrial fibrillation (PAF) from a single normal sinus rhythm (NSR) ECG. Coronary microvascular dysfunction (CMD) is a precursor for coronary artery disease, which is a known risk factor for AF.

Purpose

The aim of this study is to examine the probability of PAF, according to AI-enabled algorithm estimation, in patients with CMD.

Methods

1858 patients without persistent atrial fibrillation with signs and/or symptoms of ischemia and with non-obstructive CAD (&lt;40% stenosis) who underwent invasive coronary microvascular functional assessment and the ECG closest to the functional assessment were included in this analysis. Patients with coronary flow velocity reserve (CFR) &lt;2 in response to adenosine were labelled as endothelial-independent CMD; % increase in coronary blood flow (%ΔCBF) &lt;50% in response to acetylcholine were labelled as endothelial-dependent CMD. Patients were categorized into 4 groups. G1: Normal (NL) CFR/NL %ΔCBF; G2: Abnormal (ABN) %ΔCBF only; G3: ABN CFR only; G4: ABL CFR & %ΔCBF. The probability of having PAF (%probAF) was calculated by a previously-trained and validated AI algorithm. AF Flag = %probAF &gt;9%; which is a pre-set cut-off found to have the highest accuracy of identifying patients with PAF (Area Under the Curve = 0.87).

Results

Mean age for patients was 51.2±12.4 and 66.3% were females. 835 (45%) were in G1, 39 (2%) in G2, 911 (49%) in G3, and 73 (4%) in G4. Compared to G1 and G2, G3 and G4 were older, had more diabetes and higher smoking rates (p&lt;0.05 for all). Furthermore, G4 had a significantly higher %probAF compared to other groups (Fig. 1). G4 were also more likely to be flagged by the algorithm as having PAF, even after adjusting for cardiovascular risk factors, with an odds ratio of 1.9 [CI 95% 1.1–3.3; p=0.03]) (Fig. 2).

Conclusion

Patients with combined CMD have a significantly higher probability of having PAF based on an AI-enabled algorithm. Further research is warranted to know if patients with CMD would benefit from formal AF screening at the time of diagnosis.

Funding Acknowledgement

Type of funding source: None

Extracting hypertrophic cardiomyopathy features from cardiac magnetic resonance reports by natural language processing

Objective

Determine if information regarding hypertrophic cardiomyopathy (HCM) can be accurately retrieved from cardiac magnetic resonance (CMR) reports using natural language processing (NLP).

Background

CMR imaging is used for diagnosis and risk stratification of HCM. Manual annotation of information from CMR is time-consuming. NLP is an artificial intelligence method for automating extraction of information from narrative text.

Methods

We identified 200 HCM patients who had CMR reports from 1998 to 2018. These patients were randomly allocated into training (100 patients with 185 CMR reports) and testing sets (100 patients with 206 reports). An NLP system with 2 tiers was developed; the first extracted information regarding HCM diagnosis while second extracted categorical or numeric concepts for HCM classification. NLP performance was compared with gold-standard manual annotation.

Results

NLP algorithms achieved very high performance across all concepts with mean positive predictive value (PPV) = 0.96. An outlier was the performance for abstracting the presence of an apical pouch from CMR reports, which had noticeably lower PPV= 0.78 which be attributed to the low number of cases with this finding.

Conclusions

The algorithms developed can be translated to clinical decision support systems to increase efficiency and contribute to improved quality of care.

Funding Acknowledgement

Type of funding source: Other. Main funding source(s): Study supported by the National Heart, Lung and Blood Institute of National Institutes of Health (K01HL124045), the Mayo Clinic Center for Clinical and Translational Science (CCaTS), and the Mayo Clinic K2R award. Content is solely the responsibility of authors and does not necessarily represent official views of the National Institutes of Health.

The association of artificial intelligence-enabled electrocardiogram-derived age (physiologic age) with atherosclerotic cardiovascular events in the community

Background

We have demonstrated that artificial intelligence interpretation of ECGs (AI-ECG) can estimate an individual's physiologic age and that the gap between AI-ECG and chronologic age (Age-Gap) is associated with increased mortality. We hypothesized that Age-Gap would predict long-term atherosclerotic cardiovascular disease (ASCVD) and that Age-Gap would refine the ACC/AHA Pooled Cohort Equations' (PCE) predictive abilities.

Methods

Using the Rochester Epidemiology Project (REP) we evaluated a community-based cohort of consecutive patients seeking primary care between 1998–2000 and followed through March 2016. Inclusion criteria were age 40–79 and complete data to calculate PCE. We excluded those with known ASCVD, AF, HF or an event within 30 days of baseline.A neural network, trained, validated, and tested in an independent cohort of ∼ 500,000 independent patients, using 10-second digital samples of raw, 12 lead ECGs. PCE was categorized as low&lt;5%, intermediate 5–9.9%, high 10–19.9%, and very high≥20%. The primary endpoint was ASCVD and included fatal and non-fatal myocardial infarction and ischemic stroke; the secondary endpoint also included coronary revascularization [Percutaneous Coronary Intervention (PCI) or Coronary Artery Bypass Graft (CABG)], TIA and Cardiovascular mortality. Events were validated in duplicate. Follow-up was truncated at 10 years for PCE analysis. The association between Age-Gap with ASCVD and expanded ASCVD was assessed with cox proportional hazard models that adjusted for chronological age, sex and risk factors. Models were stratified by PCE risk categories to evaluate the effect of PCE predicted risk.

Results

We included 24,793 patients (54% women, 95% Caucasian) with mean follow up of 12.6±5.1 years. 2,366 (9.5%) developed ASCVD events and 3,401 (13.7%) the expanded ASCVD. Mean chronologic age was 53.6±11.6 years and the AI-ECG age was 54.5±10.9 years, R2=0.7865, p&lt;0.0001. The mean Age-Gap was 0.87±7.38 years. After adjusting for age and sex, those considered older by ECG, compared to their chronologic age had a higher risk for ASCVD when compared to those with &lt;−2 SD age gap (considered younger by ECG). (Figure 1A), with similar results when using the expanded definition of ASCVD (data not shown). Furthermore, Age-Gap enhanced predicted capabilities of the PCE among those with low 10-year predicted risk (&lt;5%): Age and sex adjusted HR 4.73, 95% CI 1.42–15.74, p-value=0.01 and among those with high predicted risk (&gt;20%) age and sex adjusted HR 6.90, 95% CI 1.98–24.08, p-value=0.0006, when comparing those older to younger by ECG respectively (Figure 1B).

Conclusion

The difference between physiologic AI-ECG age and chronologic age is associated with long-term ASCVD, and enhances current risk calculators (PCE) ability to identify high and low risk individuals. This may help identify individuals who should or should not be treated with newer, expensive risk-reducing therapies.

Funding Acknowledgement

Type of funding source: Foundation. Main funding source(s): Mayo Clinic

Year in Review in Cardiac Electrophysiology

In the past year, there have been numerous advances in our understanding of arrhythmia mechanisms, diagnosis, and new therapies. We have seen advances in basic cardiac electrophysiology with data suggesting that secretoneurin may be a biomarker for patients at risk of ventricular arrhythmias, and we have learned of the potential role of an NPR-C (natriuretic peptide receptor-C) in atrial fibrosis and the role of an atrial specific 2-pore potassium channel TASK-1 as a therapeutic target for atrial fibrillation. We have seen studies demonstrating the role of sensory neurons in sleep apnea-related atrial fibrillation and the association between bariatric surgery and atrial fibrillation ablation outcomes. Artificial intelligence applied to electrocardiography has yielded estimates of age, sex, and overall health. We have seen new tools for collection of patient-centered outcomes following catheter ablation. There have been significant advances in the ability to identify ventricular tachycardia termination sites through high-density mapping of deceleration zones. We have learned that right ventricular dysfunction may be a predictor of survival benefit after implantable cardioverter-defibrillator implantation in patients with nonischemic cardiomyopathy. We have seen further insights into the role of His bundle pacing on improving outcomes. As our understanding of cardiac laminopathies advances, we may have new tools to predict arrhythmic event rates in gene carriers. Finally, we have seen numerous advances in the treatment of arrhythmias in patients with congenital heart disease.

An artificial intelligence-enabled analysis of ECG changes after androgen deprivation therapy (ADT) for prostate cancer

e17535

Background: Androgen Deprivation Therapy (ADT) is commonly used to treat prostate cancer (PC), but has been associated with cardiac morbidity and mortality. The exact mechanisms of this association are unclear. We sought to use an artificial intelligence (AI) enabled algorithm to identify ECG changes in PC patients who received ADT compared to PC patients who did not receive ADT. Methods: From 1,000,000 ECGs performed on 210,414 patients between 1993 and 2017 at our institution, a convolutional neural network was developed to detect predictive signatures for cardiac pathologies. During this process, the ability to predict “estimated sex” of the patient was developed, with output values ranging from 0 (female) to 1 (male). We applied this algorithm to 8,619 ECGs performed on 1,057 men age 75 or younger treated with radiation for high-risk or recurrent PC at our institution, and compared estimated sex after receiving ADT (n = 1,065) to ADT-naive ECGs (n = 7,554). We correlated ECG-identified estimated sex with serum testosterone levels using Spearman rank correlation. Results: Patients who had received ADT had a mean (SD) estimated sex value of 0.81 (0.26) compared to 0.92 (0.17) for those who did not (p < 0.001). Difference between estimated sex in post-ADT ECGs and ADT-naive ECGs remained significant across age groups (Table). Decreased serum total testosterone correlated with decreased estimated sex values in men receiving ADT (R = .57, p < 0.001). Conclusions: ADT for prostate cancer is associated with changes in AI-identified ECG parameters, including lower estimated male sex after receiving ADT. Lower ECG male sex estimate was associated with decreased serum testosterone. In this study, we provide preliminary proof of concept for a potential non-invasive means of monitoring treatment effect and physiologic change using ECGs. [Table: see text]