Objective: Fluid therapy for sepsis patients has always been a problem that puzzles clinicians, that is, knowing when patients need fluid infusion and when they need negative fluid balance. Different clinicians may have different judgment criteria and make different decisions. Recently, studies have suggested that different fluid treatment strategies can cause different clinical outcomes. This study is intended to establish and verify a model for judging the direction of fluid therapy based on machine learning.Method: This study included 2705 sepsis patients from the Peking Union Medical College Hospital Intensive Care Medical Information System and Database (PICMISD) from January 2016 to April 2020. The training set and test set (January 2016 to June 2019) were randomly divided. Twenty-seven features were extracted for modeling, including 25 state features (bloc, vital sign, laboratory examination, blood gas assay and demographics), 1 action feature (fluid balance) and 1 outcome feature (ICU survival or death). SARSA was used to learn the data rules of the training set. Deep Q-learning (DQN) was used to learn the relationship between states and actions of the training set and predict the next balance. A double-robust estimator was used to evaluate the average expected reward of the test set in the deep Q-learning model. Lastly, we verified the difference between the predicted fluid therapy model and the actual treatment for the patient's prognoses, with sepsis patient data from July 2019 to April 2020 as the validation set.Results: The training set and test set were extracted from the same database, and the distribution of liquid balance was similar. Actions were divided into five intervals corresponding to 0–20, 20–40, 40–60, 60–80, and 80–100% percentiles of fluid balance. The higher the reward of Q(s, a) calculated by SARSA from the training set, the lower the mortality rate. Deep Q-learning indicates that both fluid balance differences that are too high and too low show an increase in mortality. The more consistent the fluid balance prediction with the real result, the lower the mortality rate. The smaller the difference between the prediction and the reality, the lower the mortality rate. The double-robust estimator shows that the model has satisfactory stability. The validation set indicates that the mortality rate of patients in the “predicted negative fluid balance and actual negative fluid balance” subgroup was the lowest, which was statistically significant, indicating that the model can be used for clinical verification.Conclusion: We used reinforcement learning to propose a possible prediction model for guiding the direction of fluid therapy for sepsis patients in the ICU. This model may accurately predict the best direction for fluid therapy, thereby improving patient prognosis.

Electrical impedance tomography (EIT) has been used to titrate positive end-expiratory pressure (PEEP). This study aims to develop a comprehensive view of the efficacy and long-term prognosis of EIT-guided PEEP compared to other conventional approaches in various clinical scenarios, including patients with acute respiratory distress syndrome (ARDS), hypoxemic acute respiratory failure (hARF) and patients undergoing surgery under general anesthesia. The literature search was conducted in PubMed, Web of Science, Embase, and Cochrane Library, from inception to July 30, 2023 (ARDS/hARF) and October 5, 2023 (surgery). The Cochrane risk of bias assessment and the methodological index for non-randomized studies were used for quality appraisal. The main outcomes were PEEP level, PaO2/FiO2 ratio, lung/respiratory system compliance (CL/Crs), driving pressure (ΔP), in-hospital mortality, and postoperative pulmonary complications (PPCs) in surgical studies. Four randomized controlled trials (RCTs), one historical control study, and six before-after studies of ARDS/hARF, as well as eight surgical RCTs, were retrieved. Subgroup analysis has been carried out and analysis of before-after studies was performed separately. Diverse PEEP strategies were adopted in the included studies, such as low/high PEEP-FiO2-table of ARDS-net, pressure–volume loop, and transpulmonary pressure. In ARDS/hARF studies, the EIT strategy did not result in considerably enhanced respiratory system mechanics, including comparable PaO2/FiO2 ratios, comparable ΔP, and increased CL/Crs. As for long-term prognosis, the rough overall meta-analysis showed decreased in-hospital mortality (risk ratio RR = 1.54, 95% CI = (1.09, 2.18), P = 0.01). In patients undergoing general anesthesia surgery, the EIT group demonstrated increased PaO2/FiO2 ratio, CL/Crs, and decreased ΔP versus the fixed 4 or 5 cmH2O PEEP. In postoperative prognosis, incidence of PPCs was generally comparable between the two groups. The EIT-derived PEEP setting strategy might be associated with potential benefits in respiratory outcomes and prognosis in ARDS/hARF and surgical patients. Current data is insufficient to provide solid evidence.

Background: Regional citrate anticoagulation (RCA) is an important local anticoagulation method during bedside continuous renal replacement therapy. To improve patient safety and achieve computer assisted dose monitoring and control, we took intensive care units patients into cohort and aiming at developing a data-driven machine learning model to give early warning of citric acid overdose and provide adjustment suggestions on citrate pumping rate and 10% calcium gluconate input rate for RCA treatment.Methods: Patient age, gender, pumped citric acid dose value, 5% NaHCO₃solvent, replacement fluid solvent, body temperature value, and replacement fluid PH value as clinical features, models attempted to classify patients who received regional citrate anticoagulation into correct outcome category. Four models, Adaboost, XGBoost, support vector machine (SVM) and shallow neural network, were compared on the performance of predicting outcomes. Prediction results were evaluated using accuracy, precision, recall and F1-score.Results: For classifying patients at the early stages of citric acid treatment, the accuracy of neutral networks model is higher than Adaboost, XGBoost and SVM, the F1-score of shallow neutral networks (90.77%) is overall outperformed than other models (88.40%, 82.17% and 88.96% for Adaboost, XGBoost and SVM). Extended experiment and validation were further conducted using the MIMIC-III database, the F1-scores for shallow neutral networks, Adaboost, XGBoost and SVM are 80.00%, 80.46%, 80.37% and 78.90%, the AUCs are 0.8638, 0.8086, 0.8466 and 0.7919 respectively.Conclusion: The results of this study demonstrated the feasibility and performance of machine learning methods for monitoring and adjusting local regional citrate anticoagulation, and further provide decision-making recommendations to clinicians point-of-care.

Patient-ventilator asynchrony (PVA) can result in ventilator-induced lung injury (VILI), prolong mechanical ventilation, and ventilator withdrawal failure. The ability of healthcare providers in China to recognize patient-ventilator asynchrony is unknown. The aim of our study was to evaluate the ability and potential influencing factors to correctly identify patient-ventilator triggering asynchrony in tertiary hospitals in China. This was an observational study carried out in 53 tertiary hospitals in China. A total of 191 healthcare providers were asked to finish entry test and evaluation test sequentially. Entry test identified qualified professionals by matching concepts with its corresponding interpretations. Evaluation test assessed the ability in recognizing patient-ventilator asynchrony waveforms by matching asynchrony waveforms with corresponding concepts. A total of 109 qualified professionals were identified. Further analysis based on professional title, role in critical care team, years of experience in managing invasive mechanical ventilation, number of published articles in the field of clinical critical respiratory medicine and training in respiratory waveform/respiratory mechanics was carried out among qualified professionals. A self-innovate Remote-VentlateView platform was used to discriminate the patient-ventilator triggering asynchrony. Among 109 qualified professionals, the average recognition accuracy was 3.45 out of 8 sets. Inconsistency of concept cognition and waveform recognition of patient-ventilator asynchrony was found among all types of asynchronies. The accuracy of the trained professionals was greater than that of the nontrained professionals for ineffective trigger [76.7% vs. 59.2% (p = 0.009)], auto-trigger [26.7% vs. 12.2% (p = 0.014)] and reverse triggers [30.8% vs. 12.2% (p = 0.002)]. Professionals who published more than 2 articles in the field of critical respiratory performed better on auto-triggers [41.7% vs. 15.9% (p = 0.001)] and reverse triggers [38.9% vs. 19.2% (p = 0.018)]. Neither experience in managing invasive mechanical ventilation nor professional title was associated with the ability of healthcare providers to identify asynchrony. Receiving training in mechanical ventilation and conducting critical respiratory clinical research may increase healthcare providers’ ability to identify patient-ventilator asynchrony by using waveform analysis. The Remote-VentlateView platform may assist in identifying patient-ventilator asynchronies.

The relationship between perfusion index (PI) and organ dysfunction in patients in the intensive care unit (ICU) is not clear. This study aimed to explore the relationship between PI and renal function in the perioperative critical care setting and evaluate the predictive efficiency of PI on patients with acute kidney injury (AKI) in the ICU. This retrospective analysis involved 12,979 patients who had undergone an operation and were admitted to the ICU in Peking Union Medical College Hospital from January 2014 to December 2019. The distribution of average PI in the first 24 h after ICU admission and its correlation with AKI was calculated by Cox regression. Receiver operating characteristic (ROC) curves were generated to compare the ability of PI, mean arterial pressure (MAP), creatinine, blood urea nitrogen (BUN), and central venous pressure (CVP) to discriminate AKI in the first 48 h in all perioperative critically ill patients. Average PI in the first 24 h served as an independent protective factor of AKI (Odds ratio [OR]=0.786, 95% confidence interval [CI]: 0.704–0.873, P 2.12). As a local blood flow indicator, the initial 24-h average PI for perioperative critically ill patients can predict AKI during their first 120 h in the ICU.

This study aimed to develop and validate a non-breath-holding contrast-enhanced electrical impedance tomography (EIT) method using low-pass filtering for b