Development and external validation of temporal fusion transformer models for continuous intraoperative blood pressure forecasting.

Background: During surgery, intraoperative hypotension is associated with postoperative morbidity and should therefore be avoided. Predicting the occurrence of hypotension in advance may allow timely interventions to prevent hypotension. Previous prediction models mostly use high-resolution waveform data, which is often not available. Methods: We utilised a novel temporal fusion transformer (TFT) algorithm to predict intraoperative blood pressure trajectories 7 min in advance. We trained the model with low-resolution data (sampled every 15 s) from 73,009 patients who were undergoing general anaesthesia for non-cardiothoracic surgery between January 1, 2017, and December 30, 2020, at the General Hospital of Vienna, Austria. The data set contained information on patient demographics, vital signs, medication, and ventilation. The model was evaluated using an internal (n = 8113) and external test set (n = 5065) obtained from the openly accessible Vital Signs Database. Findings: In the internal test set, the mean absolute error for predicting mean arterial blood pressure was 0.376 standard deviations-or 4 mmHg-and 0.622 standard deviations-or 7 mmHg-in the external test set. We also adapted the TFT model to binarily predict the occurrence of hypotension as defined by mean arterial blood pressure < 65 mmHg in the next one, three, five, and 7 min. Here, model discrimination was excellent, with a mean area under the receiver operating characteristic curve (AUROC) of 0.933 in the internal test set and 0.919 in the external test set. Interpretation: Our TFT model is capable of accurately forecasting intraoperative arterial blood pressure using only low-resolution data showing a low prediction error. When used for binary prediction of hypotension, we obtained excellent performance. Funding: No external funding.

A Comparison of Five Algorithmic Methods and Machine Learning Pattern Recognition for Artifact Detection in Electronic Records of Five Different Vital Signs: A Retrospective Analysis.

BACKGROUND: Research on electronic health record physiologic data is common, invariably including artifacts. Traditionally, these artifacts have been handled using simple filter techniques. The authors hypothesized that different artifact detection algorithms, including machine learning, may be necessary to provide optimal performance for various vital signs and clinical contexts. METHODS: In a retrospective single-center study, intraoperative operating room and intensive care unit (ICU) electronic health record datasets including heart rate, oxygen saturation, blood pressure, temperature, and capnometry were included. All records were screened for artifacts by at least two human experts. Classical artifact detection methods (cutoff, multiples of SD [z-value], interquartile range, and local outlier factor) and a supervised learning model implementing long short-term memory neural networks were tested for each vital sign against the human expert reference dataset. For each artifact detection algorithm, sensitivity and specificity were calculated. RESULTS: A total of 106 (53 operating room and 53 ICU) patients were randomly selected, resulting in 392,808 data points. Human experts annotated 5,167 (1.3%) data points as artifacts. The artifact detection algorithms demonstrated large variations in performance. The specificity was above 90% for all detection methods and all vital signs. The neural network showed significantly higher sensitivities than the classic methods for heart rate (ICU, 33.6%; 95% CI, 33.1 to 44.6), systolic invasive blood pressure (in both the operating room [62.2%; 95% CI, 57.5 to 71.9] and the ICU [60.7%; 95% CI, 57.3 to 71.8]), and temperature in the operating room (76.1%; 95% CI, 63.6 to 89.7). The CI for specificity overlapped for all methods. Generally, sensitivity was low, with only the z-value for oxygen saturation in the operating room reaching 88.9%. All other sensitivities were less than 80%. CONCLUSIONS: No single artifact detection method consistently performed well across different vital signs and clinical settings. Neural networks may be a promising artifact detection method for specific vital signs.

Concentration-effect relationship for tranexamic acid inhibition of tissue plasminogen activator-induced fibrinolysis in vitro using the viscoelastic ClotPro® TPA-test.

BACKGROUND: Tranexamic acid is an antifibrinolytic drug that is commonly administered for obstetric haemorrhage. Conventional viscoelastic tests are not sensitive to tranexamic acid, but the novel ClotPro® TPA-test can measure tranexamic acid-induced inhibition of fibrinolysis. We aimed to evaluate the TPA-test in pregnant and non-pregnant women. METHODS: We performed an in vitro study of whole blood samples spiked with tranexamic acid from pregnant women in the first, second, and third trimester (n=20 per group) and from non-pregnant women (n=20). We performed ClotPro TPA-tests of whole blood sample and ClotPro EX-tests, FIB-tests, and TPA-tests. RESULTS: Clot lysis was inhibited in a concentration-dependent manner up to a tranexamic acid concentration of 6.25 mg L-1. At tranexamic acid concentrations of 12.5 mg L-1 and above, clot lysis was completely inhibited. The concentration-effect relationship of tranexamic acid did not differ in a clinically important manner in blood from pregnant women across all three trimesters or from non-pregnant controls. A median maximum lysis cut-off value of at9 least 16% (25-75th percentiles 15-18), a median clot lysis time of 3600 s (25-75th percentiles 3600-3600), or both was associated with a tranexamic acid concentration of least 12.5 mg L-1. CONCLUSIONS: The ClotPro® TPA-test is sensitive in detecting inhibition of fibrinolysis by tranexamic acid in whole blood samples of pregnant and non-pregnant women. The concentration-effect relationship of tranexamic acid to inhibit fibrinolysis in whole blood did not differ for women in the first, second, and third trimester or for non-pregnant women.

Predicting Intraoperative Hypothermia Burden during Non-Cardiac Surgery: A Retrospective Study Comparing Regression to Six Machine Learning Algorithms.

BACKGROUND: Inadvertent intraoperative hypothermia is a common complication that affects patient comfort and morbidity. As the development of hypothermia is a complex phenomenon, predicting it using machine learning (ML) algorithms may be superior to logistic regression. METHODS: We performed a single-center retrospective study and assembled a feature set comprised of 71 variables. The primary outcome was hypothermia burden, defined as the area under the intraoperative temperature curve below 37 °C over time. We built seven prediction models (logistic regression, extreme gradient boosting (XGBoost), random forest (RF), multi-layer perceptron neural network (MLP), linear discriminant analysis (LDA), k-nearest neighbor (KNN), and Gaussian naïve Bayes (GNB)) to predict whether patients would not develop hypothermia or would develop mild, moderate, or severe hypothermia. For each model, we assessed discrimination (F1 score, area under the receiver operating curve, precision, recall) and calibration (calibration-in-the-large, calibration intercept, calibration slope). RESULTS: We included data from 87,116 anesthesia cases. Predicting the hypothermia burden group using logistic regression yielded a weighted F1 score of 0.397. Ranked from highest to lowest weighted F1 score, the ML algorithms performed as follows: XGBoost (0.44), RF (0.418), LDA (0.406), LDA (0.4), KNN (0.362), and GNB (0.32). CONCLUSIONS: ML is suitable for predicting intraoperative hypothermia and could be applied in clinical practice.

Development of a Reinforcement Learning Algorithm to Optimize Corticosteroid Therapy in Critically Ill Patients with Sepsis.

BACKGROUND: The optimal indication, dose, and timing of corticosteroids in sepsis is controversial. Here, we used reinforcement learning to derive the optimal steroid policy in septic patients based on data on 3051 ICU admissions from the AmsterdamUMCdb intensive care database. METHODS: We identified septic patients according to the 2016 consensus definition. An actor-critic RL algorithm using ICU mortality as a reward signal was developed to determine the optimal treatment policy from time-series data on 277 clinical parameters. We performed off-policy evaluation and testing in independent subsets to assess the algorithm's performance. RESULTS: Agreement between the RL agent's policy and the actual documented treatment reached 59%. Our RL agent's treatment policy was more restrictive compared to the actual clinician behavior: our algorithm suggested withholding corticosteroids in 62% of the patient states, versus 52% according to the physicians' policy. The 95% lower bound of the expected reward was higher for the RL agent than clinicians' historical decisions. ICU mortality after concordant action in the testing dataset was lower both when corticosteroids had been withheld and when corticosteroids had been prescribed by the virtual agent. The most relevant variables were vital parameters and laboratory values, such as blood pressure, heart rate, leucocyte count, and glycemia. CONCLUSIONS: Individualized use of corticosteroids in sepsis may result in a mortality benefit, but optimal treatment policy may be more restrictive than the routine clinical practice. Whilst external validation is needed, our study motivates a 'precision-medicine' approach to future prospective controlled trials and practice.

External validation of the ProVent score for prognostication of 1-year mortality of critically ill patients with prolonged mechanical ventilation: a single-centre, retrospective observational study in Austria.

OBJECTIVES: In critically ill patients requiring mechanical ventilation for at least 21 days, 1-year mortality can be estimated using the ProVent score, calculated from four variables (age, platelet count, vasopressor use and renal replacement therapy). We aimed to externally validate discrimination and calibration of the ProVent score and, if necessary, to update its underlying regression model. DESIGN: Retrospective, observational, single-centre study. SETTING: 11 intensive care units at one tertiary academic hospital. PATIENTS: 780 critically ill adult patients receiving invasive mechanical ventilation for at least 21 days. PRIMARY OUTCOME MEASURE: 1-year mortality after intensive care unit discharge. RESULTS: 380 patients (49%) had died after 1 year. One-year mortality for ProVent scores from 0 to 5 were: 15%, 27%, 57%, 66%, 72% and 76%. Area under the receiver operating characteristic curve of the ProVent probability model was 0.76 (95% CI 0.72 to 0.79), calibration intercept was -0.43 (95% CI -0.59 to -0.27) and calibration slope was 0.76 (95% CI 0.62 to 0.89). Model recalibration and extension by inclusion of three additional predictors (total bilirubin concentration, enteral nutrition and surgical status) improved model discrimination and calibration. Decision curve analysis demonstrated that the original ProVent model had negative net benefit, which was avoided with the extended ProVent model. CONCLUSIONS: The ProVent probability model had adequate discrimination but was miscalibrated in our patient cohort and, as such, could potentially be harmful. Use of the extended ProVent score developed by us could possibly alleviate this concern.

Anti-factor Xa Activity Is Not Associated With Venous Thromboembolism in Critically Ill Patients Receiving Enoxaparin for Thromboprophylaxis: A Retrospective Observational Study.

Background: Anti-factor Xa activity has been suggested as a surrogate parameter for judging the effectiveness of pharmacological thromboprophylaxis with low molecular weight heparins in critically ill patients. However, this practice is not supported by evidence associating low anti-factor Xa activity with venous thromboembolism. Methods: We performed a retrospective observational study including 1,352 critically ill patients admitted to 6 intensive care units of the Medical University of Vienna, Austria between 01/2015 and 12/2018. Included patients received prophylactically dosed enoxaparin (≤100 IU/kg body weight per day). We analyzed median peak, 12-h trough and 24-h trough anti-factor Xa activity per patient and compared anti-factor Xa activity between patients without vs. with venous thromboembolic events. Results: 19 patients (1.4%) developed a total of 22 venous thromboembolic events. We did not observe a difference of median (IQR) anti-factor Xa activity between patients without venous thromboembolism [peak 0.22 IU/mL (0.14-0.32); 12-h trough 0.1 IU/mL (<0.1-0.17), 24-h trough < 0.1 IU/mL (<0.1- <0.1)] vs. patients with venous thromboembolism [peak 0.33 IU/mL (0.14-0.34); 12-h trough 0.12 IU/mL (<0.1-0.26); 24-h trough < 0.1 IU/mL (<0.1-<0.1)]. Conclusion: Patients who developed venous thromboembolism had anti-factor Xa activities comparable to those who did not suffer from venous thromboembolism.

Comparison of the New Viscoelastic Coagulation Analyzer ClotPro® With ROTEM® Delta and Conventional Coagulation Tests in Critically Ill Patients With COVID-19.

Background: Viscoelastic coagulation testing has been suggested to help manage coagulopathy in critically ill patients with COVID-19. However, results from different viscoelastic devices are not readily comparable. ClotPro® is a novel thromboelastometry analyzer offering a wider range of commercially available assays. Methods: We compared the results from ClotPro with results from the well-established ROTEM® Delta device and conventional coagulation tests in critically ill patients with COVID-19. Results: Viscoelastic parameters indicated the presence of a potentially hypercoagulable state in the majority of patients. In up to 95 paired measurements, we found strong correlations between several parameters routinely used in clinical practice: (i) EX test vs. EXTEM CT, A5, A10, MCF, (ii) IN test vs. INTEM A5, A10, MCF, and (iii) FIB test vs. FIBTEM A5, A10, MCF (all R > 0.7 and p < 0.001). In contrast, IN test CT vs. INTEM CT showed only a moderate correlation (R = 0.53 and p < 0.001). Clot strength parameters of both devices exhibited strong correlations with platelet counts and fibrinogen levels (all R > 0.7 and p < 0.001). Divergent correlations of intrinsically activated assays with aPTT and anti-factor Xa activity were visible. Regarding absolute differences of test results, considerable delta occurred in CT, CFT, and clot strength parameters (all p < 0.001) between both devices. Conclusions: Several parameters obtained by ClotPro show strong correlations with ROTEM Delta. Due to weak correlations of intrinsically activated clotting times and considerable absolute differences in a number of parameters, our findings underline the need for device-specific algorithms in this patient cohort.

Monitoring of Enoxaparin during Hemodialysis Covered by Regional Citrate Anticoagulation in Acute Kidney Injury: A Prospective Cohort Study.

BACKGROUND: Current guidelines recommend the monitoring of anti-factor Xa (anti-Xa) levels to avoid an accumulation of low-molecular-weight heparins in patients with acute kidney injury, but there is no evidence on how to proceed with such monitoring during continuous renal replacement therapy. Against this background, we investigated the potential accumulation of enoxaparin administered subcutaneously for venous thromboembolism prophylaxis in critically ill patients during continuous renal replacement therapy covered by regional citrate anticoagulation. METHODS: Anti-Xa levels were measured at baseline (≤12 h before renal replacement therapy) and on three consecutive days (A to C) when enoxaparin had reached trough levels. Supplementary testing included modified assays of rotational thromboelastometry known to be highly sensitive for low-molecular-weight heparins. RESULTS: The 16 men and 13 women included were adults comparable in age, body mass index, thromboembolism risk assessment, and clinical severity of the disease. Throughout the four examinations, the median trough levels of anti-Xa remained below the detection limit of the test (<0.1 IU mL-1), with interquartile ranges of <0.1 to 0.14 IU mL-1 at baseline and <0.1 to 0.16 IU mL-1 on days A/B/C. All rotational thromboelastometry parameters of clot initiation and clot formation dynamics did not significantly change from baseline to day C. CONCLUSIONS: Neither anti-Xa levels nor modified assays of rotational thromboelastometry revealed any accumulation of enoxaparin administered for thromboprophylaxis during continuous renal replacement therapy covered by regional citrate anticoagulation. Although generally recommended in patients with acute kidney injury, monitoring of anti-Xa levels should be questioned in this defined setting.

Viscoelasticity and structure of blood clots generated in-vitro by rheometry: A comparison between human, horse, rat, and camel.

BACKGROUND: Although the coagulation system is evolutionary well preserved, profound species differences exist in viscoelastic as well as in common laboratory tests of coagulation. OBJECTIVE: Evaluating differences in clot formation and material characterisation of clots of four mammalian species on macro-, micro- and nanoscales by the means of rheometry, scanning electron microscopy (SEM) and small angle x-ray scattering (SAXS). METHODS: Blood samples were collected from healthy human volunteers, laboratory rats (HL/LE inbred strain), warmblood horses and dromedary camels. Clot formation was observed by oscillating shear rheometry until plateau formation of the shear storage modulus G', at which point selected clots were prepared for scanning electron microscopy. SEM images were analysed for fibre diameter and fractal dimension. Additionally, scattering profiles for plasma and whole blood samples were obtained with SAXS. RESULTS: Viscoelasticity of clots showed great interspecies variation: clots of rats and horses exhibited shorter clotting times and higher G' plateau values, when compared to human clots. Camel clots showed unique clotting characteristics with no G' plateau formation in the timeframe observed. Less differentiating features were found with SEM and SAXS, although the rat fibre network appears to be more convoluted and dense, which resulted in a higher fractal dimension. CONCLUSION: Clotting kinetic differs between the species, which is not only of clinical interest, but could also be an important finding for animal models of blood coagulation.