Differential responses of cerebral and renal oxygenation to altered perfusion conditions during experimental cardiopulmonary bypass in sheep.

We tested whether the brain and kidney respond differently to cardiopulmonary bypass (CPB) and to changes in perfusion conditions during CPB. Therefore, in ovine CPB, we assessed regional cerebral oxygen saturation (rSO2 ) by near-infrared spectroscopy and renal cortical and medullary tissue oxygen tension (PO2 ), and, in some protocols, brain tissue PO2 , by phosphorescence lifetime oximetry. During CPB, rSO2 correlated with mixed venous SO2 (r = 0.78) and brain tissue PO2 (r = 0.49) when arterial PO2 was varied. During the first 30 min of CPB, brain tissue PO2 , rSO2 and renal cortical tissue PO2 did not fall, but renal medullary tissue PO2 did. Nevertheless, compared with stable anaesthesia, during stable CPB, rSO2 (66.8 decreasing to 61.3%) and both renal cortical (90.8 decreasing to 43.5 mm Hg) and medullary (44.3 decreasing to 19.2 mm Hg) tissue PO2 were lower. Both rSO2 and renal PO2 increased when pump flow was increased from 60 to 100 mL kg-1 min-1 at a target arterial pressure of 70 mm Hg. They also both increased when pump flow and arterial pressure were increased simultaneously. Neither was significantly altered by partially pulsatile flow. The vasopressor, metaraminol, dose-dependently decreased rSO2 , but increased renal cortical and medullary PO2 . Increasing blood haemoglobin concentration increased rSO2 , but not renal PO2 . We conclude that both the brain and kidney are susceptible to hypoxia during CPB, which can be alleviated by increasing pump flow, even without increasing arterial pressure. However, increasing blood haemoglobin concentration increases brain, but not kidney oxygenation, whereas vasopressor support with metaraminol increases kidney, but not brain oxygenation.

Prolonged Non-Steroidal Anti-Inflammatory Drug Exposure After Pleurodesis Increases Pneumothorax Recurrence: A Retrospective Cohort Study.

INTRODUCTION: The use of non-steroidal anti-inflammatory drugs (NSAID) in patients undergoing pleurodesis remains controversial. Although many surgeons are comfortable prescribing NSAIDs post-operatively, some oppose this practice due to concerns of suppressing the inflammatory response and quality of pleurodesis. Only a small body of inconsistent publications exists with respect to guiding therapy in this common clinical scenario. METHODS: A retrospective cohort study was undertaken assessing effect of NSAID exposure on pleurodesis outcomes. An institutional thoracic surgery database was reviewed yielding 147 patients who underwent pleurodesis for pneumothorax between 2010 and 2018. Medical records and imaging were reviewed for patient characteristics, NSAID exposure, recurrent pneumothorax and other adverse events. RESULTS: There was no overall difference between rates of recurrence and procedural failure of pleurodesis (Relative Risk [RR] 1.67 [95% CI 0.74-3.77]). However, NSAID exposure of >48 hours was associated with increased risk of recurrent pneumothorax (RR 2.16 [95% CI 1.05-4.45]). There was no increased rate of other adverse events related to NSAID usage. CONCLUSIONS: NSAID exposure does not increase failure rates or other adverse events following pleurodesis for pneumothorax. However, prolonged NSAID exposure post-pleurodesis may increase procedural failure rates. Further large volume randomised control trials are required.

Paying attention to cardiac surgical risk: An interpretable machine learning approach using an uncertainty-aware attentive neural network.

Machine learning (ML) is increasingly applied to predict adverse postoperative outcomes in cardiac surgery. Commonly used ML models fail to translate to clinical practice due to absent model explainability, limited uncertainty quantification, and no flexibility to missing data. We aimed to develop and benchmark a novel ML approach, the uncertainty-aware attention network (UAN), to overcome these common limitations. Two Bayesian uncertainty quantification methods were tested, generalized variational inference (GVI) or a posterior network (PN). The UAN models were compared with an ensemble of XGBoost models and a Bayesian logistic regression model (LR) with imputation. The derivation datasets consisted of 153,932 surgery events from the Australian and New Zealand Society of Cardiac and Thoracic Surgeons (ANZSCTS) Cardiac Surgery Database. An external validation consisted of 7343 surgery events which were extracted from the Medical Information Mart for Intensive Care (MIMIC) III critical care dataset. The highest performing model on the external validation dataset was a UAN-GVI with an area under the receiver operating characteristic curve (AUC) of 0.78 (0.01). Model performance improved on high confidence samples with an AUC of 0.81 (0.01). Confidence calibration for aleatoric uncertainty was excellent for all models. Calibration for epistemic uncertainty was more variable, with an ensemble of XGBoost models performing the best with an AUC of 0.84 (0.08). Epistemic uncertainty was improved using the PN approach, compared to GVI. UAN is able to use an interpretable and flexible deep learning approach to provide estimates of model uncertainty alongside state-of-the-art predictions. The model has been made freely available as an easy-to-use web application demonstrating that by designing uncertainty-aware models with innately explainable predictions deep learning may become more suitable for routine clinical use.

Intraoperative hemodynamics and risk of cardiac surgery-associated acute kidney injury: An observation study and a feasibility clinical trial.

Targeting greater pump flow and mean arterial pressure (MAP) during cardiopulmonary bypass (CPB) could potentially alleviate renal hypoxia and reduce the risk of postoperative acute kidney injury (AKI). Therefore, in an observational study of 93 patients undergoing on-pump cardiac surgery, we tested whether intraoperative hemodynamic management differed between patients who did and did not develop AKI. Then, in 20 patients, we assessed the feasibility of a larger-scale trial in which patients would be randomized to greater than normal target pump flow and MAP, or usual care, during CPB. In the observational cohort, MAP during hypothermic CPB averaged 68.8 ± 8.0 mmHg (mean ± SD) in the 36 patients who developed AKI and 68.9 ± 6.3 mmHg in the 57 patients who did not (p = 0.98). Pump flow averaged 2.4 ± 0.2 L/min/m2 in both groups. In the feasibility clinical trial, compared with usual care, those randomized to increased target pump flow and MAP had greater mean pump flow (2.70 ± 0.23 vs. 2.42 ± 0.09 L/min/m2 during the period before rewarming) and systemic oxygen delivery (363 ± 60 vs. 281 ± 45 mL/min/m2 ). Target MAP ≥80 mmHg was achieved in 66.6% of patients in the intervention group but in only 27.3% of patients in the usual care group. Nevertheless, MAP during CPB did not differ significantly between the two groups. We conclude that little insight was gained from our observational study regarding the impact of variations in pump flow and MAP on the risk of AKI. However, a clinical trial to assess the effects of greater target pump flow and MAP on the risk of AKI appears feasible.

Associations Between Systemic and Cerebral Inflammation in an Ovine Model of Cardiopulmonary Bypass.

BACKGROUND: Intraoperative inflammation may contribute to postoperative neurocognitive disorders after cardiac surgery requiring cardiopulmonary bypass (CPB). However, the relative contributions of general anesthesia (GA), surgical site injury, and CPB are unclear. METHODS: In adult female sheep, we investigated (1) the temporal profile of proinflammatory and anti-inflammatory cytokines and (2) the extent of microglia activation across major cerebral cortical regions during GA and surgical trauma with and without CPB (N = 5/group). Sheep were studied while conscious, during GA and surgical trauma, with and without CPB. RESULTS: Plasma tumor necrosis factor-alpha (mean [95% confidence intervals], 3.7 [2.5-4.9] vs 1.6 [0.8-2.3] ng/mL; P = .0004) and interleukin-6 levels (4.4 [3.0-5.8] vs 1.6 [0.8-2.3] ng/mL; P = .029) were significantly higher at 1.5 hours, with a further increase in interleukin-6 at 3 hours (7.0 [3.7-10.3] vs 1.8 [1.1-2.6] ng/mL; P < .0001) in animals undergoing CPB compared with those that did not. Although cerebral oxygen saturation was preserved throughout CPB, there was pronounced neuroinflammation as characterized by greater microglia circularity within the frontal cortex of sheep that underwent CPB compared with those that did not (0.34 [0.32-0.37] vs 0.30 [0.29-0.32]; P = .029). Moreover, microglia had fewer branches within the parietal (7.7 [6.5-8.9] vs 10.9 [9.4-12.5]; P = .001) and temporal (7.8 [7.2-8.3] vs 9.9 [8.2-11.7]; P = .020) cortices in sheep that underwent CPB compared with those that did not. CONCLUSIONS: CPB enhanced the release of proinflammatory cytokines beyond that initiated by GA and surgical trauma. This systemic inflammation was associated with microglial activation across 3 major cerebral cortical regions, with a phagocytic microglia phenotype within the frontal cortex, and an inflammatory microglia phenotype within the parietal and temporal cortices. These data provide direct histopathological evidence of CPB-induced neuroinflammation in a large animal model and provide further mechanistic data on how CPB-induced cerebral inflammation might drive postoperative neurocognitive disorders in humans.

Association Between Changes in Norepinephrine Infusion Rate and Urinary Oxygen Tension After Cardiac Surgery.

OBJECTIVES: To determine if the administration of norepinephrine to patients recovering from on-pump cardiac surgery is associated with changes in urinary oxygen tension (PO2), an indirect index of renal medullary oxygenation. DESIGN: Single center, prospective observational study. SETTING: Surgical intensive care unit (ICU). PARTICIPANTS: A nonconsecutive sample of 93 patients recovering from on-pump cardiac surgery. MEASUREMENTS AND MAIN RESULTS: In the ICU, norepinephrine was the most commonly used vasopressor agent (90% of patients, 84/93), with fewer patients receiving epinephrine (48%, 45/93) or vasopressin (4%, 4/93). During the 30-to-60-minute period after increasing the infused dose of norepinephrine (n = 89 instances), urinary PO2 decreased by (least squares mean ± SEM) 1.8 ± 0.5 mmHg from its baseline level of 25.1 ± 1.1 mmHg. Conversely, during the 30-to-60-minute period after the dose of norepinephrine was decreased (n = 134 instances), urinary PO2 increased by 2.6 ± 0.5 mmHg from its baseline level of 22.7 ± 1.2 mmHg. No significant change in urinary PO2 was detected when the dose of epinephrine was decreased (n = 21). There were insufficient observations to assess the effects of increasing the dose of epinephrine (n = 11) or of changing the dose of vasopressin (n <4). CONCLUSIONS: In patients recovering from on-pump cardiac surgery, changes in norepinephrine dose are associated with reciprocal changes in urinary PO2, potentially reflecting an effect of norepinephrine on renal medullary oxygenation.

Adverse events in the digital age and where to find them.

Exponential growth of health-related data collected by digital tools is a reality within pharmaceutical and medical device research and development. Data generated through digital tools may be categorized as relevant to efficacy and/or safety. The enormity of these data requires the adoption of new approaches for processing and evaluation. Recognition of patterns within the safety data is vital for sponsors seeking regulatory approval for their new products. Nontraditional data sources may contain relevant safety information; early evaluation of these data will help to determine the product safety profile. Advanced technologies have allowed the development of digital tools to screen these data, which in some situations are classified as software as a medical devices and subject to clinical evaluation and post-marketing surveillance. Artificial intelligence may help to reduce or even eliminate noise from within these data, allowing safety experts to focus on the most pertinent evidence. We propose a data typology and provide considerations on how to define adverse events within different types of data, even where no human reporter exists. Proposals are made for the automation of screening processes. We consider validation aspects to support solutions that are proven to produce reliable results, and to deliver trusted outputs to stakeholders.

Influence of moderate hypothermia on renal and cerebral haemodynamics and oxygenation during experimental cardiopulmonary bypass in sheep.

AIM: Cardiac surgery requiring cardiopulmonary bypass (CPB) can result in renal and cerebral injury. Intraoperative tissue hypoxia could contribute to such organ injury. Hypothermia, however, may alleviate organ hypoxia. Therefore, we tested whether moderate hypothermia (30°C) improves cerebral and renal tissue perfusion and oxygenation during ovine CPB. METHODS: Ten sheep were studied while conscious, under stable anesthesia, and during 3 h of CPB. In a randomized within-animal cross-over design, five sheep commenced CPB at a target body temperature of 30°C (moderate hypothermia). After 90 min, the body temperature was increased to 36°C (standard procedure). The remaining five sheep were randomized to the opposite order of target body temperature. RESULTS: Compared with the standard procedure, moderately hypothermic CPB reduced renal oxygen delivery (-34.8% ± 19.6%, P = 0.003) and renal oxygen consumption (-42.7% ± 35.2%, P = 0.04). Nevertheless, moderately hypothermic CPB did not significantly alter either renal cortical or medullary tissue PO2 . Moderately hypothermic CPB also did not significantly alter cerebral perfusion, cerebral tissue PO2 , or cerebral oxygen saturation compared with the standard procedure. Compared with the anesthetized state, the standard procedure reduced renal medullary PO2 (-21.0 ± 13.8 mmHg, P = 0.014) and cerebral oxygen saturation (65.0% ± 7.0% to 55.4% ± 9.6%, P = 0.022) but did not significantly alter either renal cortical or cerebral PO2 . CONCLUSION: Ovine experimental CPB leads to renal medullary tissue hypoxia. Moderately hypothermic CPB did not improve cerebral or renal tissue oxygenation. In the kidney, this is probably because renal tissue oxygen consumption is matched by reduced renal oxygen delivery.

Dynamic responses of renal oxygenation at the onset of cardiopulmonary bypass in sheep and man.

INTRODUCTION: The renal medulla is susceptible to hypoxia during cardiopulmonary bypass (CPB), which may contribute to the development of acute kidney injury. But the speed of onset of renal medullary hypoxia remains unknown. METHODS: We continuously measured renal medullary oxygen tension (MPO2) in 24 sheep, and urinary PO2 (UPO2) as an index of MPO2 in 92 patients, before and after induction of CPB. RESULTS: In laterally recumbent sheep with a right thoracotomy (n = 20), even before CPB commenced MPO2 fell from (mean ± SEM) 52 ± 4 to 41 ±5 mmHg simultaneously with reduced arterial pressure (from 108 ± 5 to 88 ± 5 mmHg). In dorsally recumbent sheep with a medial sternotomy (n = 4), MPO2 was even more severely reduced (to 12 ± 12 mmHg) before CPB. In laterally recumbent sheep in which a crystalloid prime was used (n = 7), after commencing CPB, MPO2 fell abruptly to 24 ±6 mmHg within 20-30 minutes. MPO2 during CPB was not improved by adding donor blood to the prime (n = 13). In patients undergoing cardiac surgery, UPO2 fell by 4 ± 1 mmHg and mean arterial pressure fell by 7 ± 1 mmHg during the 30 minutes before CPB. UPO2 then fell by a further 12 ± 2 mmHg during the first 30 minutes of CPB but remained relatively stable for the remaining 24 minutes of observation. CONCLUSIONS: Renal medullary hypoxia is an early event during CPB. It starts to develop even before CPB, presumably due to a pressure-dependent decrease in renal blood flow. Medullary hypoxia during CPB appears to be promoted by hypotension and is not ameliorated by increasing blood hemoglobin concentration.

Intra-operative and early post-operative prediction of cardiac surgery-associated acute kidney injury: Urinary oxygen tension compared with plasma and urinary biomarkers.

Acute kidney injury (AKI) is a common and serious post-operative complication of cardiac surgery. The value of a predictive biomarker is determined not only by its predictive efficacy, but also by how early this prediction can be made. For a biomarker of cardiac surgery-associated AKI, this is ideally during the intra-operative period. Therefore, in 82 adult patients undergoing cardiac surgery requiring cardiopulmonary bypass (CPB), we prospectively compared the predictive efficacy of various blood and urinary biomarkers with that of continuous measurement of urinary oxygen tension (UPO2 ) at pre-determined intra- and post-operative time-points. None of the blood or urine biomarkers we studied showed predictive efficacy for post-operative AKI when measured intra-operatively. When treated as a binary variable (≤ or > median for the whole cohort), the earliest excess risk of AKI was predicted by an increase in urinary neutrophil gelatinase-associated lipocalin (NGAL) at 3 h after entry into the intensive care unit (odds ratio [95% confidence limits], 2.86 [1.14-7.21], p = 0.03). Corresponding time-points were 6 h for serum creatinine (3.59 [1.40-9.20], p = 0.008), and 24 h for plasma NGAL (4.54 [1.73-11.90], p = 0.002) and serum cystatin C (6.38 [2.35-17.27], p = 0.001). In contrast, indices of intra-operative urinary hypoxia predicted AKI after weaning from CPB, and in the case of a fall in UPO2 to ≤10 mmHg, during the rewarming phase of CPB (3.00 [1.19-7.56], p = 0.02). We conclude that continuous measurement of UPO2 predicts AKI earlier than plasma or urinary NGAL, serum cystatin C, or early post-operative changes in serum creatinine.

Adverse Events in the Digital Age: Finding the Sharpest Tool in the Box.

Digital health arrived to society and it generates data which is growing exponentially. Pharma and medical device industry recently embarked on this journey. Digital tools became everyday experience both in development as well as post-market settings. These non-traditional data sources may contain relevant safety, efficacy, effectiveness and other knowledge, which are valuable for understanding and further characterizing safety profile of a given medicinal and medical device product. It is however unclear what new responsibilities are associated with the use of such tools and data generated / collected by them. Current regulatory framework does not provide very clear guidance on it. Teams are struggling to interpret expectations originated from regulations, ethics and patients. We present practical approaches for data management and we suggest that a detailed assessment of projects is conducted to identify obligations for screening digitally-sourced data. Compliance with regulations is obligatory, but it is also incumbent on the sponsor to define a data management strategy covering the purpose of the activities, and the value of the data gathered. Decisions thereafter should be aligned with the mission, vision, and objectives defined by the sponsor.

Renal and Cerebral Hypoxia and Inflammation During Cardiopulmonary Bypass.

Cardiac surgery-associated acute kidney injury and brain injury remain common despite ongoing efforts to improve both the equipment and procedures deployed during cardiopulmonary bypass (CPB). The pathophysiology of injury of the kidney and brain during CPB is not completely understood. Nevertheless, renal (particularly in the medulla) and cerebral hypoxia and inflammation likely play critical roles. Multiple practical factors, including depth and mode of anesthesia, hemodilution, pump flow, and arterial pressure can influence oxygenation of the brain and kidney during CPB. Critically, these factors may have differential effects on these two vital organs. Systemic inflammatory pathways are activated during CPB through activation of the complement system, coagulation pathways, leukocytes, and the release of inflammatory cytokines. Local inflammation in the brain and kidney may be aggravated by ischemia (and thus hypoxia) and reperfusion (and thus oxidative stress) and activation of resident and infiltrating inflammatory cells. Various strategies, including manipulating perfusion conditions and administration of pharmacotherapies, could potentially be deployed to avoid or attenuate hypoxia and inflammation during CPB. Regarding manipulating perfusion conditions, based on experimental and clinical data, increasing standard pump flow and arterial pressure during CPB appears to offer the best hope to avoid hypoxia and injury, at least in the kidney. Pharmacological approaches, including use of anti-inflammatory agents such as dexmedetomidine and erythropoietin, have shown promise in preclinical models but have not been adequately tested in human trials. However, evidence for beneficial effects of corticosteroids on renal and neurological outcomes is lacking. © 2021 American Physiological Society. Compr Physiol 11:1-36, 2021.

Reversal of renal tissue hypoxia during experimental cardiopulmonary bypass in sheep by increased pump flow and arterial pressure.

AIM: Renal tissue hypoxia during cardiopulmonary bypass could contribute to the pathophysiology of acute kidney injury. We tested whether renal tissue hypoxia can be alleviated during cardiopulmonary bypass by the combined increase in target pump flow and mean arterial pressure. METHODS: Cardiopulmonary bypass was established in eight instrumented sheep under isoflurane anaesthesia, at a target continuous pump flow of 80 mL·kg-1 min-1 and mean arterial pressure of 65 mmHg. We then tested the effects of simultaneously increasing target pump flow to 104 mL·kg-1 min-1 and mean arterial pressure to 80 mmHg with metaraminol (total dose 0.25-3.75 mg). We also tested the effects of transitioning from continuous flow to partially pulsatile flow (pulse pressure ~15 mmHg). RESULTS: Compared with conscious sheep, at the lower target pump flow and mean arterial pressure, cardiopulmonary bypass was accompanied by reduced renal blood flow (6.8 ± 1.2 to 1.95 ± 0.76 mL·min-1 kg-1) and renal oxygen delivery (0.91 ± 0.18 to 0.24 ± 0.11 mL·O2 min-1 kg-1). There were profound reductions in cortical oxygen tension (PO2) (33 ± 13 to 6 ± 6 mmHg) and medullary PO2 (31 ± 12 to 8 ± 8 mmHg). Increasing target pump flow and mean arterial pressure increased renal blood flow (to 2.6 ± 1.0 mL·min-1 kg-1) and renal oxygen delivery (to 0.32 ± 0.13 mL·O2 min-1kg-1) and returned cortical PO2 to 58 ± 60 mmHg and medullary PO2 to 28 ± 16 mmHg; levels similar to those of conscious sheep. Partially pulsatile pump flow had no significant effects on renal perfusion or oxygenation. CONCLUSIONS: Renal hypoxia during experimental CPB can be corrected by increasing target pump flow and mean arterial pressure within a clinically feasible range.

Intraoperative renal hypoxia and risk of cardiac surgery-associated acute kidney injury.

BACKGROUND: Acute kidney injury (AKI) is common after cardiac surgery requiring cardiopulmonary bypass. Renal hypoxia may precede clinically detectable AKI. We compared the efficacy of two indices of renal hypoxia, (i) intraoperative urinary oxygen tension (UPO2 ) and (ii) the change in plasma erythropoietin (pEPO) during surgery, in predicting AKI. We also investigated whether the performance of these prognostic markers varies with preoperative patient characteristics. METHODS: In 82 patients undergoing on-pump cardiac surgery, blood samples were taken upon induction of anesthesia and upon entry into the intensive care unit. UPO2 was continuously measured throughout surgery. RESULTS: Thirty-two (39%) patients developed postoperative AKI. pEPO increased during surgery, but this increase did not predict AKI, regardless of risk of postoperative mortality assessed by EuroSCORE-II. For patients categorized at higher risk by EuroSCORE-II >1.98 (median score for the cohort), UPO2 ≤10 mmHg at any time during surgery predicted a 4.04-fold excess risk of AKI (p = .04). However, UPO2 did not significantly predict AKI in lower-risk patients. UPO2 significantly predicted AKI in patients who were older, had previous myocardial infarction, diabetes, lower preoperative serum creatinine, or shorter bypass times. pEPO and UPO2 were only weakly correlated. CONCLUSIONS: Intraoperative change in pEPO does not predict AKI. However, UPO2 shows promise, particularly in patients with higher risk of operative mortality. The disparity between these two markers of renal hypoxia may indicate that UPO2 reflects medullary oxygenation whereas pEPO reflects cortical oxygenation.

Influence of blood haemoglobin concentration on renal haemodynamics and oxygenation during experimental cardiopulmonary bypass in sheep.

AIM: Blood transfusion may improve renal oxygenation during cardiopulmonary bypass (CPB). In an ovine model of experimental CPB, we tested whether increasing blood haemoglobin concentration [Hb] from ~7 g dL-1 to ~9 g dL-1 improves renal tissue oxygenation. METHODS: Ten sheep were studied while conscious, under stable isoflurane anaesthesia, and during 3 hours of CPB. In a randomized cross-over design, 5 sheep commenced bypass at a high target [Hb], achieved by adding 600 mL donor blood to the priming solution. After 90 minutes of CPB, PlasmaLyte® was added to the blood reservoir to achieve low target [Hb]. For the other 5 sheep, no blood was added to the prime, but after 90 minutes of CPB, 800-900 mL of donor blood was given to achieve a high target [Hb]. RESULTS: Overall, CPB was associated with marked reductions in renal oxygen delivery (-50 ± 12%, mean ± 95% confidence interval) and medullary tissue oxygen tension (PO2 , -54 ± 29%). Renal fractional oxygen extraction was 17 ± 10% less during CPB at high [Hb] than low [Hb] (P = .04). Nevertheless, no increase in tissue PO2 in either the renal medulla (0 ± 6 mmHg change, P > .99) or cortex (-19 ± 13 mmHg change, P = .08) was detected with high [Hb]. CONCLUSIONS: In experimental CPB blood transfusion to increase Hb concentration from ~7 g dL-1 to ~9 g dL-1 did not improve renal cortical or medullary tissue PO2 even though it decreased whole kidney oxygen extraction.

Machine Learning Algorithms for Predicting and Risk Profiling of Cardiac Surgery-Associated Acute Kidney Injury.

Using a large national database of cardiac surgical procedures, we applied machine learning (ML) to risk stratification and profiling for cardiac surgery-associated acute kidney injury. We compared performance of ML to established scoring tools. Four ML algorithms were used, including logistic regression (LR), gradient boosted machine (GBM), K-nearest neighbor, and neural networks (NN). These were compared to the Cleveland Clinic score, and a risk score developed on the same database. Five-fold cross-validation repeated 20 times was used to measure the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. Risk profiles from GBM and NN were generated using Shapley additive values. A total of 97,964 surgery events in 96,653 patients were included. For predicting postoperative renal replacement therapy using pre- and intraoperative data, LR, GBM, and NN achieved an AUC (standard deviation) of 0.84 (0.01), 0.85 (0.01), 0.84 (0.01) respectively outperforming the highest performing scoring tool with 0.81 (0.004). For predicting cardiac surgery-associated acute kidney injury, LR, GBM, and NN each achieved 0.77 (0.01), 0.78 (0.01), 0.77 (0.01) respectively outperforming the scoring tool with 0.75 (0.004). Compared to scores and LR, shapely additive values analysis of black box model predictions was able to generate patient-level explanations for each prediction. ML algorithms provide state-of-the-art approaches to risk stratification. Explanatory modeling can exploit complex decision boundaries to aid the clinician in understanding the risks specific to individual patients.

Renal functional reserve: from physiological phenomenon to clinical biomarker and beyond.

Glomerular filtration rate (GFR) is acutely increased following a high-protein meal or systemic infusion of amino acids. The mechanisms underlying this renal functional response remain to be fully elucidated. Nevertheless, they appear to culminate in preglomerular vasodilation. Inhibition of the tubuloglomerular feedback signal appears critical. However, nitric oxide, vasodilator prostaglandins, and glucagon also appear important. The increase in GFR during amino acid infusion reveals a "renal reserve," which can be utilized when the physiological demand for single nephron GFR increases. This has led to the concept that in subclinical renal disease, before basal GFR begins to reduce, renal functional reserve can be recruited in a manner that preserves renal function. The extension of this concept is that once a decline in basal GFR can be detected, renal disease is already well progressed. This concept likely applies both in the contexts of chronic kidney disease and acute kidney injury. Critically, its corollary is that deficits in renal functional reserve have the potential to provide early detection of renal dysfunction before basal GFR is reduced. There is growing evidence that the renal response to infusion of amino acids can be used to identify patients at risk of developing either chronic kidney disease or acute kidney injury and as a treatment target for acute kidney injury. However, large multicenter clinical trials are required to test these propositions. A renewed effort to understand the renal physiology underlying the response to amino acid infusion is also warranted.