Impact of the Covid-19 pandemic on the performance of machine learning algorithms for predicting perioperative mortality.

BACKGROUND: Machine-learning models are susceptible to external influences which can result in performance deterioration. The aim of our study was to elucidate the impact of a sudden shift in covariates, like the one caused by the Covid-19 pandemic, on model performance. METHODS: After ethical approval and registration in Clinical Trials (NCT04092933, initial release 17/09/2019), we developed different models for the prediction of perioperative mortality based on preoperative data: one for the pre-pandemic data period until March 2020, one including data before the pandemic and from the first wave until May 2020, and one that covers the complete period before and during the pandemic until October 2021. We applied XGBoost as well as a Deep Learning neural network (DL). Performance metrics of each model during the different pandemic phases were determined, and XGBoost models were analysed for changes in feature importance. RESULTS: XGBoost and DL provided similar performance on the pre-pandemic data with respect to area under receiver operating characteristic (AUROC, 0.951 vs. 0.942) and area under precision-recall curve (AUPR, 0.144 vs. 0.187). Validation in patient cohorts of the different pandemic waves showed high fluctuations in performance from both AUROC and AUPR for DL, whereas the XGBoost models seemed more stable. Change in variable frequencies with onset of the pandemic were visible in age, ASA score, and the higher proportion of emergency operations, among others. Age consistently showed the highest information gain. Models based on pre-pandemic data performed worse during the first pandemic wave (AUROC 0.914 for XGBoost and DL) whereas models augmented with data from the first wave lacked performance after the first wave (AUROC 0.907 for XGBoost and 0.747 for DL). The deterioration was also visible in AUPR, which worsened by over 50% in both XGBoost and DL in the first phase after re-training. CONCLUSIONS: A sudden shift in data impacts model performance. Re-training the model with updated data may cause degradation in predictive accuracy if the changes are only transient. Too early re-training should therefore be avoided, and close model surveillance is necessary.

Sex-specific issues of central and peripheral arginine-vasopressin concentrations in neurocritical care patients.

BACKGROUND: Arginine-Vasopressin (AVP) is a nonapeptide that exerts multiple functions within the central nervous system and in the blood circulation that might contribute to outcome in critically ill patients. Sex differences have been found for mental and physical effects of AVP. For example, stress response and response due to hemorrhage differ between males and females, at least in animal studies. Data on humans -especially on AVP within the central nervous system (CNS)-are scarce, as cerebrospinal fluid (CSF) which is said to represent central AVP activity, has to be collected by means of invasive procedures. Here we present data on 30 neurocritical care patients where we simultaneously collected blood, CSF and saliva to analyze concentrations in the central and peripheral compartments. PATIENTS AND METHODS: 30 neurocritical care patients were included (13 male, 13 postmenopausal female, 4 premenopausal female) with a median age of 60 years. CSF, plasma and saliva were obtained simultaneously once in each patient and analyzed for AVP concentrations. Correlations between the central compartment represented by CSF, and the peripheral compartment represented by plasma and saliva, were identified. Relations between AVP concentrations and serum sodium and hematocrit were also determined. RESULTS: In the whole patient collective, only very weak to weak correlations could be detected between AVP plasma/CSF, plasma/saliva and CSF/saliva as well as between AVP concentrations in each of the compartments and serum sodium/hematocrit. Regarding the subgroup of postmenopausal females, a significant moderate correlation could be detected for AVP in plasma and CSF and AVP CSF and serum sodium. CONCLUSION: Absolute concentrations of AVP in central and peripheral compartments did not show sex differences. However, correlations between AVP plasma and CSF and AVP CSF and serum sodium in postmenopausal females indicate differences in AVP secretion and AVP response to triggers that deserve further examination.

The influence of PEEP and tidal volume on central blood volume.

BACKGROUND AND OBJECTIVE: Measurement of central blood volumes (CBV), such as global end-diastolic volume (GEDV) and right ventricular end-diastolic volume (RVEDV) are considered appropriate estimates of intravascular volume status. However, to apply those parameters for preload assessment in mechanically ventilated patients, the influence of tidal volume (TV) and positive endexpiratory airway pressure (PEEP) on those parameters must be known. METHODS: In 13 mechanically ventilated piglets, the effect of low (10 mL kg(-1)) and high (20 mL kg(-1)) TVs on CBV was investigated in absence and presence of PEEP (0 and 15 cm H(2)O). GEDV, RVEDV, right heart (RHEDV) and left heart end-diastolic volume (LHEDV) were measured by thermodilution. Blood flow on the descending thoracic aorta measured with an ultrasonic flow-probe served to determine stroke volume (SV). Measurements were performed during baseline conditions, after volume loading with previously extracted haemodilution blood (20 mL kg(-1)) and following haemorrhage (30 mL kg(-1)). RESULTS: Application of PEEP decreased GEDV and SV significantly (P < 0.05). Augmenting TV did not reduce GEDV systematically, but significantly reduced SV (P < 0.05). Changes in ventilator settings only influenced RVEDV following volume loading (P < 0.05). RHEDV and LHEDV decreased following application of PEEP, but only RHEDV decreased after augmenting TV at baseline and following volume loading. Correlation of SV with parameters of CBV was r = 0.487 (P < 0.01) for GEDV, r = 0.553 (P < 0.01) for RVEDV, r = 0.596 (P < 0.01) for RHEDV and r = 0.303 (P < 0.01) for LHEDV. CONCLUSION: Application of PEEP decreases CBV and SV. Augmenting TV reduces SV but not CBV. There is a moderate correlation between parameters of CBV and cardiac performance.

[Acute normovolemic hemodilutin (ANH). Effects of ANH on the diastolic function of the left ventricle]. / Akute normovolämische Hämodilution (ANH). Effekte der ANH auf die diastolische Funktion des linken Ventrikels.

Ischemia-induced changes of diastolic leftventricular (LV) properties commonly precede corresponding ECG-changes. In the present experimental study the consequences of acute normovolemic hemodilution (ANH) induced dilutional anemia (hematocrit, hct 20%) for LV diastolic function were investigated. A total of 22 anaesthetized, splenectomized beagle dogs breathing room air were hemodiluted with isooncotic hydroxyethylstarch solution (6% HAES 200,000/0.5) until a hct value of 20% was reached. Before and after ANH intravascular blood volume (indocyaningreen dilution technique), global and regional myocardial blood flow (radioactive microspheres technique) and the following parameters reflecting LV diastolic properties were ascertained: 1) the maximum rate of LV pressure decrease (LVdp/dtmin), 2) slope and intercept of the enddiastolic pressure-volume relationship (EDPVR, conductance technique) and 3) the time-constant of isovolumic LV pressure decline "tau". After ANH to hct 20% diastolic LV function was found unchanged. Particularly the load-independent parameters (EDPVR-slope and tau) remained constant. The decrease of LV dp/dtmin (-2724 +/- 479 vs. -2388 +/- 408 mmHg.sek-1; p < 0.05) reflects ANH induced changes of LV pre- and afterload. Signs of subendocardial perfusion mismatch were not encountered. Presumed that the coronary vascular system is intact ANH to hct 20% does not provoque changes of LV diastolic function. Moreover neither myocardial perfusion and oxygenation nor myocardial function are endangered by this degree of dilutional anemia.

Effect of acute normovolemic hemodilution on distribution of blood flow and tissue oxygenation in dog skeletal muscle.

Acute normovolemic hemodilution (ANH) is efficient in reducing allogenic blood transfusion needs during elective surgery. Tissue oxygenation is maintained by increased cardiac output and oxygen extraction and, presumably, a more homogeneous tissue perfusion. The aim of this study was to investigate blood flow distribution and oxygenation of skeletal muscle. ANH from hematocrit of 36 +/- 3 to 20 +/- 1% was performed in 22 splenectomized, anesthetized beagles (17 analyzed) ventilated with room air. Normovolemia was confirmed by measurement of blood volume. Distribution of perfusion within skeletal muscle was determined by using radioactive microspheres. Tissue oxygen partial pressure was assessed with a polarographic platinum surface electrode. Cardiac index (3.69 +/- 0.79 vs. 4.79 +/- 0.73 l. min-1. m-2) and muscle perfusion (4.07 +/- 0.44 vs. 5.18 +/- 0.36 ml. 100 g-1. min-1) were increased at hematocrit of 20%. Oxygen delivery to skeletal muscle was reduced to 74% of baseline values (0.64 +/- 0.06 vs. 0.48 +/- 0.03 ml O2. 100 g-1. min-1). Nevertheless, tissue PO2 was preserved (27.4 +/- 1.3 vs. 29.9 +/- 1. 4 Torr). Heterogeneity of muscle perfusion (relative dispersion) was reduced after ANH (20.0 +/- 2.2 vs. 13.9 +/- 1.5%). We conclude that a more homogeneous distribution of perfusion is one mechanism for the preservation of tissue oxygenation after moderate ANH, despite reduced oxygen delivery.

IV perflubron emulsion versus autologous transfusion in severe normovolemic anemia: effects on left ventricular perfusion and function.

Intact cardiac compensatory mechanisms are necessary to maintain adequate tissue oxygenation during acute normovolemic hemodilution (ANH). Left ventricular (LV) perfusion, oxygenation and function were analyzed in an experimental whole-body model of profound ANH (Hct 9%) and effectiveness of a perfluorocarbon-based oxygen carrier in maintaining myocardial oxygenation and function was evaluated. A total of 22 anesthetized dogs were hemodiluted to Hct 20% followed by a simulated, controlled blood-loss phase in which dogs were randomized to either: (1) 1:1 exchange of lost blood with autologous red blood cells (RBC-group), (2) 1:1 exchange with a colloid (control-group) and (3) 1:1 exchange with a colloid after a single dose of 1.8 g/kg BW perflubron i.v. (PFC-group). Myocardial oxygen delivery and consumption as well as endocardial perfusion were determined using radioactive microspheres. LV myocardial contractility (LV MC) was assessed from: (1) the relationship between maximum rate of LV pressure increase (LVdp/dtmax) and LV enddiastolic volume (LVEDV) and (2) analysis of the LV endsystolic pressure volume relationship (ESPVR). LV diastolic properties were reflected by (1) minimum rate of LV pressure increase (LVdp/dtmin), (2) slope and intercept of the enddiastolic pressure-volume relationship (EDPVR) and (3) the time-constant of isovolumic LV pressure decline "tau 1/2". Full sets of LV MC data were obtained from 18 dogs (n = 6 per group). LV MC (LVdp/dtmax-LVEDV relation) increased after perflubron administration. At the lowest Hct level, all parameters reflecting LV MC as well as LVdp/dtmin were significantly higher in the PFC-group than in the control-group. After profound normovolemic hemodilution (Hct 9%) superiority of LV MC and LV diastolic properties was found, when myocardial oxygenation was supported by i.v. perflubron emulsion, a temporary O2 carrier.

Normovolaemic haemodilution and hyperoxia have no effect on fractal dimension of regional myocardial perfusion in dogs.

Hypervolaemic haemodilution makes myocardial perfusion more homogenous as reflected by reduced fractal dimension of regional myocardial perfusion. The clinically more commonly performed acute normovolaemic haemodilution, however, has not yet been studied in this respect. Hyperoxic ventilation with 100% oxygen is used in conjunction with haemodilution to compensate for low oxygen content by increasing physically dissolved oxygen in plasma. Since hyperoxia is known to cause disturbance in microcirculatory regulation we studied the effects of acute normovolaemic haemodilution to haematocrit (hct) 20 +/- 1% and hyperoxia on regional myocardial perfusion heterogeneity in 22 anaesthetized dogs using fractal and correlation analysis. Regional myocardial perfusion was assessed with radioactive microspheres. The results of the study were that heart rate, blood volume and arterial pressure were unchanged during haemodilution. Cardiac index was 3.6 +/- 0.7 L min-1 m-2 before and 4.6 +/- 0.7 L min-1 m-2 after haemodilution (P < 0.05). Fractal dimension (D) of regional myocardial perfusion was 1.17 +/- 0.10 at baseline. Neither haemodilution (D = 1.19 +/- 0.10) nor hyperoxia (D = 1.17 +/- 0.10) altered fractal properties of regional myocardial perfusion. Spatial correlation of blood flow to adjacent tissue samples before haemodilution was 0.58 +/- 0.15. Haemodilution and hyperoxia did not significantly influence spatial correlation (0.57 +/- 0.12 vs. 0.60 +/- 0.09; ns). We conclude that neither acute normovolaemic haemodilution nor haemodilution in combination with hyperoxic ventilation alter physiological myocardial perfusion heterogeneity.

Effects of hyperoxic ventilation on hemodilution-induced changes in anesthetized dogs.

BACKGROUND: In subjects who have undergone acute preoperative normovolemic hemodilution (ANH), intraoperative hemorrhage is generally treated by immediate return of autologous blood collected during ANH. Simply increasing blood oxygen content by hyperoxic ventilation (HV, inspiratory fraction [FIO2] 1.0) might compensate for the acute anemia, allow further ANH, and delay onset of autologous blood return. STUDY DESIGN AND METHODS: This study 1) evaluated the effects of HV (FIO2 1.0) upon ANH to a hemoglobin (Hb) concentration of 7 g per dL in anesthetized dogs ventilated with room air and 2) compared the effects of subsequent profound ANH (Hb, 3 g/dL) with and without an intravenous perfluorocarbon emulsion (perflubron 60% wt/vol) versus those of autologous red cell transfusion. The results of the entire study are presented in two parts. Organ tissue oxygenation was assessed in skeletal muscle and liver, and systemic oxygenation status was evaluated. Myocardial contractility was deduced from left ventricular pressure-volume relationship. Seven of 22 dogs underwent further hemodilution while breathing 100-percent O2, for a determination of the Hb concentration at which HV-induced effects were abolished. RESULTS: HV completely reversed the ANH-induced increase in cardiac index (4.6 +/- 0.7 vs. 3.8 +/- 0.9 L/min/m2 before and during HV; p < 0.05) and partially reversed the decrease in systemic vascular resistance (1784 +/- 329 vs. 2087 +/- 524 dyn x cm-5 x sec x m-2; p < 0.05). Despite unchanged global O2 delivery, organ tissue oxygenation improved during HV (mixed venous partial pressure of O2: 40 +/- 3 vs. 59 +/- 7 torr; coronary venous pressure of O2: 30 +/- 4 vs. 43 +/- 6 torr; p < 0.05; liver surface: 31 +/- 11 vs. 39 +/- 13 torr; skeletal muscle surface: 30 +/- 14 vs. 41 +/- 22 torr; p < 0.05). This improvement was due to an increased contribution of physically dissolved O2 in plasma to O2 delivery (3.2 +/- 0.2% before HV vs. 14.6 +/- 1% during HV; p < 0.05) and O2 consumption (whole body: 6 +/- 1% vs. 47 +/- 8%, p < 0.05; myocardium: 4.3 +/- 0.9% vs. 31 +/- 6%, p < 0.05). The beneficial effects of HV were lost after an additional volume-compensated exchange of 19 percent of blood volume (Hb, 5.6 g/dL). CONCLUSION: In anesthetized dogs ventilated with room air and hemodiluted to a Hb of 7 g per dL, simple oxygen therapy by HV (FIO2 1.0) rapidly improves tissue oxygenation and permits extended hemodilution to Hb of 5.8 g per dL until the HV-induced effects are lost.

Hemodilution and intravenous perflubron emulsion as an alternative to blood transfusion: effects on tissue oxygenation during profound hemodilution in anesthetized dogs.

BACKGROUND: Intravenously administered perfluorocarbon (PFC) emulsions increase oxygen solubility in plasma. PFC might therefore temporarily replace red cells (RBCs) lost during intraoperative hemorrhage. In patients who have undergone hemodilution, the return of autologous blood may be delayed by the administration of PFC, and autologous RBCs may be saved for transfusion after surgical bleeding is stopped and PFC is cleared by the reticuloendothelial system. STUDY DESIGN AND METHODS: In 22 anesthetized, hemodiluted dogs (hemoglobin [Hb] 7 g/dL) breathing 100-percent O2, an intraoperative volume-compensated blood loss was simulated. The efficacy of three therapeutic regimens in maintaining tissue oxygenation was compared: 1) RBC group (n = 7): maintenance of a Hb > 7 g per dL by transfusion of autologous RBCs; 2) PFC group (n = 7): bolus application of a second-generation PFC emulsion (60% wt/vol perflubron) and further acute normovolemic hemodilution (ANH) to a Hb of 3 g per dL; and 3) control group (n = 7): further ANH alone to a Hb of 3 g per dL. Systemic and myocardial oxygenation status and tissue oxygenation were assessed. RESULTS: Autologous RBCs transfused to maintain a Hb of 7 g per dL preserved hemodynamics and tissue oxygenation during blood loss. In the PFC and control groups, heart rate and cardiac index increased significantly in response to further ANH. Tissue oxygenation was not different in the PFC and the RBC groups. Direct comparison of the PFC and control groups revealed better tissue oxygenation in the PFC group, as reflected by significantly higher mixed venous, coronary venous, and local tissue pO2 on liver and skeletal muscle. CONCLUSION: Bolus intravenous administration of 60-percent (wt/vol) perflubron emulsion and further hemodilution from a Hb of 7 g per dL to one of 3 g per dL were as effective as autologous RBC transfusion in maintaining tissue oxygenation during volume-compensated blood loss designed to mimic surgical bleeding.

Hemodilution and hyperoxia locally change distribution of regional pulmonary perfusion in dogs.

In seven anesthetized dogs, the effects of acute normovolemic hemodilution (ANH) to a hematocrit of 20 and 8% and the effects of hyperoxic ventilation (100% oxygen) on distribution of regional pulmonary blood flow (rPBF; radioactive microspheres) were investigated. Normovolemia was monitored with blood volume measurements (indocyanine green dilution kinetics). Before ANH, fractal dimension (D) of rPBF in the whole lung was 1.19 +/- 0.09 (mean +/- SD). Spatial correlation (rho) of rPBF in the whole lung was 0.6 +/- 0.08. D is a resolution-independent measure for global rPBF distribution, and rho is the averaged flow relationship of directly neighboring lung samples. With regard to the entire lung, neither ANH nor hyperoxia changed D or rho. With regard to horizontal, isogravitational planes, ANH induced opposite changes of rPBF heterogeneity depending on the vertical location of the plane and the parameter used. In ventral planes, a change in relative dispersion (SD/mean) indicated decreased homogeneity. However, rho suggested more homogeneous perfusion. Hyperoxia restored baseline rPBF distribution. Our data suggest that ANH causes different alterations of heterogeneity of rPBF depending on location within the lung.

Effects of hemodilution on splanchnic perfusion and hepatorenal function. I. Splanchnic perfusion.

Perfusion of intestinal organs increases in response to acute normovolemic hemodilution (ANH). However, detailed studies on distribution of regional splanchnic organ perfusion during ANH are lacking. We therefore carried out this study to test the hypothesis that ANH does not cause disturbance of physiologic patterns of regional splanchnic organ blood flow. After governmental permission, 22 anesthetized dogs were instrumented to allow invasive hemodynamic measurements and intracardial injection of radioactive microspheres (diameter 15 micro m) for determination of regional organ perfusion. Measurements were made at baseline (hematocrit 37 +/- 3%) and after ANH with 6% hydroxyethyl starch (mol. wt. 200000 / 0.5) to hct 20 +/- 1%. After completion of the protocol, splanchnic organs were removed and dissected into small samples according to anatomical and functional principles. Regional perfusion was determined based on the microsphere content of each sample. Hepatic, intestinal, and pancreatic blood flow increased with ANH. Hepatic arterial blood flow rose by 86%, whereas portal venous perfusion increased by 28%. Small intestine mucosal perfusion was augmented by 68% while the non-mucosal tissue compartment of the gut wall received 32% more blood flow after ANH which is in proportion to the increase in cardiac index after ANH. This redistribution of intestinal flow might be the basis for the preservation of tissue oxygenation during moderate isovolemic anemia.

Effects of hemodilution on splanchnic perfusion and hepatorenal function. II. Renal perfusion and hepatorenal function.

Hepatorenal perfusion and function were assxssed in 22 dogs undergoing acute normovolemic hemodilution (ANH) to a hematocrit (Hct) of 20% using 6% hydroxyethyl starch (200.000/0.5) as the diluent. Organ perfusion was determined with the radioactive microspheres method. Renal function was assessed by urinary output, creatinine clearance and fractional sodium excretion. Blood volume as well as hepatic function were derived from indocyanine green (ICG) dilution kinetics. Hepatocellular integrity was determined by serum enzymatic activity of glutamate-oxalacetate-transaminase (GOT) and glutamate-pyruvate- transaminase (GPT). ANH to Hct 20% did not change blood volume and mean aortic pressure, while heart rate was slightly elevated (p<0.05) by 5 beats per minute and cardiac output increased by 29% (p<0.05). In contrast to the liver, where arterial and portal venous blood flow increased (86% and 28%, respectively; p<0.05), total renal blood flow as well as intraorgan distribution of renal blood flow remained unchanged post-ANH. While creatinine clearance remained unchanged following ANH, urinary output and fractional urinary excretion increased (p<0.05). In response to enhanced hepatic blood flow after ANH, intravascular half-life of ICG was reduced (p<0.05) and ICG clearance increased (p<0.05). Serum enzymatic activity of GPT decreased upon ANH (p<0.05), while GOT activity remained unchanged. ANH to a Hct 20% does not impair hepatorenal function. Increased urinary output points out the necessity for proper adjustment of crystalloid infusion to maintain normal intravascular volume and avoid hypovolemia and the associated risk of tissue hypoxia.

Effects of hemodilution on gastric regional perfusion and intramucosal pH.

UNLABELLED: Acute normovolemic hemodilution (ANH) has been shown to be a cost-effective method of reducing allogenic blood transfusion during elective surgery. ANH has been implicated with impaired oxygenation in isolated canine gastric flaps. The present study was designed to investigate the effects of ANH on gastric mucosal oxygenation using a model closely imitating the clinical situation. Sixteen splenectomized anesthetized beagles were isovolemically hemodiluted to a hematocrit of 20 +/- 1% (6% hydroxyethyl starch; mol wt 200,000; substitution ratio 0.45-0.55). Blood volume (indocyanine green), cardiorespiratory parameters, gastric intramucosal pH (pHi), and gastric regional blood flow (radioactive microspheres; 15 microns) were measured before and after ANH. RESULTS: blood volume was unchanged (87 +/- 8 ml/kg before and 88 +/- 7 ml/kg after ANH). Median total gastric mucosal blood flow at baseline was 0.51 +/- 0.35 ml.min-1.g-1 and did not change significantly upon ANH. The mean pHi was 7.29 +/- 0.05 before and 7.28 +/- 0.05 after hemodilution. There was a homogenization of blood flow distribution in gastric mucosa. Severe hemodilution to a hematocrit of 20% does not impair gastric mucosal oxygenation and poses no risk to gastric mucosal integrity.

The effect of acute normovolemic hemodilution (ANH) on myocardial contractility in anesthetized dogs.

The influence of severe acute normovolemic hemodilution (ANH) on myocardial contractility (MC) was investigated in 14 splenectomized, anesthetized dogs. MC was assessed by the maximum rate of left ventricular pressure increase (LVdp/dt(max)), end-systolic elastance (Ees), and preload recruitable stroke work (PRSW) (conductance catheter, left ventricular pressure-volume relationship). Measurements of myocardial perfusion and oxygenation (radioactive microsphere technique) assured comparability of the model to previously performed studies. Global and regional myocardial blood flow increased significantly upon hemodilution with preference to midmyocardium and subendocardium. This resulted in preservation of both myocardial oxygen delivery and consumption after ANH. Myocardial oxygen extraction as well as coronary venous Po2 were unaffected by ANH, while coronary venous lactate concentration decreased, indicating that myocardial oxygen need was met. LVdp/dt(max) decreased significantly after hemodilution (2278 +/- 577 vs 1884 +/- 381 mm Hg/s, P < 0.01), whereas Ees and PRSW increased significantly (1.76 +/- 0.54 vs 2.15 +/- 0.75 mm Hg/mL, P < 0.05, for Ees and 33 +/- 14 vs 45 +/- 14 mm Hg.mL, P < 0.05, for PRSW). While the decrease of LVdp/dt(max) most likely reflects ANH-induced changes of ventricular pre- and afterload, the increase of Ees and PRSW indicates a true increase of myocardial contractility during ANH in anesthetized dogs.