Using Machine Learning to Predict AKI and Augment Mannitol Concentrations in Cardioplegia Solution During Coronary Artery Bypass

Authors: Parker Wilson, BS, Jessa Deckwa, BS, Dr. Zain Khalpey, MD, PhD, FACS

Introduction:

Artificial Intelligence (AI) and Machine Learning (ML) can be used to enhance prevention of acute kidney injury (AKI) in the peri-operative period with coronary artery bypass (CABG) patients. We suggest that an algorithm can be created to identify patients at highest risk for AKI pre-operatively so that interventions can be made to mitigate this risk. Additionally, we believe this algorithm could provide recommendations for intervention through augmentation of Mannitol concentrations in del Nido cardioplegic solutions used in CABG. The concentration can be changed based on pre-operative and intra-operative measures of renal function including urine output, Creatinine, NephroCheck and Near-Infrared Spectroscopy (NIRS). Because mannitol is renal-protective, we believe that useful machine learning algorithms can be devised from existing evidence and our own research to create optimal concentrations of mannitol in cardioplegic solutions to maximally reduce AKI incidence in CABG patients on cardiopulmonary bypass.

Acute Kidney Injury:

Acute kidney injury (AKI) following Coronary Artery Bypass Graft (CABG) is a peri-operative adverse event that worsens short- and long-term outcomes. AKI occurs in 20-30% of procedures that utilitze cardiopulmonary bypass (CPB), such as CABG. Additionally, there is a reported AKI-related mortality of 4.5% in all cardiac surgery.1 Renal replacement therapy (RRT) is required in 2-4% of vascular and cardiac surgery patients.2 For this reason, it is imperative to reduce the risk and incidence of AKI in cardiac surgery, especially CABG when CPB is often utilized.

It is theorized that kidney injury during CPB is often caused by renal hypoperfusion. Measures are often taken to monitor renal function throughout surgery including measuring urine output and metabolic panels, but few interventions are possible to reduce kidney injury before or during operations. Crystalloid solution in addition to cell-saving techniques with CPB are often infused throughout surgery which can reduce renal hypoperfusion, yet we still see high incidences of AKI despite these standard interventions.

The Role of Mannitol:

Mannitol is a sugar alcohol often used in del Nido cardioplegic solutions because it was shown to reduce post-operative arrhythmia and myocardial damage due to free radical scavenging ability of mannitol.3,4,5 As a beneficial side effect, mannitol has also been shown to be renal-protective, therefore, preventative of AKI. Mannitol increases renal perfusion through vasodilatation, thereby improving renal blood flow (RBF) and glomerular filtration rate (GFR).6 Thus mannitol has become an accepted and often used constituent of cardioplegia solutions, though evidence in recent decades have questioned these early studies. A 2019 systematic analysis by Waskowski et al reviewed 7 randomized-controlled trials (RCTs) investigating mannitol and renal protection in elective cardiac surgery patients.7 Fischer et al showed increased urine output (UOP), a surrogate for renal function, in patients receiving increasing concentrations of mannitol compared to placebo.8 Ip-Yam et al, however, showed no significant differences in a variety of renal function measurements between normothermic and hypothermic groups receiving either 0.5 mg/kg mannitol in CPB prime solution versus standard of care.9 Other RCTs often showed no significant differences between mannitol groups and placebo. Evidence for prevention of renal dysfuction with mannitol has, therefore, become mixed.

Our Study:

We at Khalpey AI Lab have sought to add to the body of evidence in support of the renal effects of Mannitol in cardioplegic solutions used in CABG using a variety of markers for renal function including NIRS, which is a novel non-invasive measure of perfusion since the outflow of evidence in the 1990s. We performed a single-center retrospective analysis, which included CABG performed by a single surgeon (2019-2021). Thirty patients (65% male, mean age 67 + 8.25 yrs) were stratified into receiving varying concentrations of Mannitol in del Nido cardioplegic solutions (0 mg, 12.5 mg, 25 mg). Markers of urinary perfusion and function were recorded throughout the peri-operative period including near-infrared spectroscopy (NIRS). AKI in these patients was defined by KDIGO criteria.

Of the patients included in our study, 40% (n=4/10) of subjects receiving 25 mg Mannitol developed AKI whereas 50% (n=5/10) and 40% (n=4/10) of subjects receiving 12.5 mg and 0 mg respectively developed AKI. A significant decrease in NIRS (-5.9, p = 0.024) for patients receiving 0 mg Mannitol was found along with a significant difference between mean post-op urine output between patients in the 25 mg and 0 mg groups. Mean changes in creatinine peri-operatively (pre-op to 72 hrs post-op) did not differ significantly between the groups (-0.08 vs 0.4 vs -0.05) and neither did incidence of AKI. Otherwise there were no significant differences in renal functions between the 3 groups.

Discussion:

Our data did not significantly show that mannitol in del Nido cardioplegic solution is protective against AKI in CABG patients, although increased UOP with increasing concentrations of mannitol aligns with previous studies that indicated this as improved renal function. Additionally, significantly decreased NIRS in the control group without mannitol may indicate decreased renal perfusion. We believe NIRS can be another useful measurement for confirmation of the protective effects of mannitol, but further investigation is necessary. A larger or better stratified study would provide more significant evidence for this conclusion.

Artificial Intelligence

When studies further show the renal protective effects of mannitol, a spectrum of renal function measurements can be compiled with a stratification of patients to build a database on which a machine learning algorithm can be devised. This algorithm will be able to show what patients are at highest risk of developing AKI using pre-operative measurements. It would then designate the optimal mannitol concentration for patients in order to mitigate AKI risk. It could even evaluate intra-operative deliverance of mannitol if it is linked with the non-invasive measurements we already monitor. If NIRS, UOP and serum creatinine are continuously evaluated by this algorithm intra-operatively, mannitol concentrations could feasibly be augmented to optimally improve renal function, thereby, reducing incidence of AKI.

Algorithms such as this one are part of our vision for the future of healthcare. Evidence-based ML algorithms can serve to improve and optimize care for patients inside the operating room, helping surgeons to reduce adverse events. Predictive algorithms that allow better, more comprehensive screening of patients will allow surgeons to better prepare for caring for critically ill patients. The development of such technology is part of the revolution of AI inside the OR and we at Khalpey AI Lab are leading the way.