The impact of ethnic background on ICU care and outcome in sepsis and septic shock - A retrospective multicenter analysis on 17,949 patients.

BACKGROUND: Previous studies have been inconclusive about racial disparities in sepsis. This study evaluated the impact of ethnic background on management and outcome in sepsis and septic shock. METHODS: This analysis included 17,146 patients suffering from sepsis and septic shock from the multicenter eICU Collaborative Research Database. Generalized estimated equation (GEE) population-averaged models were used to fit three sequential regression models for the binary primary outcome of hospital mortality. RESULTS: Non-Hispanic whites were the predominant group (n = 14,124), followed by African Americans (n = 1,852), Hispanics (n = 717), Asian Americans (n = 280), Native Americans (n = 146) and others (n = 830). Overall, the intensive care treatment and hospital mortality were similar between all ethnic groups. This finding was concordant in patients with septic shock and persisted after adjusting for patient-level variables (age, sex, mechanical ventilation, vasopressor use and comorbidities) and hospital variables (teaching hospital status, number of beds in the hospital). CONCLUSION: We could not detect ethnic disparities in the management and outcomes of critically ill septic patients and patients suffering from septic shock. Disparate outcomes among critically ill septic patients of different ethnicities are a public health, rather than a critical care challenge.

Automatically detecting Crohn's disease and Ulcerative Colitis from endoscopic imaging.

BACKGROUND: The SI-CURA project (Soluzioni Innovative per la gestione del paziente e il follow up terapeutico della Colite UlceRosA) is an Italian initiative aimed at the development of artificial intelligence solutions to discriminate pathologies of different nature, including inflammatory bowel disease (IBD), namely Ulcerative Colitis (UC) and Crohn's disease (CD), based on endoscopic imaging of patients (P) and healthy controls (N). METHODS: In this study we develop a deep learning (DL) prototype to identify disease patterns through three binary classification tasks, namely (1) discriminating positive (pathological) samples from negative (healthy) samples (P vs N); (2) discrimination between Ulcerative Colitis and Crohn's Disease samples (UC vs CD) and, (3) discrimination between Ulcerative Colitis and negative (healthy) samples (UC vs N). RESULTS: The model derived from our approach achieves a high performance of Matthews correlation coefficient (MCC) > 0.9 on the test set for P versus N and UC versus N, and MCC > 0.6 on the test set for UC versus CD. CONCLUSION: Our DL model effectively discriminates between pathological and negative samples, as well as between IBD subgroups, providing further evidence of its potential as a decision support tool for endoscopy-based diagnosis.

Prediction of blood lactate values in critically ill patients: a retrospective multi-center cohort study.

Elevations in initially obtained serum lactate levels are strong predictors of mortality in critically ill patients. Identifying patients whose serum lactate levels are more likely to increase can alert physicians to intensify care and guide them in the frequency of tending the blood test. We investigate whether machine learning models can predict subsequent serum lactate changes. We investigated serum lactate change prediction using the MIMIC-III and eICU-CRD datasets in internal as well as external validation of the eICU cohort on the MIMIC-III cohort. Three subgroups were defined based on the initial lactate levels: (i) normal group (< 2 mmol/L), (ii) mild group (2-4 mmol/L), and (iii) severe group (> 4 mmol/L). Outcomes were defined based on increase or decrease of serum lactate levels between the groups. We also performed sensitivity analysis by defining the outcome as lactate change of > 10% and furthermore investigated the influence of the time interval between subsequent lactate measurements on predictive performance. The LSTM models were able to predict deterioration of serum lactate values of MIMIC-III patients with an AUC of 0.77 (95% CI 0.762-0.771) for the normal group, 0.77 (95% CI 0.768-0.772) for the mild group, and 0.85 (95% CI 0.840-0.851) for the severe group, with only a slightly lower performance in the external validation. The LSTM demonstrated good discrimination of patients who had deterioration in serum lactate levels. Clinical studies are needed to evaluate whether utilization of a clinical decision support tool based on these results could positively impact decision-making and patient outcomes.

Management of intoxicated patients - a descriptive outcome analysis of 4,267 ICU patients.

INTRODUCTION: Intoxications are common in intensive care units (ICUs). The number of causative substances is large, mortality usually low. This retrospective cohort study aims to characterize differences of intoxicated compared to general ICU patients, point out variations according to causative agents, as well as to highlight differences between survivors and non-survivors among intoxicated individuals in a large-scale multi-center analysis. METHODS: A total of 105,998 general ICU patients and 4,267 individuals with the admission diagnoses "overdose" and "drug toxicity" from the years 2014 and 2015 where included from the eICU Collaborative Research Database. In addition to comparing these groups with respect to baseline characteristics, intensive care measures and outcome parameters, differences between survivors and non-survivors from the intoxication group, as well as the individual groups of causative substances were investigated. RESULTS: Intoxicated patients were younger (median 41 vs. 66 years; p<0.001), more often female (55 vs. 45%; p<0.001), and normal weighted (36% vs. 30%; p<0.001), whereas more obese individuals where observed in the other group (37 vs. 31%; p<0.001). Intoxicated individuals had a significantly lower mortality compared to general ICU patients (1% vs. 10%; aOR 0.07 95%CI 0.05-0.11; p<0.001), a finding which persisted after multivariable adjustment (aOR 0.17 95%CI 0.12-0.24; p<0.001) and persisted in all subgroups. Markers of disease severity (SOFA-score: 3 (1-5) vs. 4 (2-6) pts.; p<0.001) and frequency of vasopressor use (5 vs. 15%; p<0.001) where lower, whereas rates of mechanical ventilation where higher (24 vs. 26%; p<0.001) in intoxicated individuals. There were no differences with regard to renal replacement therapy in the first three days (3 vs. 4%; p=0.26). In sensitivity analysis (interactions for age, sex, ethnicity, hospital category, maximum initial lactate, mechanical ventilation, and vasopressor use), a trend towards lower mortality in intoxicated patients persisted in all subgroups. CONCLUSION: This large-scale retrospective analysis indicates a significantly lower mortality of intoxicated individuals compared to general ICU patients.

Red Cell Distribution Width Is Independently Associated with Mortality in Sepsis.

BACKGROUND: Mortality in sepsis remains high. Studies on small cohorts have shown that red cell distribution width (RDW) is associated with mortality. The aim of this study was to validate these findings in a large multicenter cohort. METHODS: We conducted this retrospective analysis of the multicenter eICU Collaborative Research Database in 16,423 septic patients. We split the cohort in patients with low (≤15%; n = 7,129) and high (>15%; n = 9,294) RDW. Univariable and multivariable multilevel logistic regressions were used to fit regression models for the binary primary outcome of hospital mortality and the secondary outcome intensive care unit (ICU) mortality with hospital unit as random effect. Optimal cutoffs were calculated using the Youden index. RESULTS: Patients with high RDW were more often older than 65 years (57% vs. 50%; p < 0.001) and had higher Acute Physiology and Chronic Health Evaluation (APACHE) IV scores (69 vs. 60 pts.; p < 0.001). Both hospital (adjusted odds ratios [aOR] 1.18; 95% CI: 1.16-1.20; p < 0.001) and ICU mortality (aOR 1.16; 95% CI: 1.14-1.18; p < 0.001) were associated with RDW as a continuous variable. Patients with high RDW had a higher hospital mortality (20 vs. 9%; aOR 2.63; 95% CI: 2.38-2.90; p < 0.001). This finding persisted after multivariable adjustment (aOR 2.14; 95% CI: 1.93-2.37; p < 0.001) in a multilevel logistic regression analysis. The optimal RDW cutoff for the prediction of hospital mortality was 16%. CONCLUSION: We found an association of RDW with mortality in septic patients and propose an optimal cutoff value for risk stratification. In a combined model with lactate, RDW shows equivalent diagnostic performance to Sequential Organ Failure Assessment (SOFA) score and APACHE IV score.

Machine learning predicts mortality based on analysis of ventilation parameters of critically ill patients: multi-centre validation.

BACKGROUND: Mechanical Ventilation (MV) is a complex and central treatment process in the care of critically ill patients. It influences acid-base balance and can also cause prognostically relevant biotrauma by generating forces and liberating reactive oxygen species, negatively affecting outcomes. In this work we evaluate the use of a Recurrent Neural Network (RNN) modelling to predict outcomes of mechanically ventilated patients, using standard mechanical ventilation parameters. METHODS: We performed our analysis on VENTILA dataset, an observational, prospective, international, multi-centre study, performed to investigate the effect of baseline characteristics and management changes over time on the all-cause mortality rate in mechanically ventilated patients in ICU. Our cohort includes 12,596 adult patients older than 18, associated with 12,755 distinct admissions in ICUs across 37 countries and receiving invasive and non-invasive mechanical ventilation. We carry out four different analysis. Initially we select typical mechanical ventilation parameters and evaluate the machine learning model on both, the overall cohort and a subgroup of patients admitted with respiratory disorders. Furthermore, we carry out sensitivity analysis to evaluate whether inclusion of variables related to the function of other organs, improve the predictive performance of the model for both the overall cohort as well as the subgroup of patients with respiratory disorders. RESULTS: Predictive performance of RNN-based model was higher with Area Under the Receiver Operating Characteristic (ROC) Curve (AUC) of 0.72 (± 0.01) and Average Precision (AP) of 0.57 (± 0.01) in comparison to RF and LR for the overall patient dataset. Higher predictive performance was recorded in the subgroup of patients admitted with respiratory disorders with AUC of 0.75 (± 0.02) and AP of 0.65 (± 0.03). Inclusion of function of other organs further improved the performance to AUC of 0.79 (± 0.01) and AP 0.68 (± 0.02) for the overall patient dataset and AUC of 0.79 (± 0.01) and AP 0.72 (± 0.02) for the subgroup with respiratory disorders. CONCLUSION: The RNN-based model demonstrated better performance than RF and LR in patients in mechanical ventilation and its subgroup admitted with respiratory disorders. Clinical studies are needed to evaluate whether it impacts decision-making and patient outcomes. TRIAL REGISTRATION: NCT02731898 ( https://clinicaltrials.gov/ct2/show/NCT02731898 ), prospectively registered on April 8, 2016.

Monitoring and detecting faults in wastewater treatment plants using deep learning.

Wastewater treatment plants use many sensors to control energy consumption and discharge quality. These sensors produce a vast amount of data which can be efficiently monitored by automatic systems. Consequently, several different statistical and learning methods are proposed in the literature which can automatically detect faults. While these methods have shown promising results, the nonlinear dynamics and complex interactions of the variables in wastewater data necessitate more powerful methods with higher learning capacities. In response, this study focusses on modelling faults in the oxidation and nitrification process. Specifically, this study investigates a method based on deep neural networks (specifically, long short-term memory) compared with statistical and traditional machine-learning methods. The network is specifically designed to capture temporal behaviour of sensor data. The proposed method is evaluated on a real-life dataset containing over 5.1 million sensor data points. The method achieved a fault detection rate (recall) of over 92%, thus outperforming traditional methods and enabling timely detection of collective faults.

Stress modelling and prediction in presence of scarce data.

OBJECTIVE: Stress at work is a significant occupational health concern. Recent studies have used various sensing modalities to model stress behaviour based on non-obtrusive data obtained from smartphones. However, when the data for a subject is scarce it becomes a challenge to obtain a good model. METHODS: We propose an approach based on a combination of techniques: semi-supervised learning, ensemble methods and transfer learning to build a model of a subject with scarce data. Our approach is based on the comparison of decision trees to select the closest subject for knowledge transfer. RESULTS: We present a real-life, unconstrained study carried out with 30 employees within two organisations. The results show that using information (instances or model) from similar subjects can improve the accuracy of the subjects with scarce data. However, using transfer learning from dissimilar subjects can have a detrimental effect on the accuracy. Our proposed ensemble approach increased the accuracy by ≈10% to 71.58% compared to not using any transfer learning technique. CONCLUSIONS: In contrast to high precision but highly obtrusive sensors, using smartphone sensors for measuring daily behaviours allowed us to quantify behaviour changes, relevant to occupational stress. Furthermore, we have shown that use of transfer learning to select data from close models is a useful approach to improve accuracy in presence of scarce data.

Artificial intelligence of imaging and clinical neurological data for predictive, preventive and personalized (P3) medicine for Parkinson Disease: The NeuroArtP3 protocol for a multi-center research study.

BACKGROUND: The burden of Parkinson Disease (PD) represents a key public health issue and it is essential to develop innovative and cost-effective approaches to promote sustainable diagnostic and therapeutic interventions. In this perspective the adoption of a P3 (predictive, preventive and personalized) medicine approach seems to be pivotal. The NeuroArtP3 (NET-2018-12366666) is a four-year multi-site project co-funded by the Italian Ministry of Health, bringing together clinical and computational centers operating in the field of neurology, including PD. OBJECTIVE: The core objectives of the project are: i) to harmonize the collection of data across the participating centers, ii) to structure standardized disease-specific datasets and iii) to advance knowledge on disease's trajectories through machine learning analysis. METHODS: The 4-years study combines two consecutive research components: i) a multi-center retrospective observational phase; ii) a multi-center prospective observational phase. The retrospective phase aims at collecting data of the patients admitted at the participating clinical centers. Whereas the prospective phase aims at collecting the same variables of the retrospective study in newly diagnosed patients who will be enrolled at the same centers. RESULTS: The participating clinical centers are the Provincial Health Services (APSS) of Trento (Italy) as the center responsible for the PD study and the IRCCS San Martino Hospital of Genoa (Italy) as the promoter center of the NeuroartP3 project. The computational centers responsible for data analysis are the Bruno Kessler Foundation of Trento (Italy) with TrentinoSalute4.0 -Competence Center for Digital Health of the Province of Trento (Italy) and the LISCOMPlab University of Genoa (Italy). CONCLUSIONS: The work behind this observational study protocol shows how it is possible and viable to systematize data collection procedures in order to feed research and to advance the implementation of a P3 approach into the clinical practice through the use of AI models.

An interpretable deep learning model for time-series electronic health records: Case study of delirium prediction in critical care.

Deep Learning (DL) models have received increasing attention in the clinical setting, particularly in intensive care units (ICU). In this context, the interpretability of the outcomes estimated by the DL models is an essential step towards increasing adoption of DL models in clinical practice. To address this challenge, we propose an ante-hoc, interpretable neural network model. Our proposed model, named double self-attention architecture (DSA), uses two attention-based mechanisms, including self-attention and effective attention. It can capture the importance of input variables in general, as well as changes in importance along the time dimension for the outcome of interest. We evaluated our model using two real-world clinical datasets covering 22840 patients in predicting onset of delirium 12 h and 48 h in advance. Additionally, we compare the descriptive performance of our model with three post-hoc interpretable algorithms as well as with the opinion of clinicians based on the published literature and clinical experience. We find that our model covers the majority of the top-10 variables ranked by the other three post-hoc interpretable algorithms as well as the clinical opinion, with the advantage of taking into account both, the dependencies among variables as well as dependencies between varying time-steps. Finally, our results show that our model can improve descriptive performance without sacrificing predictive performance.

Hyperlactatemia and altered lactate kinetics are associated with excess mortality in sepsis : A multicenter retrospective observational study.

Severe hyperlactatemia (>10mmol/L) or impaired lactate metabolism are known to correlate with increased mortality. The maximum lactate concentration on day 1 of 10,724 septic patients from the eICU Collaborative Research Database was analyzed and patients were divided into three groups based on maximum lactate in the first 24 h (<5mmol/l; ≥5mmol/l & <10mmol/l; ≥10mmol/l). In addition, delta lactate was calculated using the following formula: (maximum lactate day 1 minus maximum lactate day 2) divided by maximum lactate day 1. A multilevel regression analysis was performed, with hospital mortality serving as the primary study end point. Significant differences in hospital mortality were found in patients with hyperlactatemia (lactate ≥10mmol/l: 79%, ≥5mmol/l & <10mmol/l: 43%, <5mmol/l, 13%; p<0.001). The sensitivity of severe hyperlactatemia (≥10mmol/l) for hospital mortality was 17%, the specificity was 99%. In patients with negative delta lactate in the first 24 h, hospital mortality was excessive (92%). In conclusion, mortality in patients with severe hyperlactatemia is very high, especially if it persists for more than 24 h. Severe hyperlactatemia, together with clinical parameters, could therefore provide a basis for setting treatment limits.

Machine learning models predict liver steatosis but not liver fibrosis in a prospective cohort study.

INTRODUCTION: Screening for liver fibrosis continues to rely on laboratory panels and non-invasive tests such as FIB-4-score and transient elastography. In this study, we evaluated the potential of machine learning (ML) methods to predict liver steatosis on abdominal ultrasound and liver fibrosis, namely the intermediate-high risk of advanced fibrosis, in individuals participating in a screening program for colorectal cancer. METHODS: We performed ultrasound on 5834 patients admitted between 2006 and 2020, and transient elastography on a subset of 1240 patients. Steatosis on ultrasound was diagnosed if liver areas showed a significantly increased echogenicity compared to the renal parenchyma. Liver fibrosis was defined as a liver stiffness measurement ≥8 kPa in transient elastography. We evaluated the performance of three algorithms, namely Extreme Gradient Boosting, Feed-Forward neural network and Logistic Regression, deriving the models using data from patients admitted from January 2007 up to January 2016 and prospectively evaluating on the data of patients admitted from January 2016 up to March 2020. We also performed a performance comparison with the standard clinical test based on Fibrosis-4 Index (FIB-4). RESULTS: The mean age was 58±9 years with 3036 males (52%). Modelling laboratory parameters, clinical parameters, and data on eight food types/dietary patterns, we achieved high performance in predicting liver steatosis on ultrasound with AUC of 0.87 (95% CI [0.87-0.87]), and moderate performance in predicting liver fibrosis with AUC of 0.75 (95% CI [0.74-0.75]) using XGBoost machine learning algorithm. Patient-reported variables did not significantly improve predictive performance. Gender-specific analyses showed significantly higher performance in males with AUC of 0.74 (95% CI [0.73-0.74]) in comparison to female patients with AUC of 0.66 (95% CI [0.65-0.66]) in prediction of liver fibrosis. This difference was significantly smaller in prediction of steatosis with AUC of 0.85 (95% CI [0.83-0.87]) in female patients, in comparison to male patients with AUC of 0.82 (95% CI [0.80-0.84]). CONCLUSION: ML based on point-prevalence laboratory and clinical information predicts liver steatosis with high accuracy and liver fibrosis with moderate accuracy. The observed gender differences suggest the need to develop gender-specific models.

Deep ROC Analysis and AUC as Balanced Average Accuracy, for Improved Classifier Selection, Audit and Explanation.

Optimal performance is desired for decision-making in any field with binary classifiers and diagnostic tests, however common performance measures lack depth in information. The area under the receiver operating characteristic curve (AUC) and the area under the precision recall curve are too general because they evaluate all decision thresholds including unrealistic ones. Conversely, accuracy, sensitivity, specificity, positive predictive value and the F1 score are too specific-they are measured at a single threshold that is optimal for some instances, but not others, which is not equitable. In between both approaches, we propose deep ROC analysis to measure performance in multiple groups of predicted risk (like calibration), or groups of true positive rate or false positive rate. In each group, we measure the group AUC (properly), normalized group AUC, and averages of: sensitivity, specificity, positive and negative predictive value, and likelihood ratio positive and negative. The measurements can be compared between groups, to whole measures, to point measures and between models. We also provide a new interpretation of AUC in whole or part, as balanced average accuracy, relevant to individuals instead of pairs. We evaluate models in three case studies using our method and Python toolkit and confirm its utility.

Underweight but not overweight is associated with excess mortality in septic ICU patients.

BACKGROUND: Higher survival has been shown for overweight septic patients compared with normal or underweight patients in the past. This study aimed at investigating the management and outcome of septic ICU patients in different body mass index (BMI) categories in a large multicenter database. METHODS: In total, 16,612 patients of the eICU collaborative research database were included. Baseline characteristics and data on organ support were documented. Multilevel logistic regression analysis was performed to fit three sequential regression models for the binary primary outcome (ICU mortality) to evaluate the impact of the BMI categories: underweight (<18.5 kg/m2), normal weight (18.5 to < 25 kg/m2), overweight (25 to < 30 kg/m2) and obesity (≥ 30 kg/m2). Data were adjusted for patient level characteristics (model 2) as well as management strategies (model 3). RESULTS: Management strategies were similar across BMI categories. Underweight patients evidenced higher rates of ICU mortality. This finding persisted after adjusting in model 2 (aOR 1.54, 95% CI 1.15-2.06; p = 0.004) and model 3 (aOR 1.57, 95%CI 1.16-2.12; p = 0.003). No differences were found regarding ICU mortality between normal and overweight patients (aOR 0.93, 95%CI 0.81-1.06; p = 0.29). Obese patients evidenced a lower risk of ICU mortality compared to normal weight, a finding which persisted across all models (model 2: aOR 0.83, 95%CI 0.69-0.99; p = 0.04; model 3: aOR 0.82, 95%CI 0.68-0.98; p = 0.03). The protective effect of obesity and the negative effect of underweight were significant in individuals > 65 years only. CONCLUSION: In this cohort, underweight was associated with a worse outcome, whereas obese patients evidenced lower mortality. Our analysis thus supports the thesis of the obesity paradox.

Delirium prediction in the ICU: designing a screening tool for preventive interventions.

Introduction: Delirium occurrence is common and preventive strategies are resource intensive. Screening tools can prioritize patients at risk. Using machine learning, we can capture time and treatment effects that pose a challenge to delirium prediction. We aim to develop a delirium prediction model that can be used as a screening tool. Methods: From the eICU Collaborative Research Database (eICU-CRD) and the Medical Information Mart for Intensive Care version III (MIMIC-III) database, patients with one or more Confusion Assessment Method-Intensive Care Unit (CAM-ICU) values and intensive care unit (ICU) length of stay greater than 24 h were included in our study. We validated our model using 21 quantitative clinical parameters and assessed performance across a range of observation and prediction windows, using different thresholds and applied interpretation techniques. We evaluate our models based on stratified repeated cross-validation using 3 algorithms, namely Logistic Regression, Random Forest, and Bidirectional Long Short-Term Memory (BiLSTM). BiLSTM represents an evolution from recurrent neural network-based Long Short-Term Memory, and with a backward input, preserves information from both past and future. Model performance is measured using Area Under Receiver Operating Characteristic, Area Under Precision Recall Curve, Recall, Precision (Positive Predictive Value), and Negative Predictive Value metrics. Results: We evaluated our results on 16 546 patients (47% female) and 6294 patients (44% female) from eICU-CRD and MIMIC-III databases, respectively. Performance was best in BiLSTM models where, precision and recall changed from 37.52% (95% confidence interval [CI], 36.00%-39.05%) to 17.45 (95% CI, 15.83%-19.08%) and 86.1% (95% CI, 82.49%-89.71%) to 75.58% (95% CI, 68.33%-82.83%), respectively as prediction window increased from 12 to 96 h. After optimizing for higher recall, precision and recall changed from 26.96% (95% CI, 24.99%-28.94%) to 11.34% (95% CI, 10.71%-11.98%) and 93.73% (95% CI, 93.1%-94.37%) to 92.57% (95% CI, 88.19%-96.95%), respectively. Comparable results were obtained in the MIMIC-III cohort. Conclusions: Our model performed comparably to contemporary models using fewer variables. Using techniques like sliding windows, modification of threshold to augment recall and feature ranking for interpretability, we addressed shortcomings of current models.

COVID-19 machine learning model predicts outcomes in older patients from various European countries, between pandemic waves, and in a cohort of Asian, African, and American patients.

BACKGROUND: COVID-19 remains a complex disease in terms of its trajectory and the diversity of outcomes rendering disease management and clinical resource allocation challenging. Varying symptomatology in older patients as well as limitation of clinical scoring systems have created the need for more objective and consistent methods to aid clinical decision making. In this regard, machine learning methods have been shown to enhance prognostication, while improving consistency. However, current machine learning approaches have been limited by lack of generalisation to diverse patient populations, between patients admitted at different waves and small sample sizes. OBJECTIVES: We sought to investigate whether machine learning models, derived on routinely collected clinical data, can generalise well i) between European countries, ii) between European patients admitted at different COVID-19 waves, and iii) between geographically diverse patients, namely whether a model derived on the European patient cohort can be used to predict outcomes of patients admitted to Asian, African and American ICUs. METHODS: We compare Logistic Regression, Feed Forward Neural Network and XGBoost algorithms to analyse data from 3,933 older patients with a confirmed COVID-19 diagnosis in predicting three outcomes, namely: ICU mortality, 30-day mortality and patients at low risk of deterioration. The patients were admitted to ICUs located in 37 countries, between January 11, 2020, and April 27, 2021. RESULTS: The XGBoost model derived on the European cohort and externally validated in cohorts of Asian, African, and American patients, achieved AUC of 0.89 (95% CI 0.89-0.89) in predicting ICU mortality, AUC of 0.86 (95% CI 0.86-0.86) for 30-day mortality prediction and AUC of 0.86 (95% CI 0.86-0.86) in predicting low-risk patients. Similar AUC performance was achieved also when predicting outcomes between European countries and between pandemic waves, while the models showed high calibration quality. Furthermore, saliency analysis showed that FiO2 values of up to 40% do not appear to increase the predicted risk of ICU and 30-day mortality, while PaO2 values of 75 mmHg or lower are associated with a sharp increase in the predicted risk of ICU and 30-day mortality. Lastly, increase in SOFA scores also increase the predicted risk, but only up to a value of 8. Beyond these scores the predicted risk remains consistently high. CONCLUSION: The models captured both the dynamic course of the disease as well as similarities and differences between the diverse patient cohorts, enabling prediction of disease severity, identification of low-risk patients and potentially supporting effective planning of essential clinical resources. TRIAL REGISTRATION NUMBER: NCT04321265.

Disease-Course Adapting Machine Learning Prognostication Models in Elderly Patients Critically Ill With COVID-19: Multicenter Cohort Study With External Validation.

BACKGROUND: The COVID-19 pandemic caused by SARS-CoV-2 is challenging health care systems globally. The disease disproportionately affects the elderly population, both in terms of disease severity and mortality risk. OBJECTIVE: The aim of this study was to evaluate machine learning-based prognostication models for critically ill elderly COVID-19 patients, which dynamically incorporated multifaceted clinical information on evolution of the disease. METHODS: This multicenter cohort study (COVIP study) obtained patient data from 151 intensive care units (ICUs) from 26 countries. Different models based on the Sequential Organ Failure Assessment (SOFA) score, logistic regression (LR), random forest (RF), and extreme gradient boosting (XGB) were derived as baseline models that included admission variables only. We subsequently included clinical events and time-to-event as additional variables to derive the final models using the same algorithms and compared their performance with that of the baseline group. Furthermore, we derived baseline and final models on a European patient cohort, which were externally validated on a non-European cohort that included Asian, African, and US patients. RESULTS: In total, 1432 elderly (≥70 years old) COVID-19-positive patients admitted to an ICU were included for analysis. Of these, 809 (56.49%) patients survived up to 30 days after admission. The average length of stay was 21.6 (SD 18.2) days. Final models that incorporated clinical events and time-to-event information provided superior performance (area under the receiver operating characteristic curve of 0.81; 95% CI 0.804-0.811), with respect to both the baseline models that used admission variables only and conventional ICU prediction models (SOFA score, P<.001). The average precision increased from 0.65 (95% CI 0.650-0.655) to 0.77 (95% CI 0.759-0.770). CONCLUSIONS: Integrating important clinical events and time-to-event information led to a superior accuracy of 30-day mortality prediction compared with models based on the admission information and conventional ICU prediction models. This study shows that machine-learning models provide additional information and may support complex decision-making in critically ill elderly COVID-19 patients. TRIAL REGISTRATION: ClinicalTrials.gov NCT04321265; https://clinicaltrials.gov/ct2/show/NCT04321265.

Sex-specific outcomes and management in critically ill septic patients.

BACKGROUND: Female and male critically ill septic patients might differ with regards to risk distribution, management, and outcomes. We aimed to compare male versus female septic patients in a large collective with regards to baseline risk distribution and outcomes. METHODS: In total, 17,146 patients were included in this analysis, 8781 (51%) male and 8365 (49%) female patients. The primary endpoint was ICU-mortality. Baseline characteristics and data on organ support were documented. Multilevel logistic regression analyses were used to assess sex-specific differences. RESULTS: Female patients had lower SOFA scores (5 ± 5 vs. 6 ± 6; p<0.001) and creatinine (1.20±1.35 vs. 1.40±1.54; p<0.001). In the total cohort, the ICU mortality was 10% and similar between female and male (10% vs. 10%; p = 0.34) patients. The ICU remained similar between sexes after adjustment in model-1 (aOR 1.05 95% CI 0.95-1.16; p = 0.34); model-2 (aOR 0.91 95% CI 0.79-1.05; p = 0.18) and model-3 (aOR 0.93 95% CI 0.80-1.07; p = 0.29). In sensitivity analyses, no major sex-specific differences in mortality could be detected. CONCLUSION: In this study no clinically relevant sex-specific mortality differences could be detected in critically ill septic patients. Possible subtle gender differences could play a minor role in the acute situation due to the severity of the disease in septic patients.

Machine learning predicts mortality in septic patients using only routinely available ABG variables: a multi-centre evaluation.

PURPOSE: To evaluate the application of machine learning methods, specifically Deep Neural Networks (DNN) models for intensive care (ICU) mortality prediction. The aim was to predict mortality within 96 hours after admission to mirror the clinical situation of patient evaluation after an ICU trial, which consists of 24-48 hours of ICU treatment and then "re-triage". The input variables were deliberately restricted to ABG values to maximise real-world practicability. METHODS: We retrospectively evaluated septic patients in the multi-centre eICU dataset as well as single centre MIMIC-III dataset. Included were all patients alive after 48 hours with available data on ABG (n = 3979 and n = 9655 ICU stays for the multi-centre and single centre respectively). The primary endpoint was 96 -h-mortality. RESULTS: The model was developed using long short-term memory (LSTM), a type of DNN designed to learn temporal dependencies between variables. Input variables were all ABG values within the first 48 hours. The SOFA score (AUC of 0.72) was moderately predictive. Logistic regression showed good performance (AUC of 0.82). The best performance was achieved by the LSTM-based model with AUC of 0.88 in the multi-centre study and AUC of 0.85 in the single centre study. CONCLUSIONS: An LSTM-based model could help physicians with the "re-triage" and the decision to restrict treatment in patients with a poor prognosis.

Machine Learning Models Cannot Replace Screening Colonoscopy for the Prediction of Advanced Colorectal Adenoma.

Screening for colorectal cancer (CRC) continues to rely on colonoscopy and/or fecal occult blood testing since other (non-invasive) risk-stratification systems have not yet been implemented into European guidelines. In this study, we evaluate the potential of machine learning (ML) methods to predict advanced adenomas (AAs) in 5862 individuals participating in a screening program for colorectal cancer. Adenomas were diagnosed histologically with an AA being ≥ 1 cm in size or with high-grade dysplasia/villous features being present. Logistic regression (LR) and extreme gradient boosting (XGBoost) algorithms were evaluated for AA prediction. The mean age was 58.7 ± 9.7 years with 2811 males (48.0%), 1404 (24.0%) of whom suffered from obesity (BMI ≥ 30 kg/m²), 871 (14.9%) from diabetes, and 2095 (39.1%) from metabolic syndrome. An adenoma was detected in 1884 (32.1%), as well as AAs in 437 (7.5%). Modelling 36 laboratory parameters, eight clinical parameters, and data on eight food types/dietary patterns, moderate accuracy in predicting AAs with XGBoost and LR (AUC-ROC of 0.65-0.68) could be achieved. Limiting variables to established risk factors for AAs did not significantly improve performance. Moreover, subgroup analyses in subjects without genetic predispositions, in individuals aged 45-80 years, or in gender-specific analyses showed similar results. In conclusion, ML based on point-prevalence laboratory and clinical information does not accurately predict AAs.