Faculty Bibliography

BACKGROUND/OBJECTIVE/OBJECTIVE:Studies show an association between elevated blood pressure, obesity, and insulin resistance with a higher risk of developing diabetes. As sodium is closely linked to elevated blood pressure and hypertension, and potassium is a counterbalancing nutrient to sodium, this study examines the association between intake of sodium, potassium, and sodium: potassium (Na:K) ratio and the incidence of diabetes. DESIGN SETTING AND PARTICIPANTS/METHODS:Retrospective data analysis of dietary intake measured by a validated food frequency questionnaire in a prospective cohort of veterans participating in the Million Veteran Program (MVP) between 2011 and 2020, who were free of diabetes at baseline. MAIN OUTCOME MEASUREMENT/METHODS:The main outcome is clinically diagnosed diabetes defined by phenotyping algorithms applied to electronic health records. RESULTS:In this study of 198,049 veterans (mean age: 63.8 ± 13.1 years, 89 % male), 7260 were diagnosed with diabetes over a mean follow-up of 4.3 years. The mean sodium intake was 1218 mg/day. A higher sodium intake was associated with an 11 % higher rate of developing diabetes (hazard ratio, HR) comparing extreme quintiles: 1.11, 95 % CI: 1.03-1.20). The average daily potassium intake was 2589 mg and the highest quintile of potassium intake was associated with a 13 % lower rate of diabetes (HR: 0.87, 95 % CI: 0.81, 0.94) compared to the lowest quintile of potassium. Highest quintile of Na:K ratio was associated with a 21 % higher rate for diabetes (HR: 1.21, 95% CI: 1.12, 1.30). The pattern of associations between Na:K ratio and diabetes closely followed the pattern of dietary sodium intake and diabetes associations. CONCLUSION/CONCLUSIONS:A higher sodium intake and a higher Na:K ratio were associated with a higher risk of diabetes in this large cohort of veterans. These findings may be applied in future work to identify personalized lifestyle and dietary supports to prevent and treat T2DM.

BACKGROUND:Limited contemporary evidence exists on risk prediction by stress imaging and exercise electrocardiography (ECG) among patients with chronic coronary syndromes (CCS). Objectives From the ISCHEMIA (International Study of Comparative Health Effectiveness with Medical and Invasive Approaches) study, prognosis was examined by core laboratory-defined stress imaging and exercise ECG findings in CCS patients. METHODS:A total of 5,179 patients (qualifying by stress nuclear imaging [n = 2,567], echocardiography [n = 1,085], cardiac magnetic resonance [CMR] [n = 257], and ECG [n = 1,270]) were randomized. Cox models assessed associations between trial endpoints and the number of scarred and ischemic segments, rest/stress left ventricular ejection fraction (LVEF), and ST-segment depression. HRs and 95% CIs were calculated per millimeter, segment, or 5% of LVEF. We examined prognostic models for the following trial endpoints: 1) the trial's primary endpoint of cardiovascular (CV) death, myocardial infarction (MI), resuscitated cardiac arrest, or hospitalization for unstable angina or heart failure; 2) CV death; 3) spontaneous MI; 4) procedural MI; and 5) type 2 MI. RESULTS:The number of scarred segments (HR: 1.07 [95% CI: 1.02-1.13]; P = 0.0209), rest LVEF (HR: 0.88 [95% CI: 0.83-0.93]; P < 0.001), and stress LVEF (HR: 0.87 [95% CI: 0.83-0.91]; P < 0.001) predicted the trial's primary endpoint of CV death, MI, resuscitated cardiac arrest, or hospitalization for unstable angina or heart failure. The extent of scar and rest/stress LVEF on echocardiography and nuclear imaging predicted several trial endpoints. The number of ischemic segments predicted spontaneous (HR: 1.08 [95% CI: 1.03-1.14]; P = 0.0104) and procedural MI (HR: 1.14 [95% CI: 1.03-1.25]; P = 0.0015) but was of borderline significance for the trial's primary endpoint (P = 0.0746). Ischemia extent by CMR predicted the trial's primary endpoint (P = 0.0068) and spontaneous MI (P = 0.0042). CONCLUSIONS:ISCHEMIA trial findings from 320 worldwide centers revealed that stress imaging and exercise ECG measures exhibited a variable association with key trial endpoints delineating risk patterns for ischemia and infarction. Stress CMR ischemia predicted several trial endpoints, supporting an expanded role in the evaluation of patients with CCS (ISCHEMIA [International Study of Comparative Health Effectiveness With Medical and Invasive Approaches]; NCT01471522).

BACKGROUND/UNASSIGNED:The relationship between the extent and severity of stress-induced ischemia and the extent and severity of anatomic coronary artery disease (CAD) in patients with obstructive CAD is multifactorial and includes the intensity of stress achieved, type of testing used, presence and extent of prior infarction, collateral blood flow, plaque characteristics, microvascular disease, coronary vasomotor tone, and genetic factors. Among chronic coronary disease participants with site-determined moderate or severe ischemia, we investigated associations between ischemia severity on stress testing and the extent of CAD on coronary computed tomography angiography. METHODS/UNASSIGNED:Clinically indicated stress testing included nuclear imaging, echocardiography, cardiac magnetic resonance imaging, or nonimaging exercise tolerance test. Among those with preserved renal function who underwent coronary computed tomography angiography, we examined relationships between ischemia and CAD by coronary computed tomography angiography, overall, and by stress test modality, regardless of subsequent randomization. Core laboratories categorized ischemia as severe, moderate, mild, or none, while the extent and severity of anatomic CAD were categorized based on the modified Duke prognostic index. RESULTS/UNASSIGNED:Among 3601 participants with interpretable stress tests and coronary computed tomography angiography, ischemia severity was weakly associated with CAD extent/severity (r=0.27), with modest variability in strength of association by modality: nuclear (n=1532; r=0.40), echocardiography (n=827; r=0.15), cardiac magnetic resonance imaging (n=108; r=0.31), and exercise tolerance test (n=1134; r=0.18). The extent of infarction on nuclear imaging and echocardiography was weakly associated with CAD extent/severity. CONCLUSIONS/UNASSIGNED:Overall, ischemia severity on stress testing showed weak to moderate associations with the anatomic extent of CAD in this cohort with moderate or severe ischemia on local interpretation and controlled symptoms. REGISTRATION/UNASSIGNED:URL: https://www.clinicaltrials.gov; Unique identifier: NCT01471522.

BACKGROUND:Previous studies evaluated the ability of large language models (LLMs) in medical disciplines; however, few have focused on image analysis, and none specifically on cardiovascular imaging or nuclear cardiology. This study assesses four LLMs-GPT-4, GPT-4 Turbo, GPT-4omni (GPT-4o) (Open AI), and Gemini (Google Inc.)-in responding to questions from the 2023 American Society of Nuclear Cardiology Board Preparation Exam, reflecting the scope of the Certification Board of Nuclear Cardiology (CBNC) examination. METHODS:We used 168 questions: 141 text-only and 27 image-based, categorized into four sections mirroring the CBNC exam. Each LLM was presented with the same standardized prompt and applied to each section 30 times to account for stochasticity. Performance over six weeks was assessed for all models except GPT-4o. McNemar's test compared correct response proportions. RESULTS:GPT-4, Gemini, GPT-4 Turbo, and GPT-4o correctly answered median percentages of 56.8% (95% confidence interval 55.4% - 58.0%), 40.5% (39.9% - 42.9%), 60.7% (59.5% - 61.3%), and 63.1% (62.5%-64.3%) of questions, respectively. GPT-4o significantly outperformed other models (P = .007 vs GPT-4 Turbo, P < .001 vs GPT-4 and Gemini). GPT-4o excelled on text-only questions compared to GPT-4, Gemini, and GPT-4 Turbo (P < .001, P < .001, and P = .001), while Gemini performed worse on image-based questions (P < .001 for all). CONCLUSION/CONCLUSIONS:GPT-4o demonstrated superior performance among the four LLMs, achieving scores likely within or just outside the range required to pass a test akin to the CBNC examination. Although improvements in medical image interpretation are needed, GPT-4o shows potential to support physicians in answering text-based clinical questions.

Electronic health record (EHR)-embedded tools are known to improve prescribing of guideline-directed medical therapy (GDMT) for patients with heart failure. However, physicians may perceive EHR tools to be unhelpful, and may be therefore hesitant to implement these in their practice. We surveyed cardiologists about two effective EHR-tools to improve heart failure care, and they perceived the EHR tools to be easy to use, helpful, and improve the overall management of their patients with heart failure.

BACKGROUND/UNASSIGNED:Previous studies evaluated the ability of large language models (LLMs) in medical disciplines; however, few have focused on image analysis, and none specifically on cardiovascular imaging or nuclear cardiology. OBJECTIVES/UNASSIGNED:This study assesses four LLMs - GPT-4, GPT-4 Turbo, GPT-4omni (GPT-4o) (Open AI), and Gemini (Google Inc.) - in responding to questions from the 2023 American Society of Nuclear Cardiology Board Preparation Exam, reflecting the scope of the Certification Board of Nuclear Cardiology (CBNC) examination. METHODS/UNASSIGNED:We used 168 questions: 141 text-only and 27 image-based, categorized into four sections mirroring the CBNC exam. Each LLM was presented with the same standardized prompt and applied to each section 30 times to account for stochasticity. Performance over six weeks was assessed for all models except GPT-4o. McNemar's test compared correct response proportions. RESULTS/UNASSIGNED:GPT-4, Gemini, GPT4-Turbo, and GPT-4o correctly answered median percentiles of 56.8% (95% confidence interval 55.4% - 58.0%), 40.5% (39.9% - 42.9%), 60.7% (59.9% - 61.3%) and 63.1% (62.5 - 64.3%) of questions, respectively. GPT4o significantly outperformed other models (p=0.007 vs. GPT-4Turbo, p<0.001 vs. GPT-4 and Gemini). GPT-4o excelled on text-only questions compared to GPT-4, Gemini, and GPT-4 Turbo (p<0.001, p<0.001, and p=0.001), while Gemini performed worse on image-based questions (p<0.001 for all). CONCLUSION/UNASSIGNED:GPT-4o demonstrated superior performance among the four LLMs, achieving scores likely within or just outside the range required to pass a test akin to the CBNC examination. Although improvements in medical image interpretation are needed, GPT-4o shows potential to support physicians in answering text-based clinical questions.