Rohan Goswami

Mayo Clinic, United States

Title: Ai In Predicting Heart Recovery In Shock

---

Biography

Rohan Goswami, M.D. is an advanced heart failure and transplantcardiologist in the Transplant Center at Mayo Clinic hospital in Jacksonville,Florida. Goswami received his medical degree from the American University ofthe Caribbean School of Medicine, Sint Maarten, Netherlands Antilles, and wenton to complete an internal medicine residency at The Stamford Hospital/ColumbiaUniversity College of Physicians and Surgeons in Stamford, Connecticut, wherehe was Chief Resident. During his time at Stamford Hospital, he developed andimplemented a resident education system to help improve resident knowledge,patient care and patient outcomes. He then completed a fellowship incardiovascular diseases at the University of Tennessee Health Science Center inMemphis, Tennessee, and his advanced heart failure and transplant cardiologyfellowship at Mayo Clinic College of Medicine in Jacksonville.  Dr. Goswami has a background in informationtechnology and is focused on innovation and clinical research.  He has been invited to speak at the World AISummit and keynotes on AI in Transplant. He also is an avid mountaineer andclimbed Mount Kilimanjaro with his father in 2016.

Abstract

Background: Machine learning (ML) for heart failure cardiogenic shock (HFCS) management has not been performed with high fidelity.  Multiple data platforms for retrospectively analysing data have been published to dichotomize shock patients.  In the era of more frequently deployed temporary mechanical circulatory support for HFCS, we used ML-based assessment to develop clinical profiles to aid in medical management and assess for early LV remodeling after Impella 5.5 placement. 

Objectives: To evaluate the ability of machine learning algorithms to define clinical profiles to guide practical management and provide early markers of left ventricular (LV) remodeling in patients with acute or chronic HFCS. 

Methods: We performed a retrospective cohort study analysing multiple hemodynamic, echo, and demographic parameters utilizing specific machine-learning methods to create outcome-based clusters related to early signs of LV remodelling within two weeks after Impella 5.5 insertion for HFCS. Models were assessed through model accuracy, precision, recall, and f1-scores.  

Results: A total of 131,072 parameter combinations were assessed within our ML model.  Three distinct clinical and patient profiles were associated with varying degrees of LV change after Impella 5.5 placement.  When assessing the likelihood of LV remodelling, our approach achieved an overall prediction accuracy of 81.67%.  This signifies a high level of reliability in identifying patients with substantial potential for heart remodelling within two weeks of Impella 5.5 placement.  Our analysis identified three distinct patient clusters based on hemodynamic parameters, cardiac function metrics, and baseline characteristics, revealing significant variations in cardiovascular profiles among patients after Impella 5.5 device insertion   The median age of our cohort at Impella placement was 62 years (55 – 68) old.  The median ejection fraction was 20% (15 – 24) with a baseline left ventricular end-diastolic dimension of 68mm (62 – 71.2).  The median duration of support after Impella 5.5 was 22 days (14 – 35).  The average right atrial pressure was 11mmHg, with a mean PA average of 36mmHg, PCWP of 26mmHg, and Fick cardiac index of 2.08 L/min/m2.

Conclusions: This proof of concept paper demonstrates the utility of ML to identify features that are not immediately apparent to the clinician and may provide the basis for further exploration into the role of temporary mechanical circulatory support in patients with HFCS and the outcome of transplant, durable LVAD, or native cardiac recovery.