Publications

BACKGROUND: Subtle, prognostically important ECG features may not be apparent to physicians. In the course of supervised machine learning, thousands of ECG features are identified. These are not limited to conventional ECG parameters and morphology. We aimed to investigate whether neural network-derived ECG features could be used to predict future cardiovascular disease and mortality and have phenotypic and genotypic associations.

METHODS: We extracted 5120 neural network-derived ECG features from an artificial intelligence-enabled ECG model trained for 6 simple diagnoses and applied unsupervised machine learning to identify 3 phenogroups. Using the identified phenogroups, we externally validated our findings in 5 diverse cohorts from the United States, Brazil, and the United Kingdom. Data were collected between 2000 and 2023.

RESULTS: In total, 1 808 584 patients were included in this study. In the derivation cohort, the 3 phenogroups had significantly different mortality profiles. After adjusting for known covariates, phenogroup B had a 20% increase in long-term mortality compared with phenogroup A (hazard ratio, 1.20 [95% CI, 1.17-1.23]; P<0.0001; phenogroup A mortality, 2.2%; phenogroup B mortality, 6.1%). In univariate analyses, we found phenogroup B had a significantly greater risk of mortality in all cohorts (log-rank P<0.01 in all 5 cohorts). Phenome-wide association study showed phenogroup B had a higher rate of future atrial fibrillation (odds ratio, 2.89; P<0.00001), ventricular tachycardia (odds ratio, 2.00; P<0.00001), ischemic heart disease (odds ratio, 1.44; P<0.00001), and cardiomyopathy (odds ratio, 2.04; P<0.00001). A single-trait genome-wide association study yielded 4 loci. SCN10A, SCN5A, and CAV1 have roles in cardiac conduction and arrhythmia. ARHGAP24 does not have a clear cardiac role and may be a novel target.

CONCLUSIONS: Neural network-derived ECG features can be used to predict all-cause mortality and future cardiovascular diseases. We have identified biologically plausible and novel phenotypic and genotypic associations that describe mechanisms for the increased risk identified.

BACKGROUND: Pacing-induced cardiomyopathy (PICM) is a frequent complication of right ventricular pacing that often requires reoperation for biventricular or conduction system pacing. Better methods for predicting PICM may inform initial pacing strategy and follow-up monitoring.

OBJECTIVE: The purpose of this study was to determine whether the spatial ventricular gradient (SVG), a vectorcardiographic marker of ventricular electrical and mechanical heterogeneity, is associated with the subsequent development of PICM.

METHODS: This was a retrospective study of patients with pacemakers implanted between 2003 and 2012 at the Hospital of the University of Pennsylvania. Baseline demographic, echocardiographic, and electrocardiographic parameters, including SVG magnitude, elevation, and azimuth, were measured from standard 12-lead electrocardiograms. Adjusted Cox proportional hazards modeling was used to assess the associations between the SVG and the risk of PICM over follow-up.

RESULTS: Of the 203 patients with a median age of 74 years (p25-p75 64-79), 110 (54%) male, and median baseline left ventricular ejection fraction 65% (p25-p75 57-70), 44 (22%) developed PICM during follow-up. In unadjusted Cox regression, male sex, native QRS duration in patients without bundle branch block, and both native and paced mean adjusted SVG azimuth predicted future PICM. After multivariable adjustment, higher tertile (tertile 3 vs tertiles 1-2) of the mean adjusted SVG azimuth before (adjusted hazard ratio 1.95; P = .047) and immediately after (adjusted hazard ratio 2.55; P=.003) pacemaker implantation remained significant predictors of PICM.

CONCLUSION: Assessment of the SVG both before and immediately after pacemaker implantation can help identify patients at elevated risk of PICM and may identify a cohort of patients who would be better served with initial biventricular or conduction system pacing.

Degenerative mitral valve disease is common. Up to a quarter of patients with degenerative mitral valve disease may be asymptomatic despite having severe valve regurgitation. Current guideline indications for intervention in asymptomatic patient are centred on left ventricular dimensions and ejection fraction and may include consideration in atrial fibrillation, pulmonary hypertension and those with left atrial dilatation. However, despite intervention according to these recommendations, patients remain at risk of post-operative heart failure and mortality. Newer risk markers have been developed including left ventricular and atrial strain, myocardial fibrosis demonstrated using late gadolinium enhancement, mitral annular disjunction and ventricular arrhythmia burden. Translating newer markers into clinical practice will require integrating and identifying high-risk phenotypes that benefit from early intervention using machine learning techniques and artificial intelligence. Valve repair is the recommended intervention. However, repair rate and durability are dependent on both operator and centre volumes as well as valve characteristics. Recent advancements, including robotic surgery, may enhance repair rates; however, larger datasets are necessary to confirm these improvements. Efforts should focus on establishing high-volume regional centres of excellence for mitral valve repair.

BACKGROUND: In 2019, the US Food and Drug Administration issued a warning that symptomatic relief from claudication using paclitaxel-coated devices might be associated with an increase in mortality over 5 years. We designed a discrete-choice experiment (DCE) to quantify tradeoffs that patients would accept between a decreased risk of clinically driven target-vessel revascularization (CDTVR) and increased mortality risk.

METHODS: Patients with claudication symptoms were recruited from seven medical centers to complete a web-based survey including eight DCE questions that presented pairs of hypothetical device profiles defined by varying risks of CDTVR and overall mortality at 2 and 5 years. Random-parameters logit models were used to estimate relative preference weights, from which the maximum-acceptable increase in 5-year mortality risk was derived.

RESULTS: A total of 272 patients completed the survey. On average, patients would accept a device offering reductions in CDTVR risks from 30% to 10% at 2 years and from 40% to 30% at 5 years if the 5-year mortality risk was less than 12.6% (95% CI: 11.8-13.4%), representing a cut-point of 4.6 percentage points above a baseline risk of 8%. However, approximately 40% chose the device alternative with the lower 5-year mortality risk in seven (20.6%) or eight (18.0%) of the eight DCE questions regardless of the benefit offered.

CONCLUSIONS: Most patients in the study would accept some incremental increase in 5-year mortality risk to reduce the 2-year and 5-year risks of CDTVR by 20 and 10 percentage points, respectively. However, significant patient-level variability in risk tolerance underscores the need for systematic approaches to support benefit-risk decision making.

BACKGROUND: Patients who undergo cardiac surgery for drug use-associated infective endocarditis (DUA-IE) have high rates of readmissions for recurrent endocarditis, substance use disorder (SUD), and septicemia. Our primary objective was to assess whether exposure to an addiction consult team (ACT) was associated with reduced readmissions in this population.

METHODS: This single-center retrospective analysis identified patients who underwent cardiac surgery for DUA-IE between 1/2012-9/2022 using the Society for Thoracic Surgeons database, and compared the cumulative incidence of readmissions at 1, 3, 6, and 12 months among those cared for before and after the implementation of an ACT in 9/2017, accounting for competing risk of mortality and adjusted for measured confounders using inverse probability of treatment weighting.

RESULTS: The 58 patients (35 pre-ACT and 23 post-ACT) were young (36.4 +/- 7.7 years) and predominantly White (53.4%) and male (70.7%). The post-ACT cohort had a significantly lower risk of readmission at 1 month (adjusted risk difference [RD] -23.8% [95% CI -94.4%, -8.3%], P = 0.005) and 3 months (RD -34.1% [-55.1%, -13.1%], P = 0.005), but not at 6 or 12 months. In a sensitivity analysis, the post-ACT cohort also had significantly lower risk of readmissions for SUD complications at 3 months.

DISCUSSION AND CONCLUSION: ACT exposure was associated with a lower risk of short-term readmission among patients with surgically managed DUA-IE, possibly due to a reduction in SUD-related complications. Additional studies are needed to replicate these findings and to identify ways to sustain the potential benefits of ACTs over the longer term.

BACKGROUND: High and rising prescription drug costs for asthma and chronic obstructive pulmonary disease (COPD) contribute to medication nonadherence and poor clinical outcomes. The recently enacted Inflation Reduction Act includes provisions that will cap out-of-pocket prescription drug spending at $2,000 per year and expand low-income subsidies. However, little is known about how these provisions will impact out-of-pocket drug spending for Medicare beneficiaries with asthma and COPD.

OBJECTIVE: To estimate the impact of the Inflation Reduction Act's out-of-pocket spending cap and expansion of low-income subsidies on Medicare beneficiaries with obstructive lung disease.

DESIGN: We calculated the number of Medicare beneficiaries ≥ 65 years with asthma and/or COPD and out-of-pocket prescription drug spending > $2,000/year, and then estimated their median annual out-of-pocket savings under the Inflation Reduction Act's spending cap. We then estimated the number of beneficiaries with incomes > 135% and ≤ 150% of the federal poverty level who would become newly eligible for low-income subsidies under this policy.

PARTICIPANTS: Respondents to the 2016-2019 Medical Expenditure Panel Survey (MEPS).

MAIN MEASURES: Annual out-of-pocket prescription drug spending.

KEY RESULTS: An annual estimated 5.2 million Medicare beneficiaries had asthma and/or COPD. Among them, 360,160 (SE ± 38,021) experienced out-of-pocket drug spending > $2,000/year, with median out-of-pocket costs of $3,003/year (IQR $2,360-$3,941). Therefore, median savings under the Inflation Reduction Act's spending cap would be $1,003/year (IQR $360-$1,941), including $738/year and $1,137/year for beneficiaries with asthma and COPD, respectively. Total annual estimated savings would be $504 million (SE ± $42 M). In addition, 232,155 (SE ± 4,624) beneficiaries would newly qualify for low-income subsidies, which will further reduce prescription drug costs.

CONCLUSIONS: The Inflation Reduction Act will have major implications on out-of-pocket prescription drug spending for Medicare beneficiaries with obstructive lung disease resulting in half-a-billion dollars in total out-of-pocket savings per year, which could ultimately have implications on medication adherence and clinical outcomes.

OBJECTIVE: To assess the association between state policies and sociodemographic characteristics and state mean fair share spending at non-profit hospitals. Fair share spending is a hospital's charity care and community investment less the estimated value of their tax-exempt status.

BACKGROUND: Hospitals with non-profit status in the United States are exempt from paying taxes. In return, they are expected to provide community benefits by subsidizing medical care for those who cannot pay and investing in the health and social needs of their community.

METHODS: We used a multivariable linear regression model to determine the association of state-level sociodemographics and policies with state-level mean fair share spending in 2019. Fair share spending data was obtained from the Lown Institute.

RESULTS: We found no association between the percentage of people living in poverty, in rural areas, or U.S. region and fair share spending. Similarly, there was no association found for state minimum community benefit and reporting requirements. The state percentage of racial/ethnic minorities was associated with higher mean fair share spending [+$1.48 million for every 10% increase (95% CI: 0.01 to 2.96 million)]. Medicaid expansion status was associated with a 6.9-million-dollar decrease (95% CI: -10.4 to -3.3 million).

CONCLUSIONS: State-level community benefit policies have been ineffective at raising community benefit spending to levels comparable to the value of non-profit hospital tax-exempt status.

Social determinants of health (SDoH) significantly impacts health outcomes and disparities. While the Centers for Medicare and Medicaid Services has mandated hospitals to collect standardized SDoH data, existing tools lack key elements. This systematic review identified 78 studies and 20 screening tools addressing various SDoH domains. However, most tools were missing several key domains and lacked standardization. We propose a comprehensive tool meeting essential criteria: validated questions, brevity, actionability, cultural appropriateness, workflow integration, and community linkage. Our tool addresses gaps in available tools and incorporates standardized and validated questions to enable patient-centered screening for diverse social and environmental determinants of health. It uniquely includes detailed race/ethnicity data collection, housing characteristics, physical activity assessment, access to healthy food measures, and environmental exposure evaluation. The tool aims to provide actionable data for immediate interventions while informing broader population health strategies and policy initiatives. By offering a holistic assessment of SDoH across multiple domains, our tool enables standardized data collection, risk stratification, and focused initiatives to address health inequities at both individual and population levels. Further research is needed to develop evidence-based pathways for integrating SDoH data into real-world patient care workflows, improve risk prediction algorithms, address health-related social needs, and reduce disparities.