"Big Data" Analyses Underlie Clinical Discoveries at the Aortic Institute.

This issue of the Yale Journal of Biology and Medicine (YJBM) focuses on Big Data and precision analytics in medical research. At the Aortic Institute at Yale New Haven Hospital, the vast majority of our investigations have emanated from our large, prospective clinical database of patients with thoracic aortic aneurysm (TAA), supplemented by ultra-large genetic sequencing files. Among the fundamental clinical and scientific discoveries enabled by application of advanced statistical and artificial intelligence techniques on these clinical and genetic databases are the following: From analysis of Traditional "Big Data" (Large data sets). 1. Ascending aortic aneurysms should be resected at 5 cm to prevent dissection and rupture. 2. Indexing aortic size to height improves aortic risk prognostication. 3. Aortic root dilatation is more malignant than mid-ascending aortic dilatation. 4. Ascending aortic aneurysm patients with bicuspid aortic valves do not carry the poorer prognosis previously postulated. 5. The descending and thoracoabdominal aorta are capable of rupture without dissection. 6. Female patients with TAA do more poorly than male patients. 7. Ascending aortic length is even better than aortic diameter at predicting dissection. 8. A "silver lining" of TAA disease is the profound, lifelong protection from atherosclerosis. From Modern "Big Data" Machine Learning/Artificial Intelligence analysis: 1. Machine learning models for TAA: outperforming traditional anatomic criteria. 2. Genetic testing for TAA and dissection and discovery of novel causative genes. 3. Phenotypic genetic characterization by Artificial Intelligence. 4. Panel of RNAs "detects" TAA. Such findings, based on (a) long-standing application of advanced conventional statistical analysis to large clinical data sets, and (b) recent application of advanced machine learning/artificial intelligence to large genetic data sets at the Yale Aortic Institute have advanced the diagnosis and medical and surgical treatment of TAA.

Bicuspid aortopathy does not require earlier surgical intervention.

OBJECTIVES: Guidelines for surgical correction of patients with ascending thoracic aortic aneurysm (ATAA) with a bicuspid aortic valve (BAV) have oscillated over the years. In this study, we outline the natural history of the ascending aorta in patients with BAV and trileaflet aortic valve (TAV) ATAA followed over time, to ascertain if their behavior differs and to determine if a different threshold for intervention is required. METHODS: Aortic diameters and long-term complications (ie, adverse aortic events) of 2428 patients (554 BAV and 1874 TAV) with ATAA before operative repair were reviewed. Growth rates, yearly complication rates, event-free survival, and risk of complications as a function of aortic size were calculated. Long-term follow-up and precise cause of death granularity was achieved via a comprehensive 6-pronged approach. RESULTS: Aortic growth rate in patients with BAV vs TAV ATAA was 0.20 and 0.17 cm/year, respectively (P = .009), with the rate increasing with increasing aortic size. Yearly adverse aortic events rates increased with ATAA size and were lower for patients with BAV. The relative risk of adverse aortic events exhibited an exponential increase with aortic diameter. Patients with BAV had a lower all-cause and ascending aorta-specific adverse aortic events hazard. Age-adjusted 10-year event-free survival was significantly better for patients with BAV, and BAV emerged as a protective factor against type A dissection, rupture, and ascending aortic death. CONCLUSIONS: The threshold for surgical repair of ascending aneurysm with BAV should not differ from that of TAV. Prophylactic surgery should be considered at 5.0 cm for patients with TAV (and BAV) at expert centers.

Drp1/p53 interaction mediates p53 mitochondrial localization and dysfunction in septic cardiomyopathy.

BACKGROUND: Previous studies have implicated p53-dependent mitochondrial dysfunction in sepsis induced end organ injury, including sepsis-induced myocardial dysfunction (SIMD). However, the mechanisms behind p53 localization to the mitochondria have not been well established. Dynamin-related protein 1 (Drp1), a mediator of mitochondrial fission, may play a role in p53 mitochondrial localization. Here we examined the role of Drp1/p53 interaction in SIMD using in vitro and murine models of sepsis. METHODS: H9c2 cardiomyoblasts and BALB/c mice were exposed to lipopolysaccharide (LPS) to model sepsis phenotype. Pharmacologic inhibitors of Drp1 activation (ψDrp1) and of p53 mitochondrial binding (pifithrin µ, PFTµ) were utilized to assess interaction between Drp1 and p53, and the subsequent downstream impact on mitochondrial morphology and function, cardiomyocyte function, and sepsis phenotype. RESULTS: Both in vitro and murine models demonstrated an increase in physical Drp1/p53 interaction following LPS treatment, which was associated with increased p53 mitochondrial localization, and mitochondrial dysfunction. This Drp1/p53 interaction was inhibited by ΨDrp1, suggesting that this interaction is dependent on Drp1 activation. Treatment of H9c2 cells with either ΨDrp1 or PFTµ inhibited the LPS mediated localization of Drp1/p53 to the mitochondria, decreased oxidative stress, improved cellular respiration and ATP production. Similarly, treatment of BALB/c mice with either ΨDrp1 or PFTµ decreased LPS-mediated mitochondrial localization of p53, mitochondrial ROS in cardiac tissue, and subsequently improved cardiomyocyte contractile function and survival. CONCLUSION: Drp1/p53 interaction and mitochondrial localization is a key prodrome to mitochondrial damage in SIMD and inhibiting this interaction may serve as a therapeutic target.

A machine learning approach for predicting complications in descending and thoracoabdominal aortic aneurysms.

OBJECTIVE: To use machine learning to predict rupture, dissection, and all-cause mortality for patients with descending and thoracoabdominal aortic aneurysms in an effort to improve on diameter-based surgical intervention criteria. METHODS: Retrospective data from 1083 patients with descending aortic diameters 3.0 cm or greater were collected, with a mean follow-up time of 3.52 years and an average descending diameter of 4.13 cm. Six machine learning classifiers were trained using 44 variables to predict the occurrence of dissection, rupture, or all-cause mortality within 1, 2, or 5 years of initial patient encounter for a total of 54 (6 × 3 × 3) separate classifiers. Classifier performance was measured using area under the receiver operator curve. RESULTS: Machine learning models achieved area under the receiver operator curves of 0.842 to 0.872 when predicting type B dissection, 0.847 to 0.856 when predicting type B dissection or rupture, and 0.820 to 0.845 when predicting type B dissection, rupture, or all-cause mortality. All models consistently outperformed descending aortic diameter across all end points (area under the receiver operator curve = 0.713-0.733). Feature importance inspection showed that other features beyond aortic diameter, such as a history of myocardial infarction, hypertension, and patient sex, play an important role in improving risk prediction. CONCLUSIONS: This study provides surgeons with a more accurate, machine learning-based, risk-stratification metric to predict complications for patients with descending aortic aneurysms.

A Selective Inhibitor of Cardiac Troponin I Phosphorylation by Delta Protein Kinase C (δPKC) as a Treatment for Ischemia-Reperfusion Injury.

Myocardial infarction is the leading cause of cardiovascular mortality, with myocardial injury occurring during ischemia and subsequent reperfusion (IR). We previously showed that the inhibition of protein kinase C delta (δPKC) with a pan-inhibitor (δV1-1) mitigates myocardial injury and improves mitochondrial function in animal models of IR, and in humans with acute myocardial infarction, when treated at the time of opening of the occluded blood vessel, at reperfusion. Cardiac troponin I (cTnI), a key sarcomeric protein in cardiomyocyte contraction, is phosphorylated by δPKC during reperfusion. Here, we describe a rationally-designed, selective, high-affinity, eight amino acid peptide that inhibits cTnI's interaction with, and phosphorylation by, δPKC (ψTnI), and prevents tissue injury in a Langendorff model of myocardial infarction, ex vivo. Unexpectedly, we also found that this treatment attenuates IR-induced mitochondrial dysfunction. These data suggest that δPKC phosphorylation of cTnI is critical in IR injury, and that a cTnI/δPKC interaction inhibitor should be considered as a therapeutic target to reduce cardiac injury after myocardial infarction.

Drp1/Fis1-Dependent Pathologic Fission and Associated Damaged Extracellular Mitochondria Contribute to Macrophage Dysfunction in Endotoxin Tolerance.

OBJECTIVES: Recent publications have shown that mitochondrial dynamics can govern the quality and quantity of extracellular mitochondria subsequently impacting immune phenotypes. This study aims to determine if pathologic mitochondrial fission mediated by Drp1/Fis1 interaction impacts extracellular mitochondrial content and macrophage function in sepsis-induced immunoparalysis. DESIGN: Laboratory investigation. SETTING: University laboratory. SUBJECTS: C57BL/6 and BALB/C mice. INTERVENTIONS: Using in vitro and murine models of endotoxin tolerance (ET), we evaluated changes in Drp1/Fis1-dependent pathologic fission and simultaneously measured the quantity and quality of extracellular mitochondria. Next, by priming mouse macrophages with isolated healthy mitochondria (MC) and damaged mitochondria, we determined if damaged extracellular mitochondria are capable of inducing tolerance to subsequent endotoxin challenge. Finally, we determined if inhibition of Drp1/Fis1-mediated pathologic fission abrogates release of damaged extracellular mitochondria and improves macrophage response to subsequent endotoxin challenge. MEASUREMENTS AND MAIN RESULTS: When compared with naïve macrophages (NMs), endotoxin-tolerant macrophages (ETM) demonstrated Drp1/Fis1-dependent mitochondrial dysfunction and higher levels of damaged extracellular mitochondria (Mitotracker-Green + events/50 µL: ETM = 2.42 × 106 ± 4,391 vs NM = 5.69 × 105 ± 2,478; p < 0.001). Exposure of NMs to damaged extracellular mitochondria (MH) induced cross-tolerance to subsequent endotoxin challenge, whereas MC had minimal effect (tumor necrosis factor [TNF]-α [pg/mL]: NM = 668 ± 3, NM + MH = 221 ± 15, and NM + Mc = 881 ± 15; p < 0.0001). Inhibiting Drp1/Fis1-dependent mitochondrial fission using heptapeptide (P110), a selective inhibitor of Drp1/Fis1 interaction, improved extracellular mitochondrial function (extracellular mitochondrial membrane potential, JC-1 [R/G] ETM = 7 ± 0.5 vs ETM + P110 = 19 ± 2.0; p < 0.001) and subsequently improved immune response in ETMs (TNF-α [pg/mL]; ETM = 149 ± 1 vs ETM + P110 = 1,150 ± 4; p < 0.0001). Similarly, P110-treated endotoxin tolerant mice had lower amounts of damaged extracellular mitochondria in plasma (represented by higher extracellular mitochondrial membrane potential, TMRM/MT-G: endotoxin tolerant [ET] = 0.04 ± 0.02 vs ET + P110 = 0.21 ± 0.02; p = 0.03) and improved immune response to subsequent endotoxin treatment as well as cecal ligation and puncture. CONCLUSIONS: Inhibition of Drp1/Fis1-dependent mitochondrial fragmentation improved macrophage function and immune response in both in vitro and in vivo models of ET. This benefit is mediated, at least in part, by decreasing the release of damaged extracellular mitochondria, which contributes to endotoxin cross-tolerance. Altogether, these data suggest that alterations in mitochondrial dynamics may play an important role in sepsis-induced immunoparalysis.

Machine Learning Applied to Electronic Health Records: Identification of Chemotherapy Patients at High Risk for Preventable Emergency Department Visits and Hospital Admissions.

PURPOSE: Acute care use (ACU) is a major driver of oncologic costs and is penalized by a Centers for Medicare & Medicaid Services quality measure, OP-35. Targeted interventions reduce preventable ACU; however, identifying which patients might benefit remains challenging. Prior predictive models have made use of a limited subset of the data in the electronic health record (EHR). We aimed to predict risk of preventable ACU after starting chemotherapy using machine learning (ML) algorithms trained on comprehensive EHR data. METHODS: Chemotherapy patients treated at an academic institution and affiliated community care sites between January 2013 and July 2019 who met inclusion criteria for OP-35 were identified. Preventable ACU was defined using OP-35 criteria. Structured EHR data generated before chemotherapy treatment were obtained. ML models were trained to predict risk for ACU after starting chemotherapy using 80% of the cohort. The remaining 20% were used to test model performance by the area under the receiver operator curve. RESULTS: Eight thousand four hundred thirty-nine patients were included, of whom 35% had preventable ACU within 180 days of starting chemotherapy. Our primary model classified patients at risk for preventable ACU with an area under the receiver operator curve of 0.783 (95% CI, 0.761 to 0.806). Performance was better for identifying admissions than emergency department visits. Key variables included prior hospitalizations, cancer stage, race, laboratory values, and a diagnosis of depression. Analyses showed limited benefit from including patient-reported outcome data and indicated inequities in outcomes and risk modeling for Black and Medicaid patients. CONCLUSION: Dense EHR data can identify patients at risk for ACU using ML with promising accuracy. These models have potential to improve cancer care outcomes, patient experience, and costs by allowing for targeted, preventative interventions.

Developing machine learning models to personalize care levels among emergency room patients for hospital admission.

OBJECTIVE: To develop prediction models for intensive care unit (ICU) vs non-ICU level-of-care need within 24 hours of inpatient admission for emergency department (ED) patients using electronic health record data. MATERIALS AND METHODS: Using records of 41 654 ED visits to a tertiary academic center from 2015 to 2019, we tested 4 algorithms-feed-forward neural networks, regularized regression, random forests, and gradient-boosted trees-to predict ICU vs non-ICU level-of-care within 24 hours and at the 24th hour following admission. Simple-feature models included patient demographics, Emergency Severity Index (ESI), and vital sign summary. Complex-feature models added all vital signs, lab results, and counts of diagnosis, imaging, procedures, medications, and lab orders. RESULTS: The best-performing model, a gradient-boosted tree using a full feature set, achieved an AUROC of 0.88 (95%CI: 0.87-0.89) and AUPRC of 0.65 (95%CI: 0.63-0.68) for predicting ICU care need within 24 hours of admission. The logistic regression model using ESI achieved an AUROC of 0.67 (95%CI: 0.65-0.70) and AUPRC of 0.37 (95%CI: 0.35-0.40). Using a discrimination threshold, such as 0.6, the positive predictive value, negative predictive value, sensitivity, and specificity were 85%, 89%, 30%, and 99%, respectively. Vital signs were the most important predictors. DISCUSSION AND CONCLUSIONS: Undertriaging admitted ED patients who subsequently require ICU care is common and associated with poorer outcomes. Machine learning models using readily available electronic health record data predict subsequent need for ICU admission with good discrimination, substantially better than the benchmarking ESI system. The results could be used in a multitiered clinical decision-support system to improve ED triage.

Machine learning: principles and applications for thoracic surgery.

OBJECTIVES: Machine learning (ML) has experienced a revolutionary decade with advances across many disciplines. We seek to understand how recent advances in ML are going to specifically influence the practice of surgery in the future with a particular focus on thoracic surgery. METHODS: Review of relevant literature in both technical and clinical domains. RESULTS: ML is a revolutionary technology that promises to change the way that surgery is practiced in the near future. Spurred by an advance in computing power and the volume of data produced in healthcare, ML has shown remarkable ability to master tasks that had once been reserved for physicians. Supervised learning, unsupervised learning and reinforcement learning are all important techniques that can be leveraged to improve care. Five key applications of ML to cardiac surgery include diagnostics, surgical skill assessment, postoperative prognostication, augmenting intraoperative performance and accelerating translational research. Some key limitations of ML include lack of interpretability, low quality and volumes of relevant clinical data, ethical limitations and difficulties with clinical implementation. CONCLUSIONS: In the future, the practice of cardiac surgery will be greatly augmented by ML technologies, ultimately leading to improved surgical performance and better patient outcomes.

Use Characteristics and Triage Acuity of a Digital Symptom Checker in a Large Integrated Health System: Population-Based Descriptive Study.

BACKGROUND: Pressure on the US health care system has been increasing due to a combination of aging populations, rising health care expenditures, and most recently, the COVID-19 pandemic. Responses to this pressure are hindered in part by reliance on a limited supply of highly trained health care professionals, creating a need for scalable technological solutions. Digital symptom checkers are artificial intelligence-supported software tools that use a conversational "chatbot" format to support rapid diagnosis and consistent triage. The COVID-19 pandemic has brought new attention to these tools due to the need to avoid face-to-face contact and preserve urgent care capacity. However, evidence-based deployment of these chatbots requires an understanding of user demographics and associated triage recommendations generated by a large general population. OBJECTIVE: In this study, we evaluate the user demographics and levels of triage acuity provided by a symptom checker chatbot deployed in partnership with a large integrated health system in the United States. METHODS: This population-based descriptive study included all web-based symptom assessments completed on the website and patient portal of the Sutter Health system (24 hospitals in Northern California) from April 24, 2019, to February 1, 2020. User demographics were compared to relevant US Census population data. RESULTS: A total of 26,646 symptom assessments were completed during the study period. Most assessments (17,816/26,646, 66.9%) were completed by female users. The mean user age was 34.3 years (SD 14.4 years), compared to a median age of 37.3 years of the general population. The most common initial symptom was abdominal pain (2060/26,646, 7.7%). A substantial number of assessments (12,357/26,646, 46.4%) were completed outside of typical physician office hours. Most users were advised to seek medical care on the same day (7299/26,646, 27.4%) or within 2-3 days (6301/26,646, 23.6%). Over a quarter of the assessments indicated a high degree of urgency (7723/26,646, 29.0%). CONCLUSIONS: Users of the symptom checker chatbot were broadly representative of our patient population, although they skewed toward younger and female users. The triage recommendations were comparable to those of nurse-staffed telephone triage lines. Although the emergence of COVID-19 has increased the interest in remote medical assessment tools, it is important to take an evidence-based approach to their deployment.

The Genetics of Thoracic Aortic Aneurysms and Dissection: A Clinical Perspective.

Thoracic aortic aneurysm and dissection (TAAD) affects many patients globally and has high mortality rates if undetected. Once thought to be solely a degenerative disease that afflicted the aorta due to high pressure and biomechanical stress, extensive investigation of the heritability and natural history of TAAD has shown a clear genetic basis for the disease. Here, we review both the cellular mechanisms and clinical manifestations of syndromic and non-syndromic TAAD. We particularly focus on genes that have been linked to dissection at diameters <5.0 cm, the current lower bound for surgical intervention. Genetic screening tests to identify patients with TAAD associated mutations that place them at high risk for dissection are also discussed.