Deng, J

Abstract Rationale Propofol-based sedation in the intensive care unit (ICU) is associated with clinically significant hemodynamic instability, including hypotension. Previous studies have identified hypotension as a key predictor of adverse outcomes in critically ill patients. However, no validated tools currently exist to systematically predict this risk. Therefore, this study aimed to develop and externally validate machine-learning (ML) models for predicting the risk of hypotension in ICU patients receiving propofol sedation. Methods We developed ML models using retrospective data from adult, mechanically ventilated ICU patients who received propofol infusion for at least 24 hours across four Mayo Clinic sites between 2018 and 2024. The primary outcome was hypotension, defined as ≥ 2 consecutive MAP readings ≤60 mmHg within 15 minutes during the first 6 hours of propofol sedation. The MIMIC-IV dataset was used for external validation. Features were selected using a union of Boruta and LASSO-penalised logistic regression. Four algorithms (elastic net-penalised logistic regression, random forest, LightGBM, and multilayer perceptron) were tuned using Bayesian optimization to minimize cross-entropy loss during leave-one-site-out cross-validation (LOSO-CV). The algorithm with the lowest cross-entropy loss during LOSO-CV was chosen as the best-performing algorithm. Classification thresholds were selected to target sensitivity ≥80%. Model performance was evaluated using area under the receiver operating characteristic curve (AUC-ROC), calibration slope and intercept, sensitivity, and specificity. Feature importance was evaluated using SHapley Additive exPlanations (SHAP). Results The development cohort included 16,361 patients from Mayo Clinic, of whom 3,482 (21.3%) developed hypotension. The external validation cohort comprised 4,684 patients from MIMIC-IV, with 115 (2.5%) experiencing hypotension. Patient characteristics included age, admission Sequential Organ Failure Assessment (SOFA) scores, starting propofol dose, ICU admission source, and comorbidities. The best-performing algorithm was LightGBM. The average AUC-ROC during LOSO-CV was 0.70 (95% confidence interval [CI] 0.68-0.71), versus 0.64 during external validation. The average calibration slope and intercept during LOSO-CV was 1.08 (95% CI 0.89-1.27) and 0.08 (95% CI -0.25-0.41), respectively, versus 0.78 and -2.07 during external validation. The average sensitivity and specificity during LOSO-CV was 0.81 (95% CI 0.79-0.83) and 0.47 (95% CI 0.42-0.52), respectively, versus 0.54 and 0.65 during external validation. Conclusions We developed and externally validated an ML-based classifier that demonstrates moderate discrimination for predicting propofol-induced hypotension in mechanically ventilated ICU patients. This tool may facilitate early identification of high-risk patients and enable proactive hemodynamic management strategies. Further prospective validation and assessment of clinical impact are warranted. This abstract is funded by: None

Abstract Rationale Propofol sedation in mechanically ventilated patients often causes clinically significant hypotension. While we concurrently developed and validated a machine-learning (ML) classifier to predict hypotension risk, accurately estimating the timing of hemodynamic deterioration could enable more precise, proactive interventions. This study aimed to develop and externally validate ML models that predict not only whether hypotension will occur, but also when it is likely to develop after propofol initiation. Methods ML models were developed using retrospective data from adult, mechanically ventilated ICU patients who received propofol infusion for ≥24 hours across four Mayo Clinic sites from 2018 to 2024 and developed hypotension (≥2 consecutive MAP readings ≤60 mmHg within 15 minutes) during the first 6 hours of sedation. The MIMIC-IV dataset was used for external validation. Features were selected using a union of Boruta and LASSO-penalised quantile regression. Two quantile algorithms (LASSO-penalised quantile regression and LightGBM) were trialed to predict the conditional median and interquartile range (IQR) of time to hypotension. In addition, two probabilistic algorithms (NGBoost and Mixture Density Network) were trialed to predict the full conditional distribution. The quantile models were tuned via Bayesian optimisation to minimise pinball loss during leave-one-site-out cross-validation (LOSO-CV), whereas probabilistic models were tuned to minimise the Continuous Ranked Probability Score. The algorithm with the lowest average pinball loss was deemed the best-performing algorithm. Model performance was evaluated using median absolute error (MAE), IQR coverage, and IQR width. Feature importance was evaluated using SHapley Additive exPlanations (SHAP). Results We included 3482 patients from the Mayo Clinic for model development, with 115 patients from the MIMIC-IV dataset used for external validation. Patient characteristics in the modeling dataset included age, sex, admission SOFA scores, weight, ICU admission source, and initial propofol dose. The best-performing algorithm was LightGBM-based quantile regression. The average MAE during LOSO-CV was 4.11 hours (95% confidence interval [CI] 3.98-4.23), versus 4.83 hours during external validation. The average IQR coverage and width during LOSO-CV was 0.49 (95% CI 0.48-0.49) and 7.12 (6.89-7.35), respectively, versus 0.07 and 6.51 during external validation. Conclusions We developed and externally validated a quantile regression model that predicts both the timing and uncertainty of propofol-induced hypotension in mechanically ventilated ICU patients. This model provides clinicians with actionable time estimates that may facilitate preemptive hemodynamic interventions and optimize sedation management. Prospective validation, expanded predictors and clinical impact assessment are warranted. This abstract is funded by: None

Abstract Rationale Lung-protective ventilation protocols use predicted body weight (PBW) to set tidal volumes. However, biases exist in these formulas, with overestimation of lung capacity in older, shorter, female, and non-White patients—groups that experience higher mortality rates during invasive mechanical ventilation (IMV). This study investigated associations between pre-admission forced vital capacity (FVC), driving pressure, and patient outcomes. Methods This multicenter retrospective study included adults (18-95 years old) with pre-admission spirometry who required IMV between 01/01/2016 and 05/31/2025 across the Mayo Clinic, spanning 11 ICUs within 5 hospitals in 4 states. Patients were stratified by FVC/PBW ratio quartiles. The primary outcome was 28-day mortality. Trends were assessed using linear or logistic regression, with FVC/PBW as a continuous variable. Mediation analysis examined pathways between FVC/PBW and mortality, with driving pressure (cm H2O, standardized) as the mediator, and adjusted for covariates including age, sex, SOFA score, and BMI. Absolute risk differences and number needed to harm (NNH) were estimated for key thresholds (e.g. FVC/PBW ≥45 mL/kg, PFTs within 2 years). Results Among 4,839 patients (38.7% female, mean (±SD) age 64.5±13.5 years), FVC/PBW ratio demonstrated substantial individual variation, with a mean of 48.3 ± 13.4 mL/kg. There was systematic variation by demographics (Figure): females had significantly lower ratios than males (45.4 versus 50.2 mL/kg, P < 0.001), and ratios were lower with older age from 52.5 mL/kg in patients <65 to 43.2 mL/kg in those ≥85 years (P < 0.001 for trend). Patients in the lowest FVC/PBW quartile had higher 28-day mortality compared to the highest quartile (26.0% versus 17.0%, P < 0.001) and higher driving pressures (14.0 versus 10.0 cm H2O, P < 0.001; Figure). In mediation analysis (n = 4,147), driving pressure mediated the relationship between FVC and mortality. Each 1-SD increase in FVC/PBW was associated with 0.364-SD lower driving pressure (P < 0.001) and 2.4% lower mortality. The adjusted indirect effect through driving pressure was significant (ACME -0.024, 95% CI: -0.029 to -0.019), accounting for the mortality effect. Sensitivity analyses across binary thresholds, age quartiles, outlier exclusion, and obstructive pathology confirmed robust mediation. Patients with FVC/PBW <45 mL/kg (N = 1,644; 39.6%) had 28-day mortality of 17.8% versus 13.5% for ≥45 mL/kg (risk difference 4.3%, 95% CI 1.8-6.8%; NNH: 23 patients). Conclusions Pre-admission lung function was strongly associated with mortality in adults receiving IMV through driving pressure-mediated pathways. Driving pressure is a modifiable target, especially in patients with smaller lung capacities at highest risk for ventilator-induced lung injury. Further prospective investigation is needed to confirm these findings. This abstract is funded by: National Heart, Lung, and Blood Institute (NHLBI) of the National Institute of Health (NIH) Grant Number K23HL151671

Introduction Angina with no obstructive coronary artery disease (ANOCA) affects millions and is frequently under-recognised because diagnostic pathways and risk tools predominantly target obstructive coronary artery disease (CAD). This protocol describes shared methods for two machine-learning (ML) studies: (1) differentiating ANOCA from stable angina with obstructive CAD and (2) predicting long-term mortality among patients with ANOCA and obstructive CAD. Methods and analysis We will develop and cross-site validate ML classification models using a multicentre retrospective cohort drawn from the Alberta Provincial Project for Outcome Assessment in Coronary Heart Disease registry and institutional datasets from the University of Ottawa Heart Institute and the University Health Network. Eligible participants are adults (≥18 years) undergoing initial cardiac catheterisation for chest pain/anginal equivalents since 1995, excluding prior revascularisation, major structural heart disease and predefined non-anginal indications. Outcomes are (1) ANOCA (0% to <50% stenosis) versus obstructive CAD (≥50% stenosis) and (2) 1, 3 and 5-year mortality, modelled separately for ANOCA and obstructive CAD. Model development will use nested cross-validation with stratified k-fold inner-loop tuning and leave-one-site-out cross-validation for repeated external validation. Candidate predictors will be harmonised across sites, filtered for missingness and refined using expert/directed acyclic graph-guided selection plus Boruta and Least Absolute Shrinkage and Selection Operator. Preprocessing includes appropriate encoding, missing-data imputation (multivariate imputation by chained equations) and feature scaling. Algorithms will include elastic-net logistic regression, random forest, LightGBM and multilayer perceptron models; hyperparameters will be optimised via Bayesian optimisation. Performance and threshold tuning will be reported. Explainability and subgroup fairness will be assessed using SHapley Additive exPlanations. Final models will be deployed as a web-based clinical risk calculator. Ethics and dissemination Ethics approval has been obtained from the University of Calgary and the University Health Network (#24-5916). Analyses will use deidentified data in secure environments; only aggregate results will be reported. Findings will be disseminated via peer-reviewed publications, conferences and a web-based calculator.

Purpose: Despite the 2022 American Heart Association/American College of Cardiology guidelines recommending cardiac rehabilitation (CR) after aortic dissection repair, both patient participation in CR and the associated clinical outcomes remain poorly understood. Methods: Adults (≥18 years of age) discharged alive following thoracic aortic dissection surgery across the Mayo Clinic Enterprise (January 2012 to November 2022), with follow-up until May 2024, were included (IRB #24-001141). Data were presented using summary statistics and logistic regression modeling. Results: A total of 186 patients were referred; 37% were female, and the median age was 64 (51, 73) years. Dissections were classified as type A (43%), type B (18%), or a combination, and 55% had residual dissection. Over half (53%) did not start CR, with older age (aOR = 0.97: 95% CI, 0.94 - 0.99; P = .003), family history of aortic dissections and/or aneurysms (aOR = 0.17: 95% CI, 0.04 - 0.54; P = .005), and tobacco use (aOR = 0.35: 95% CI, 0.13 - 0.91; P = .037) being associated with lower odds of enrollment. Of the 87 who enrolled, 34 did so at a Mayo Clinic facility. There were no complications, with no difference in adverse outcomes. Among those with evaluations at graduation from CR (n = 21), quality of life (Dartmouth index 16 [14, 19] vs 25 [22, 30]; P = .002) and 6-minute walk distance (488 [375, 531] vs 302 [235, 398] m; P = .001) improved. Conclusions: Following thoracic aortic dissection repair, CR significantly improved quality of life and functional capacity without adverse events. Further work is needed to improve enrollment and establish data-driven safety margins for exercise.

Background: Propofol is a first-line sedative for critically ill adults receiving invasive mechanical ventilation (IMV). However, it can contribute to hemodynamic instability, especially in the setting of intubation. The magnitude, timing, risk factors, and variability of sedation-associated mean arterial pressure (MAP) changes remain poorly characterized in ICU settings. Objectives: To quantify MAP changes following propofol sedation, identify risk factors for hemodynamic instability, and characterize associated interventions. Methods: We included adults (≥18 years old) across 11 ICUs in the Mayo Clinic Enterprise who required IMV and received ≥6 consecutive hours of propofol infusion between 05/05/2018 and 07/31/2024. The primary outcome was MAP change within 2-hours following sedation. Secondary outcomes included vasopressor use and hypotension (MAP ≤60 mmHg). Mixed-effects modeling was utilized to account for individual patient differences. The main model used a “neutral” adjustment, estimating a 5 mmHg MAP increase per 0.05 mcg/kg/min of norepinephrine equivalents, reduced to 3 mmHg per 0.05 mcg/kg/min after reaching a dose of 0.20 mcg/kg/min. Those on any vasopressors prior to propofol sedation initiation were analyzed separately. Results: Across 16,418 patients, 25.2% were on vasopressors before sedation initiation. Among the remaining 12,281 patients, 40.3% required vasopressors and 7.7% experienced hypotension within 2 hours of sedation ( Image 1 ). Post-intubation sedation was associated with a MAP reduction within the first 30 minutes (-6.58 mmHg; 95% CI: -6.85 to -6.32; P<0.001; Images 2 and 3 ). There was substantial inter-patient variability in both baseline MAP (9.5% variation attributable to between-patient differences) and MAP decline after sedation (40.9% between-patient differences). Higher SOFA scores (-0.31 mmHg per point), older age (-0.04 mmHg per year), and male sex (-0.47 mmHg) were associated with lower MAP. Patients with higher illness severity experienced progressively greater MAP changes over time (-0.20 mmHg per hour per increasing SOFA point; P<0.001). Conclusions: Propofol sedation is associated with clinically significant hemodynamic effects requiring intervention in the early post-intubation period. The marked inter-patient variability in hemodynamic responses highlights the importance of personalized management approaches. Risk stratification may help identify patients who could benefit from preemptive hemodynamic optimization.

IMPORTANCE: Propofol is a first-line sedative for adults receiving invasive mechanical ventilation (IMV). However, it can contribute to hemodynamic instability, especially during intubation. The magnitude, timing, risk factors, and variability of sedation-associated mean arterial pressure (MAP) changes remain poorly characterized in ICU settings. OBJECTIVES: To quantify MAP changes following propofol sedation, identify risk factors for hemodynamic instability, and characterize associated interventions. DESIGN: Retrospective cohort study. The primary outcome was MAP change within 2 hours following sedation. Secondary outcomes included vasopressor use and hypotension (MAP ≤ 60 mm Hg). Mixed-effects modeling was used to account for individual patient differences. SETTING AND PARTICIPANTS: Adults (≥ 18 yrs old) who required IMV and received greater than or equal to 6 consecutive hours of propofol infusion, between May 5th, 2018, and July 31st, 2024, in 11 ICUs across the Mayo Clinic, spanning 5 hospitals in 4 states. MAIN OUTCOMES AND MEASURES: The primary outcome was the change in MAP within 2 hours following the initiation of propofol-based sedation. RESULTS: Across 16,418 patients, 25.2% were on vasopressors before sedation initiation. Among the remaining 12,281 patients, 40.3% required vasopressors and 7.7% experienced hypotension within 2 hours of sedation. Propofol-based sedation was associated with a MAP reduction within the first 30 minutes (–6.58 mm Hg; 95% CI, –6.85 to –6.32; p < 0.001). There was substantial interpatient variability in both baseline MAP, and MAP decline after sedation (9.5 and 40.9% between-patient differences, respectively). Higher Sequential Organ Failure Assessment (SOFA) scores (–0.31 mm Hg/point), older age (–0.04 mm Hg/yr), and male sex (–0.47 mm Hg) were associated with lower MAP. Patients with higher illness severity experienced progressively greater MAP decline over time (–0.20 mm Hg/hr/SOFA point; p < 0.001). CONCLUSIONS AND RELEVANCE: Propofol-based sedation was associated with clinically significant hemodynamic effects requiring intervention in the early post-intubation period. The marked interpatient variability in hemodynamic responses highlights the importance of personalized management approaches, including risk stratification based on age, sex, and illness severity.

Machine-learning (ML) models have the potential to transform health care by enabling more personalized and data-driven clinical decision making. However, their successful implementation in clinical practice requires careful consideration of factors beyond predictive accuracy. We provide an overview of essential considerations for developing clinically applicable ML models, including methods for assessing and improving calibration, selecting appropriate decision thresholds, enhancing model explainability, identifying and mitigating bias, as well as methods for robust validation. We also discuss strategies for improving accessibility to ML models and performing real-world testing. Highlights This tutorial provides clinicians with a comprehensive guide to implementing machine-learning classification models in clinical practice. Key areas covered include model calibration, threshold selection, explainability, bias mitigation, validation, and real-world testing, all of which are essential for the clinical deployment of machine-learning models. Following these guidance can help clinicians bridge the gap between machine-learning model development and real-world application and enhance patient care outcomes.

Purpose: To develop and validate an explainable machine learning (ML) tool to help clinicians predict the risk of propofol-associated hypertriglyceridemia in critically ill patients receiving propofol sedation. Methods: Patients from 11 intensive care units (ICUs) across five Mayo Clinic hospitals were included if they met the following criteria: a) ≥ 18 years of age, b) received propofol infusion while on invasive mechanical ventilation for ≥24 h, and c) had a triglyceride level measured. The primary outcome was hypertriglyceridemia (triglyceride >400 mg/dL) onset within 10 days of propofol initiation. Both COVID-inclusive and COVID-independent modeling pipelines were developed to ensure applicability post-pandemic. Decision thresholds were chosen to maintain model sensitivity >80%. Nested leave-one-site-out cross-validation (LOSO-CV) was used to externally evaluate pipeline performance. Model explainability was assessed using permutation importance and SHapley Additive exPlanations (SHAP). Results: Among 3922 included patients, 769 (19.6%) developed propofol-associated hypertriglyceridemia, and 879 (22.4%) had COVID-19 at ICU admission. During nested LOSO-CV, the COVID-inclusive pipeline achieved an average AUC-ROC of 0.71 (95% confidence interval [CI] 0.70–0.72), while the COVID-independent pipeline achieved an average AUC-ROC of 0.69 (95% CI 0.68–0.70). Age, initial propofol dose, and BMI were the top three most important features in both models. Conclusion: We developed an explainable ML-based tool with acceptable predictive performance for assessing the risk of propofol-associated hypertriglyceridemia in ICU patients. This tool can aid clinicians in identifying at-risk patients to guide triglyceride monitoring and optimize sedative selection.

Abstract Rationale: Propofol can increase the risk of hypertriglyceridemia and subsequent pancreatitis in critically ill patients. It is unclear whether modifying or continuing sedation regimens in patients experiencing hypertriglyceridemia while on propofol infusion influences the development of adverse events such as acute pancreatitis. This study compared the clinical outcomes of patients receiving invasive mechanical ventilation (IMV) who had their propofol regimen modified versus continued following the detection of hypertriglyceridemia. Methods: A targeted trial emulation was conducted using retrospective data from 5 Mayo Clinic hospitals between 05/05/2018 and 06/30/2023. Inclusion criteria included: 1) receiving IMV and continuous propofol infusion for &gt;24 hours, 2) experiencing triglyceride level &gt;400 mg/dL while on propofol, and 3) still requiring IMV at 24-hours after hypertriglyceridemia onset. Patients were then categorized based on whether they had their propofol regimen modified (i.e., dose reduction by ≥50% or any reduction with alternative sedatives started) versus continued (no reduction or &lt;50% reduction without alternative sedative initiation). Outcomes included 30-day ventilator-free days, successful extubation, acute pancreatitis (determined by chart review, in-duplicate), ICU and hospital length of stay, 30- and 90-days alive and at home, as well as all-cause mortality. Analyses were conducted using stabilized inverse probability of treatment weights with adjustments for covariates selected a priori. Sensitivity analyses, where only benzodiazepines or benzodiazepines and ketamine were considered to be eligible alternative sedatives, were done. Results: During the study period, 728 patients were included, of whom 422 had their propofol regimen modified (Image). The two groups had similar demographics, comorbidities, and illness severity. We identified 14 cases of acute pancreatitis possibly attributable to propofol, with 9 cases in the modified and 5 cases in the continued regimen group (2.1% versus 1.6%; P=0.63). Following adjustment for covariates, there were no significant differences in any outcomes of interest. After censoring for mortality, there was no significant difference in the 30-day cumulative incidence of successful extubation when comparing those with their regimen modified versus continued (Hazard ratio [HR] 0.95; 95% confidence intervals [CI] 0.77 to 1.18; P=0.70). There was also no significant difference in the 90-day survival rate between the two groups (HR 0.89; 95% CI 0.71 to 1.11; P=0.34). These findings remained consistent upon sensitivity analyses. Conclusion: There were no differences in outcomes among patients who had their propofol regimens modified versus continued following the detection of hypertriglyceridemia. Further prospective investigations are needed to identify the optimal strategy for managing sedative regimens in this patient population.