Spartan Health Sciences University is a private, for-profit medical school located in Vieux Fort, St. Lucia, in the Caribbean. Spartan confers upon its graduates the Doctor of Medicine (MD) degree. Spartan graduates can practice medicine in 46 US states.
Background: Takotsubo syndrome (TTS) is anecdotally believed to be a short-term acute phenomenon without any long-term mortality, however risk stratification at diagnosis is limited for long term adverse outcomes. We leveraged a large, real-world dataset to develop and validate a Machine Learning (ML)-based all-cause mortality prediction model and a derived integer clinical score ( TTS-SCORE ) for bedside use. Hypothesis: We hypothesized that a ML survival model using easily available baseline data at TTS diagnosis would improve mortality prediction and support a simplified bedside risk score - TTS-SCORE. Methods: We identified 19,897 patients (age 63.0 ± 13.5 years, female 85.1%) with index (ICD-10-CM I51.81) TTS diagnosis, excluding prior heart failure (ICD-10-CM I50.x) diagnosis, from the TriNetX federated research network (≥74 US health-care organizations, 2016-2026). Baseline-only features (age, sex, comorbidities, presenting labs and vitals) were used to eliminate outcome leakage. Two survival models were compared: elastic-net penalized Cox regression and Random Survival Forest (RSF). Train and Test Split was (70/30; training n = 13,928; test n = 5,969) respectively. Discrimination was assessed by Harrell's C-index and time-dependent AUC at 6 months to 4-years; calibration by a decile-based observed vs. predicted plot at 24 months; and overall accuracy by the integrated Brier score. The TTS-SCORE integer instrument was derived from the top RSF predictors using clinically rounded cut-points. Missing data was handled by median-imputing train and test set variables. Results: In our test set RSF analysis, 10-year survival probabilities of low, intermediate and high risk groups based on our TTS-SCORE ( Table 1 ) was 94%, 75%, and 51% (p < 0.0001) ( Figure 1 ).The RSF achieved a C-index of 0.804 (penalized Cox: 0.767) and time-dependent AUC of 0.841-0.847 across all follow-up durations, substantially outperforming random chance ( Figure 2 ). Calibration at 24 months was excellent across five predicted-risk strata ( Figure 3 ). The top predictors were age, hemoglobin, creatinine, malignancy, sex, heart rate, and sodium. The derived TTS-SCORE (0-17 points) successfully identified Low (0-4), Intermediate (5-8), and High (≥9) risk categories. Conclusions: Our ML model achieves strong discrimination and calibration for all-cause mortality across up to 4 years of follow-up, representing the longest follow-up reported among studies in the literature. The derived TTS-SCORE provides a practical bedside tool for risk stratification and may help identify patients who warrant close follow-up, early referral, and ongoing observation in the cardiology clinic.
Introduction: A total of 600,000 Left atrial appendage occlusion (LAAO) devices has been implanted worldwide, but randomized trials enroll selected patients in experienced centers only. Real-world outcomes across unselected multicenter institutions remain underexplored. Furthermore, bedside tools to identify patients at high risk of adverse outcomes are limited in the real world. Hypothesis We evaluated real-world clinical outcomes after LAAO and hypothesized that routinely available baseline clinical variables could be used to derive a practical bedside mortality risk score. Methods: We analyzed adults undergoing percutaneous LAAO across 111 healthcare organizations from 2016 to 2026 using TriNetX, a global federated electronic-health-record network. The index date was the first qualifying LAAO procedure (CPT 33340). Incident ischemic stroke, major bleeding, major adverse cardiovascular events (MACE), and all-cause death were estimated using Kaplan-Meier analysis. Mortality prediction models were developed using a 70/30 train-test split. Penalized Cox regression and random survival forest models were compared with CHA 2 DS 2 -VASc using Harrell C-index, time-dependent area under the curve (AUC), calibration, and integrated Brier score. A simplified 0-19-point bedside mortality score was derived from baseline clinical predictors. Results: A total of 29,554 patients were identified (mean age 75.3±7.5 years, 39.9% women, mean CHA 2 DS 2 -VASc 4.9, mean HAS-BLED 2.6, heart failure 51.2%, chronic kidney disease 35.3%, and prior major bleeding 42.1%). Kaplan-Meier cumulative incidence at 1 and 2 years was 6.9% and 13.5% for all-cause death, 2.0% and 3.7% for incident ischemic stroke, 3.9% and 5.7% for incident major bleeding, and 12.3% and 21.2% for MACE (Figure 1). Random survival forest showed the strongest mortality discrimination, with C-index 0.729 and AUCs of 0.754, 0.752, and 0.753 at 12, 24, and 36 months, respectively ( Figure 2a ). Penalized Cox also outperformed CHA 2 DS 2 -VASc (C-index 0.713 vs 0.602). Model calibration was acceptable, with integrated Brier score 0.0916 ( Figure 2b ). A bedside score incorporating age ≥75 years, heart failure, chronic kidney disease, dementia, chronic obstructive pulmonary disease, diabetes, prior major bleeding, vascular disease, cancer, and anemia stratified 12-month mortality from 2.8% in the low-risk group to 19.1% in the very-high-risk group; corresponding 24-month mortality increased from 6.3% to 32.4% ( Figure 3, Table1 ). Conclusion: In a large multicenter real-world LAAO cohort, incident ischemic stroke and major bleeding rates were low despite substantial baseline comorbidity. Machine-learning survival models improved mortality prediction compared with CHA 2 DS 2 -VASc, and a simple bedside score provided clinically practical mortality risk stratification.
Abstract Rationale Patients with rheumatoid arthritis-associated interstitial lung disease (RA-ILD) experience significant cardiopulmonary morbidity. Emerging evidence suggests sodium-glucose cotransporter-2 inhibitors (SGLT2i) may confer cardiovascular and survival benefits through pleiotropic mechanisms, yet their role in RA-ILD remains unclear. Methods We performed a retrospective, multicenter cohort study using the TriNetX U.S. Collaborative Network, which aggregates de-identified electronic health records from 109 healthcare organizations. Adults with RA-ILD and type 2 diabetes were categorized by SGLT2i exposure (dapagliflozin or empagliflozin) versus no exposure. Propensity-score matching (1:1) was applied to balance demographics, comorbidities, and cardiometabolic factors, yielding 452 patients in each group. Outcomes were assessed ≥180 days after index diagnosis and included all-cause mortality and major adverse cardiovascular events (MACE: myocardial infarction, stroke, cardiac arrest, or death). Measures of association and Kaplan-Meier analyses were conducted within the TriNetX analytic environment. Results After matching, cohorts were well balanced for age (70.4 ± 9.7 vs 70.6 ± 12.3 years), sex (61% female), and comorbid burden. SGLT2i use was associated with a numerically lower risk of all-cause mortality compared with non-users (11.7% vs 13.1%; odds ratio 0.89, 95% CI 0.60-1.32; p = 0.55). Kaplan-Meier analysis demonstrated improved survival among SGLT2i users (five-year survival 72.2% vs 19.4%; log-rank p = 0.032; hazard ratio 0.67, 95% CI 0.46-0.97). Rates of MACE were similar between groups (22.8% vs 18.8%; OR 1.27, 95% CI 0.92-1.76; p = 0.14), with a neutral effect on event-free survival (HR 0.95, 95% CI 0.71-1.26). No significant differences were observed in hospitalization frequency or event recurrence. Conclusions In this national real-world cohort of patients with RA-ILD and diabetes, SGLT2 inhibitor therapy was associated with improved long-term survival but did not significantly alter cardiovascular event rates. These findings support a potential mortality benefit independent of cardiovascular endpoints and warrant prospective validation to clarify mechanistic pathways in RA-ILD. This abstract is funded by: None
Multidrug-resistant (MDR) infections in ICUs are a growing global health challenge associated with increased morbidity, mortality, prolonged hospitalization, and rising healthcare costs. Critically ill patients are particularly vulnerable because of immune dysfunction, invasive procedures, prolonged ICU stays, and frequent exposure to broad-spectrum antibiotics, all of which increase susceptibility to MDR infections and antimicrobial selection pressure. Common MDR pathogens in ICU settings include Acinetobacter baumannii, Klebsiella pneumoniae, Pseudomonas aeruginosa, methicillin-resistant Staphylococcus aureus (MRSA), vancomycin-resistant Enterococcus (VRE), and carbapenem-resistant Enterobacterales. Major resistance mechanisms include β-lactamase and carbapenemase production, efflux pump overexpression, reduced membrane permeability, target modification, biofilm formation, and horizontal gene transfer. Effective management of MDR infections in ICUs requires an integrated, evidence-based approach combining early recognition, microbiological diagnosis, optimized antimicrobial therapy, and strict infection-prevention strategies. Pharmacological management relies on appropriate empiric therapy, antimicrobial stewardship, therapeutic drug monitoring, and pharmacokinetic/pharmacodynamic optimization to improve treatment efficacy while minimizing toxicity and resistance development. Newer antimicrobial agents, including β-lactam/β-lactamase inhibitor combinations and cefiderocol, have expanded treatment options for resistant resistant Gram-negative infections. Non-pharmacological interventions such as hand hygiene, environmental disinfection, surveillance cultures where indicated, contact precautions, and multidisciplinary infection-control programs remain essential in reducing transmission. Emerging innovations, including artificial intelligence-guided antimicrobial selection, rapid diagnostic technologies, microbiome-based therapies, and bacteriophage therapy, show promise for future management. A multidisciplinary strategy integrating prevention, stewardship, and ongoing research is essential to reduce the MDR burden in ICUs and preserve antimicrobial effectiveness.
Introduction: Takotsubo syndrome (TTS) is clinically heterogeneous, however most outcomes studies treat it as a single disease. Whether baseline clinical data identify reproducible TTS phenotypes with distinct prognosis remains unclear. Research Question: We hypothesized that unsupervised modeling of pre-index demographics and comorbidities would identify stable TTS phenotypes associated with differential survival and cardiovascular events. Methods: We studied adults with first recorded TTS (ICD-10-CM I51.81) from 2016 using the TriNetX database, a federated multicenter electronic-health-record network. Baseline features included age, sex, race, and 15 pre-index comorbidities. Latent class analysis was fit for K=2-6; K was selected using BIC trajectory, interpretability, and class size. Reproducibility was assessed by 20 bootstrap 80% subsamples using adjusted Rand index (ARI), with k-prototypes as a mixed-data confirmatory method. Outcomes included all-cause mortality, recurrent TTS diagnoses more than 90 days, incident atrial fibrillation (AF), and incident stroke. Survival was compared using Kaplan-Meier and log-rank testing. P <0.05 was considered statistically significant. Results: Among 18,759 patients admitted with TTS, four phenotypes were identified: P0 low-comorbidity (n=8,060; mean age 61.7 years), P1 neuropsychiatric (n=2,939; age 51.3 years; depression/anxiety 57%, tobacco 57%), P2 cardiometabolic (n=5,071; age 71.7 years; hypertension 71%, hyperlipidemia 67%), and P3 multimorbid/frail (n=2,689; age 68.1 years; hypertension 96%, diabetes 49%, coronary disease 53%, chronic kidney disease 29%) ( Figure 1 ). The 4-class solution was stable (bootstrap mean ARI 0.96); k-prototypes concordance was moderate (ARI 0.305), with visual separation of baseline feature structure on UMAP ( Figure 2 ). Survival differed markedly across phenotypes (p<0.001), with P3 showing the poorest long-term survival ( Figure 3 ). Recurrent TTS diagnoses were highest in P2 (44.1%; P0 31.2%, P1 34.6%, P3 33.4%; p<0.001). Incident AF was also highest in P2 (9.3%; p<0.001), while stroke varied modestly but significantly (P0 4.9%, P1 3.6%, P2 5.1%, P3 5.5%; p<0.001). Conclusion: Latent class analysis identified four stable, clinically distinguishable TTS phenotypes stratified by survival and cardiovascular events. These findings support phenotype-based risk assessment to identify high risk patients who might be prone to adverse cardiovascular events might warrant close follow up and investigation. Prospective studies are warranted with more robust hemodynamic, and cardiac imaging data to better risk stratify these patients.