Background: Patients with COVID-19 in the intensive care unit (ICU) have a high mortality rate, and methods to assess patients’ prognosis early and administer precise treatment are of great significance. Objective: The aim of this study was to use machine learning to construct a model for the analysis of risk factors and prediction of mortality among ICU patients with COVID-19. Methods: In this study, 123 patients with COVID-19 in the ICU of Vulcan Hill Hospital were retrospectively selected from the database, and the data were randomly divided into a training data set (n=98) and test data set (n=25) with a 4:1 ratio. Significance tests, correlation analysis, and factor analysis were used to screen 100 potential risk factors individually. Conventional logistic regression methods and four machine learning algorithms were used to construct the risk prediction model for the prognosis of patients with COVID-19 in the ICU. The performance of these machine learning models was measured by the area under the receiver operating characteristic curve (AUC). Interpretation and evaluation of the risk prediction model were performed using calibration curves, SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), etc, to ensure its stability and reliability. The outcome was based on the ICU deaths recorded from the database. Results: Layer-by-layer screening of 100 potential risk factors finally revealed 8 important risk factors that were included in the risk prediction model: lymphocyte percentage, prothrombin time, lactate dehydrogenase, total bilirubin, eosinophil percentage, creatinine, neutrophil percentage, and albumin level. Finally, an eXtreme Gradient Boosting (XGBoost) model established with the 8 important risk factors showed the best recognition ability in the training set of 5-fold cross validation (AUC=0.86) and the verification queue (AUC=0.92). The calibration curve showed that the risk predicted by the model was in good agreement with the actual risk. In addition, using the SHAP and LIME algorithms, feature interpretation and sample prediction interpretation algorithms of the XGBoost black box model were implemented. Additionally, the model was translated into a web-based risk calculator that is freely available for public usage. Conclusions: The 8-factor XGBoost model predicts risk of death in ICU patients with COVID-19 well; it initially demonstrates stability and can be used effectively to predict COVID-19 prognosis in ICU patients.

Background: The purpose of this study was to explore the diagnostic value of soluble triggering receptor expressed on myeloid cells 1 (sTREM-1), procalcitonin (PCT), and C-reactive protein (CRP) serum levels for differentiating sepsis from SIRS, identifying new fever caused by bacteremia, and assessing prognosis when new fever occurred. Methods: We enrolled 144 intensive care unit (ICU) patients: 60 with systemic inflammatory response syndrome (SIRS) and 84 with sepsis complicated by new fever at more than 48 h after ICU admission. Serum sTREM-1, PCT, and CRP levels were measured on the day of admission and at the occurrence of new fever (>38.3°C) during hospitalization. Based on the blood culture results, the patients were divided into a blood culture-positive bacteremia group (33 patients) and blood culture-negative group (51 patients). Based on 28-day survival, all patients, both blood culture-positive and -negative, were further divided into survivor and nonsurvivor groups. Results: On ICU day 1, the sepsis group had higher serum sTREM-1, PCT, and CRP levels compared with the SIRS group (P <0.05). The areas under the curve (AUC) for these indicators were 0.868 (95% CI, 0.798–0.938), 0.729 (95% CI, 0.637–0.821), and 0.679 (95% CI, 0.578–0.771), respectively. With 108.9 pg/ml as the cut-off point for serum sTREM-1, sensitivity was 0.83 and specificity was 0.81. There was no statistically significant difference in serum sTREM-1 or PCT levels between the blood culture-positive and -negative bacteremia groups with ICU-acquired new fever. However, the nonsurvivors in the blood culture-positive bacteremia group had higher levels of serum sTREM-1 and PCT (P <0.05), with a prognostic AUC for serum sTREM-1 of 0.868 (95% CI, 0.740–0.997). Conclusions: Serum sTREM-1, PCT, and CRP levels each have a role in the early diagnosis of sepsis. Serum sTREM-1, with the highest sensitivity and specificity of all indicators studied, is especially notable. sTREM-1, PCT, and CRP levels are of no use in determining new fever caused by bacteremia in ICU patients, but sTREM-1 levels reflect the prognosis of bacteremia. Trial registration: ClinicalTrial.gov identifier NCT01410578

Background: Modern clinical care in intensive care units is full of rich data, and machine learning has great potential to support clinical decision-making. The development of intelligent machine learning–based clinical decision support systems is facing great opportunities and challenges. Clinical decision support systems may directly help clinicians accurately diagnose, predict outcomes, identify risk events, or decide treatments at the point of care. Objective: We aimed to review the research and application of machine learning–enabled clinical decision support studies in intensive care units to help clinicians, researchers, developers, and policy makers better understand the advantages and limitations of machine learning–supported diagnosis, outcome prediction, risk event identification, and intensive care unit point-of-care recommendations. Methods: We searched papers published in the PubMed database between January 1980 and October 2020. We defined selection criteria to identify papers that focused on machine learning–enabled clinical decision support studies in intensive care units and reviewed the following aspects: research topics, study cohorts, machine learning models, analysis variables, and evaluation metrics. Results: A total of 643 papers were collected, and using our selection criteria, 97 studies were found. Studies were categorized into 4 topics—monitoring, detection, and diagnosis (13/97, 13.4%), early identification of clinical events (32/97, 33.0%), outcome prediction and prognosis assessment (46/97, 47.6%), and treatment decision (6/97, 6.2%). Of the 97 papers, 82 (84.5%) studies used data from adult patients, 9 (9.3%) studies used data from pediatric patients, and 6 (6.2%) studies used data from neonates. We found that 65 (67.0%) studies used data from a single center, and 32 (33.0%) studies used a multicenter data set; 88 (90.7%) studies used supervised learning, 3 (3.1%) studies used unsupervised learning, and 6 (6.2%) studies used reinforcement learning. Clinical variable categories, starting with the most frequently used, were demographic (n=74), laboratory values (n=59), vital signs (n=55), scores (n=48), ventilation parameters (n=43), comorbidities (n=27), medications (n=18), outcome (n=14), fluid balance (n=13), nonmedicine therapy (n=10), symptoms (n=7), and medical history (n=4). The most frequently adopted evaluation metrics for clinical data modeling studies included area under the receiver operating characteristic curve (n=61), sensitivity (n=51), specificity (n=41), accuracy (n=29), and positive predictive value (n=23). Conclusions: Early identification of clinical and outcome prediction and prognosis assessment contributed to approximately 80% of studies included in this review. Using new algorithms to solve intensive care unit clinical problems by developing reinforcement learning, active learning, and time-series analysis methods for clinical decision support will be greater development prospects in the future.

It is very important to assess the effectiveness of prone positioning (PP) in patients with severe acute respiratory distress syndrome (ARDS). However, it is difficult to identify patients who may benefit from PP. The purpose of this study was to investigate whether prone positioning potential (PPP) can be predicted by lung ultrasound in patients with ARDS. In this prospective study, 45 patients with ARDS were included for the assessment of PPP. A PP lung ultrasound examination (PLUE) protocol was performed in the dorsal regions of the lung in 16 areas at H0, H3, and H6 (0, 3, and 6 h after PP). The ultrasonography videos were blindly evaluated by two expert clinicians to classify the lung regions as normal pattern (N), moderate loss of lung aeration (B1), severe loss of lung aeration (B2), and consolidation (C). The aeration scores were collected at H0, H3, and H6. According to the ratio of partial pressure of arterial oxygen to fraction of inspired oxygen (P/F ratio) at 7 days, patients were classified into PPP-positive (P/F ratio >300) and PPP-negative groups; also, the patients were classified into survival and nonsurvival groups according to 28-day mortality. Aeration scores was compared at H0, H3, and H6. The scores were significantly reduced between H3 and H0, but there was no difference between H3 and H6. The aeration score variation (ASV) of the PPP-positive group between H3 and H0 was significantly higher than that in the PPP-negative group, and the sensitivity and specificity of ASV ≥5.5 for the PPP-positive group were 73.9% and 86.4%, respectively. The area under the receiver operating characteristic curve (AUROC) was 0.852 for the ASV. The ASV between H3 and H0 in the survival group was significantly higher than in the nonsurvival group. The sensitivity and specificity of ASV ≥7 for survival were 51.5% and 75%, respectively. The AUROC was 0.702 for the ASV. The PLUE protocol can be used to predict PPP and assess prognosis in patients with ARDS.

The gut microbiota of intensive care unit (ICU) patients displays extreme dysbiosis associated with increased susceptibility to organ failure, sepsis, and septic shock. However, such dysbiosis is difficult to characterize owing to the high dimensional complexity of the gut microbiota. We tested whether the concept of enterotype can be applied to the gut microbiota of ICU patients to describe the dysbiosis. We collected 131 fecal samples from 64 ICU patients diagnosed with sepsis or septic shock and performed 16S rRNA gene sequencing to dissect their gut microbiota compositions. During the development of sepsis or septic shock and during various medical treatments, the ICU patients always exhibited two dysbiotic microbiota patterns, or ICU-enterotypes, which could not be explained by host properties such as age, sex, and body mass index, or external stressors such as infection site and antibiotic use. ICU-enterotype I (ICU E1) comprised predominantly Bacteroides and an unclassified genus of Enterobacteriaceae, while ICU-enterotype II (ICU E2) comprised predominantly Enterococcus. Among more critically ill patients with Acute Physiology and Chronic Health Evaluation II (APACHE II) scores > 18, septic shock was more likely to occur with ICU E1 (P = 0.041). Additionally, ICU E1 was correlated with high serum lactate levels (P = 0.007). Therefore, different patterns of dysbiosis were correlated with different clinical outcomes, suggesting that ICU-enterotypes should be diagnosed as independent clinical indices. Thus, the microbial-based human index classifier we propose is precise and effective for timely monitoring of ICU-enterotypes of individual patients. This work is a first step toward precision medicine for septic patients based on their gut microbiota profiles.

Background: The prevalence of carbapenem-resistant Klebsiella pneumoniae bloodstream infection (CRKP-BSI) is increasing worldwide. CRKP-BSI is associated with high rates of morbidity and mortality due to limited antibiotic choices. Here, we aim to identify the prevalence and risk factors for infection and mortality of CRKP BSI. Methods: This was a retrospective study of the past data from January 1st, 2012 to December 31st, 2019 of adult patients with KP-BSI in Xiangya Hospital, China. Results: Among the 706 incidences included in this study, 27.4% of them (212/753) being CR-KP strains. The occurrence of CRKP-BSI was increased from 20.69 to 37.40% from 2012 to 2019. Hematologic malignancies and ICU acquired infection were identified to be substantial risk factors of carbapenem resistance. The overall 28-day mortality rates of CRKP-BSI patients was significantly higher than that of CSKP-BSI (P < 0.001). Logistic regression analysis identified severe sepsis or septic shock incidents, inadequate empirical antimicrobial therapy and corticosteroids use preceding infection onset as the independent predictors of 28-day mortality of CRKP-BSI patients. However, high dose carbapenem combination therapy was identified as anticipated factors of low 28-day mortality. Conclusion: The occurrence of CRKP-BSI was significantly increased during the study period. Hematologic malignancies and ICU acquired infection were associated with the development of CRKP BSI. Severe sepsis or septic shock incidents, inadequate empirical antimicrobial therapy and corticosteroids use preceding infection onset caused significant increase of mortality rates in CRKP-BSI patients. High dose carbapenem combination therapy was associated with better outcome.