SFSaul Flores

Saul Flores

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Saul Flores has a total of 314 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Machine Learning to Distinguish Physiologic Causes of Hypoxemia After Bidirectional Glenn Using Bedside Markers

    by Fabio Savorgnan, Pranathi Pilla, Sarah Visokay, Kaitlin Ness, Thomas Fiedorek, Saul Flores

    Background: Hypoxemia after bidirectional Glenn palliation can result from distinct physiological mechanisms that may produce similar arterial oxygen saturation. We evaluated whether routinely available bedside markers could recover physiological states encoded in a mechanistic virtual-patient model. Methods: A steady-state zero-dimensional closed-loop model generated five causes of Glenn hypoxemia: elevated pulmonary vascular resistance, veno-venous collateral shunting, pulmonary gas-exchange impairment, increased metabolic demand, and ventricular dysfunction. Hidden physiological variables were withheld from the classifier; inputs were SpO2, heart rate, mean arterial pressure, hemoglobin, Glenn/SVC pressure, and SVC oxygen saturation. Because bedside patterns arise downstream from prespecified equations and phenotype perturbations, classification was interpreted as internal mechanistic recoverability rather than clinical diagnostic performance. Results: Of 150,000 generated states, 127,162 passed broad physiological criteria and 66,269 met strict hypoxemia/hemodynamic criteria. After balancing, 26,400 virtual patients were analyzed (5280 per mechanism; 19,800 training and 6600 held-out testing). SpO2 alone achieved 26.1% accuracy (macro-AUROC 0.580); the primary bedside model achieved 56.1% accuracy (95% CI 55.0–57.4%) and macro-AUROC 0.845 (95% CI 0.840–0.851). Elevated PVR was the most readily recovered phenotype. Glenn/SVC pressure and SVC saturation provided the largest incremental contributions, while broader measurement-error assumptions substantially reduced performance. Conclusions: Within this mechanistic simulation, multivariable bedside patterns allowed partial recovery of physiological states encoded by the model. Because phenotypes and their observable consequences were generated from prespecified assumptions, the reported performance quantifies internal recoverability under those assumptions and should not be interpreted as diagnostic accuracy or as independent evidence of bedside information content in patients. The model identifies testable physiologic signatures and areas of overlap for prospective clinical validation.

  • Open Access

    Human milk and breastfeeding support in United States paediatric cardiac ICUs: a multicentre cohort and survey study

    by Kristin M. Elgersma, Brian F. Joy, Zuofu Huang, Monique R. Radman, Kimberly I. Mills, Jennifer E. Schramm, Joshua H. Wong, Meghan M. Chlebowski, Asaad G. Beshish, Raya Safa, et al.

    Human milk and direct breastfeeding are recommended for all infants, but rates are low in CHD with limited data on institutional lactation support. We aimed to describe lactation-supportive practices within US paediatric cardiac ICUs and to determine associations between these practices and human milk/breastfeeding outcomes. This retrospective cohort study and site survey included centres from Collaborative Research from the Pediatric Cardiac Intensive Care Society. Linear regression assessed associations between the total number of lactation-supportive practices and sites’ human milk and breastfeeding percentages, during the neonatal period and at discharge. Potentially strong clinical practices were identified through a phased analysis. Survey respondents represented 34 sites, with detailed data for 832 patients from 24 sites. Implementation of 22 lactation-supportive practices varied (median: 10; range: 2–20); smaller sites reported fewer practices ( p = 0.006). Common practices included in-room breast pumps (100%), feeding protocols (91%), and staff training (79%), while skin-to-skin protocols (41%) and automatic lactation orders (44%) were underutilised. While total practices did not correlate with outcomes, sites with high versus low implementation of “strongest” practices demonstrated generally better lactation outcomes, especially at discharge: 11.7% higher human milk volume (95% CI: 0.1%–23.3%); 8.2% higher breastfeeding rate (95% CI: −0.3%–16.6%). These strongest practices involved institutional formalisation of lactation support. Lactation-supportive practices varied among sites. Findings suggest formalisation of practices including a developmental care protocol, comprehensive parent education/support, and protocols recommending human milk/breastfeeding may improve lactation outcomes in critical CHD.

  • Open Access

    Machine Learning and High-Fidelity Data in Medicine and Surgery

    by Rohit S. Loomba, Saul Flores, Izzet Turkalp Akbasli, Wesam Sourour, Fabio Savorgnan

    Artificial intelligence has become increasingly integrated into medicine and surgery, yet much of its clinical impact has historically relied on models trained using low resolution and intermittently sampled data. High fidelity clinical data, defined by continuous acquisition, fine temporal resolution, and direct physiologic measurement, constitute a fundamentally different substrate for machine learning algorithms. Continuous monitoring tools such as near infrared spectroscopy and central venous oxygen saturations provide an uninterrupted stream of information that captures the dynamic behavior of human physiology. These data streams reveal complex patterns of physiologic compensation, adaptation, and eventual failure that remain largely invisible to traditional static clinical datasets. By tracking the delicate balance between oxygen delivery and systemic oxygen extraction in real time, clinicians can better understand the adequacy of perfusion before overt hemodynamic collapse occurs. When paired with machine learning methods capable of modeling temporal dependence, nonlinear interactions, and physiologic coupling, this rich stream of information enables a crucial transition from static risk prediction toward dynamic physiologic understanding. This chapter examines the conceptual foundations, analytic paradigms, data engineering requirements, and specific clinical applications of machine learning using high fidelity data, with particular emphasis on pediatric cardiovascular surgery and cardiac intensive care settings. Key methodologic challenges, interpretability considerations, and the ethical, regulatory, and operational constraints of deploying these computational models at the bedside are thoroughly discussed. The review concludes by highlighting future directions toward real time learning health systems and patient specific digital physiologic models that prioritize the continuous optimization of systemic oxygen delivery and extraction.

  • Artificial intelligence in paediatric cardiac critical care: from predictive analytics to continuous physiologic intelligence

    by Fabio Savorgnan, Pranathi Pilla, Rohit Loomba, Saul Flores

    Objective: Artificial intelligence is increasingly applied in paediatric cardiology and cardiac intensive care, where CHD, postoperative physiologic instability, and continuous bedside monitoring generate complex datasets. This review summarises current applications of artificial intelligence in paediatric cardiology and the cardiac ICU, focusing on cardiac imaging, electrocardiography, predictive analytics, waveform intelligence, and postoperative outcome prediction. Data sources: A narrative review of PubMed, MEDLINE, and Scopus was performed for studies published from 2018 through 2026 evaluating artificial intelligence, machine learning, or deep learning in paediatric cardiology, CHD, congenital heart surgery, and cardiac critical care. Study selection: Studies addressing paediatric or CHD-focused applications in echocardiography, cardiac MRI, electrocardiography, cardiac ICU monitoring, waveform analytics, or postoperative complication prediction were included, together with relevant reviews. Data extraction: Information was extracted on study population, artificial intelligence method, clinical task, model inputs, outcomes, validation strategy, and major limitations. Data synthesis: Artificial intelligence is most developed in imaging and electrocardiography, while paediatric and complex CHD evidence remains less mature. In cardiac ICU, models using electronic health record variables, high-frequency physiologic data, and risk indices have been evaluated for cardiac arrest, low cardiac output physiology, hyperlactatemia, extubation failure, and vasoactive de-escalation. External validation, prospective deployment, calibration, and workflow-safe implementation remain major gaps. Conclusions: Artificial intelligence may improve diagnostic consistency, risk stratification, and early recognition of deterioration, but prospective validation and measurable clinical benefit are needed before autonomous use.

  • Open Access

    A Digital Model to Explain the Necessity of Prostaglandin E1 After Balloon Atrial Septostomy in D-Transposition of the Great Arteries

    by Fabio Savorgnan, Saul Flores, Julia Garcia-Mancebo, Adel Hassan, Rohit S. Loomba, Sebastian Acosta

    Objective: Patients with D-transposition of the great arteries (TGA) often require prostaglandin E1 (PGE) even after balloon atrial septostomy (BAS). This study builds a computer simulation that analyzes whether or not elevated pulmonary vascular resistance (PVR) could explain the profound hypoxemia seen in some patients after PGE discontinuation. Methods: We developed a systems-based mathematical model of TGA incorporating systemic and pulmonary circulations, an atrial septal defect (ASD), and a patent ductus arteriosus (PDA), with bidirectional atrial mixing. The PGE-on pre-BAS state represented restrictive atrial communication, with systemic arterial saturation in the clinically expected 60–70% range. BAS was modeled by reducing ASD resistance and increasing the atrial mixing parameter; PGE withdrawal was modeled by increasing PDA resistance. A reproducible Monte Carlo cohort of 500 virtual patients was generated using independent probability distributions for heart rate, PVR, SVR, ASD resistance, PDA resistance, and atrial mixing. Prespecified sensitivity analyses varied ASD resistance reduction, PDA resistance, and pulmonary and systemic vascular responses to PGE withdrawal. Results: The PGE-on pre-BAS cohort had a median systemic arterial saturation of 64.7% (interquartile range, 59.1–68.8%), which increased to 74.7% (70.6–77.9%) after BAS. Following PGE withdrawal, saturation decreased to 70.9% (66.0–74.4%), while systemic flow increased from 1.83 to 2.04 L/min/m2 and systemic oxygen delivery increased from 271.4 to 283.9 mL O2/min/m2. For the response from PGE-on post-BAS to PGE-off post-BAS, each 1-WU·m2 increase in PVR was associated with a 0.56-percentage-point greater decrease in saturation, a 6.56% smaller increase in systemic flow, a 3.96% greater decrease in effective pulmonary flow, and a 7.87% smaller increase in oxygen delivery (all p < 0.01). Conclusions: In a model representing clinically restrictive pre-BAS atrial communication, PVR strongly modified the response to PGE withdrawal after BAS. Higher PVR was associated with larger decreases in saturation and effective pulmonary flow and smaller improvements in systemic flow and oxygen delivery. Saturation and oxygen delivery may move in opposite directions; therefore, assessment after PGE withdrawal should integrate systemic perfusion rather than rely on saturation alone. These findings are mechanistic and require external clinical validation.

  • Open Access

    A Digital Model to Explain the Necessity of Prostaglandin E1 After Balloon Atrial Septostomy in D-Transposition of the Great Arteries

    by Fabio Savorgnan, Saul Flores, Rohit S. Loomba

    Objective: Patients with D-Transposition of the Great Arteries (TGA) often require prostaglandin E1 (PGE) even after balloon atrial septostomy (BAS). This study builds a computer simulation that analyzes whether or not elevated pulmonary vascular resistance (PVR) could explain the profound hypoxemia seen in some patients after PGE discontinuation. Methods: We developed a systems-based mathematical model of TGA incorporating systemic and pulmonary circulations, atrial septal defect (ASD), and patent ductus arteriosus (PDA) with bidirectional atrial mixing. The effects of BAS and PGE withdrawal were modeled by altering atrial and ductal resistances, respectively. Using Monte Carlo sampling, 500 virtual patients with variable pulmonary vascular resistance (PVR) and systemic vascular resistance (SVR) were simulated. Key outcomes included relative changes in systemic oxygen saturation (ΔSsa), systemic flow (ΔQs/Qs), effective pulmonary flow (ΔQp,eff/Qp,eff), and oxygen delivery (ΔDO₂/DO₂). Results: Regression analysis demonstrated that each unit increase in PVR was associated with a 0.4% decrease in ΔSsa, an 8.8% reduction in ΔQs/Qs, an 8.8% reduction in ΔQp,eff/Qp,eff, and a 9.5% reduction in ΔDO₂/DO₂ (all p< 0.01). In contrast, SVR had smaller but favorable effects, with each unit increase associated with a 0.3% improvement in ΔSsa, 0.6% increase in ΔQs/Qs, 1.6% increase in ΔQp,eff/Qp,eff, and 1.0% increase in ΔDO₂/DO₂. Importantly, discontinuation of PGE and ductal closure often resulted in lower systemic oxygen saturation but higher systemic oxygen delivery due to increased systemic blood flow, an effect most pronounced in patients with low PVR. Conclusions: The effectiveness of BAS in eliminating ductal dependence is strongly influenced by vascular tone. Elevated PVR markedly limits the benefit of atrial decompression, while higher SVR exerts modestly protective effects. These findings explain clinical variability in BAS outcomes and highlight that systemic oxygen delivery—not saturation alone—should guide assessment of post-BAS physiology in neonates with TGA.

  • Open Access

    Hyperacute effects of non-code dose bolus epinephrine in paediatric cardiac intensive care patients: insights from high-fidelity physiologic data

    by Wesam Sourour, Michael Evans, Kevin Le, Saul Flores, Juan S. Farias, Rohit Seth Loomba

    Background: Non-code dose boluses of epinephrine are utilised in critically ill paediatric patients during periods of hemodynamic deterioration, often with the hopes of preventing a cardiac arrest. Data regarding the physiologic effects of these administrations are limited. The primary aim of this study was to use high-fidelity physiologic data to characterise the effects of intravenous non-code dose bolus epinephrine. Methods: Paediatric patients in the cardiac ICU who received non-code dose bolus epinephrine were identified. Those who received fluid boluses or chest compressions within 2 minutes of bolus epinephrine were excluded. Autoregressive integrated moving average analyses with exogenous variables were conducted to characterise the time-dependent changes in hemodynamic indices. Cluster analyses were then conducted to determine patterns in hemodynamic changes associated with bolus epinephrine. Results: A total of 71 non-code dose bolus epinephrine administrations were included in the final analyses. Heart rate, blood pressure, and renal near infrared spectroscopy all demonstrated statistically significant changes after bolus epinephrine administration. Peak change in each was 40%, 52%, and 9%, respectively, with peaks occurring between 60 seconds and 120 seconds after administration. Three response-based clusters were identified. Conclusion: Non-code dose bolus epinephrine is associated with a significant increase in heart rate, blood pressure, and systemic oxygen delivery. Cluster analysis using the peak change identified distinct clinical clusters.

  • Open Access

    The Agreement Between Pulse Oximetry and Measured Arterial Oxygen Saturations in Postoperative Functionally Univentricular Patients

    by Fabio Savorgnan, Sebastian Acosta, Joshua Prabhu, Pranathi Pilla, Vikram Shah, Saul Flores, Rohit S. Loomba

    Background: Pulse oximetry is widely used to estimate arterial oxygen saturation, yet accuracy may vary for a number of reasons. Data on children with functionally univentricular circulation are limited. The primary aim of this study was to evaluate the agreement between arterial oxygen saturation measured by blood gas and pulse oximetry in children with functionally univentricular circulations. Methods: A retrospective analysis was performed of paired arterial blood gas and pulse oximetry oxygen saturation measurements following Norwood, Glenn, or Fontan procedures. Signed difference was defined as arterial oxygen saturation by blood gas—arterial oxygen saturation by pulse oximetry. Bland–Altman analyses, multivariable regressions, and generalized additive modeling were performed. Results: Mean bias was −4.9 percentage points, indicating pulse oximetry overestimated arterial saturation. The 95% limits of agreement were wide, from −20.7 to 10.8. The agreement was similar in Black and White patients. Fontan physiology demonstrated reduced overestimation by pulse oximetry by multivariable regression. Nonlinear modeling demonstrated more bias in agreement at lower arterial oxygen saturation levels, with arterial oxygen saturation levels explaining 50% of the variance. Conclusions: In functionally univentricular patients, pulse oximetry using the Nellcor MAXN-NS pulse oximeter (Medtronic, Dublin, Ireland) systematically overestimates arterial saturation, particularly in the setting of hypoxemia. Saturation level, rather than race, was the dominant determinant of bias.

  • Good Practice Concepts in Pediatric Critical Care Point-of-Care Ultrasound: A Modified Delphi Consensus Initiative

    by Jamie S. Penk, Vidit Bhargava, Harsha Chandnani, Grace Chong, Thomas Conlon, Ryan L. DeSanti, J. Wesley Diddle, Alejandro Floh, Saul Flores, Anyir Hsieh, et al.

    OBJECTIVE: We assembled a workgroup within the Pediatric Research Collaborative on Critical UltraSound (PeRCCUS), a subgroup of the Pediatric Acute Lung Injury and Sepsis Investigators (PALISI), to define early guidance for point-of-care ultrasound (POCUS) institutional practice and foster future comprehensive guidelines for its broad adoption in pediatric critical care medicine. DESIGN: A modified Delphi method was used for creating the statements. The first meeting was an open proposal session for workgroup members to suggest items for consideration. This was followed by a cycle of voting for levels of agreement along a 7-point Likert-type scale. Items were reviewed, with only items receiving a score of greater than or equal to 6 progressing to the next stage of voting and lower-scoring items reconsidered, with only consensus items proceeding to the next stage for additional rounds of voting until consensus was reached. SETTING: Multi-institutional, multidisciplinary, workgroup of experts on POCUS organized within PeRCCUS as a subgroup of PALISI. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Consensus was obtained for 25 recommendations across five domains: clinical application, quality assurance, equipment, education, and research. CONCLUSIONS: We report consensus recommendations for institutions on clinical use, educational programs, quality assurance, technical requirements, and future research opportunities for the adoption of pediatric critical care medicine POCUS.