Artificial neural networks improve and simplify intensive care mortality prognostication: a national cohort study of 217,289 first-time intensive care unit admissions

Artificial intelligence and machine learning: Definition of terms and current concepts in critical care research

With increasing computing power, artificial intelligence (AI) and machine learning (ML) have prospered, which facilitate the analysis of large datasets, especially those found in critical care. It is important to define these terminologies, to inform a standardized approach to critical care research. This manuscript hopes to clarify these terms with examples from medical literature. Three major components that are required for a successful ML implementation: (i) reliable dataset, (ii) ML algorithm, and (iii) unbiased model evaluation, are discussed. A reliable dataset can be structured or unstructured with limited noise, outliers, and missing values. ML, a subset of AI, is typically focused on supervised or unsupervised learning tasks in which the output is based on inputs and derived from iterative pattern recognition algorithms, while AI is the overall ability of a machine to "think" or mimic human behavior; and to analyze data free from human influence. Even with successful implementation, advanced AI and ML algorithms have faced challenges in adoption into practice, mainly due to their lack of interpretability, which hinders trust, buy-in, and engagement from clinicians. Consequently, traditional algorithms, such as linear and logistic regression, that may have reduced predictive power but are highly interpretable, continue to be widely used.

ICURE: Intensive care unit (ICU) risk evaluation for 30-day mortality. Developing and evaluating a multivariable machine learning prediction model for patients admitted to the general ICU in Sweden

BACKGROUND

A prediction model that estimates mortality at admission to the intensive care unit (ICU) is of potential benefit to both patients and society. Logistic regression models like Simplified Acute Physiology Score 3 (SAPS 3) and APACHE are the traditional ICU mortality prediction models. With the emergence of machine learning (machine learning) and artificial intelligence, new possibilities arise to create prediction models that have the potential to sharpen predictive accuracy and reduce the likelihood of misclassification in the prediction of 30-day mortality.

METHODS

We used the Swedish Intensive Care Registry (SIR) to identify and include all patients ≥18 years of age admitted to general ICUs in Sweden from 2008 to 2022 with SAPS 3 score registered. Only data collected within 1 h of ICU admission was used. We had 153 candidate predictors including baseline characteristics, previous medical conditions, blood works, physiological parameters, cause of admission, and initial treatment. We stratified the data randomly on the outcome variable 30-day mortality and created a training set (80% of data) and a test set (20% of data). We evaluated several hundred prediction models using multiple ML frameworks including random forest, gradient boosting, neural networks, and logistic regression models. Model performance was evaluated by comparing the receiver operator characteristic area under the curve (AUC-ROC). The best performing model was fine-tuned by optimizing hyperparameters. The model's calibration was evaluated by a calibration belt. Ultimately, we simplified the best performing model with the top 1-20 predictors.

RESULTS

We included 296,344 first-time ICU admissions. We found age, Glasgow Coma Scale, creatinine, systolic blood pressure, and pH being the most important predictors. The AUC-ROC was 0.884 in test data using all predictors, specificity 95.2%, sensitivity 47.0%, negative predictive value of 87.9% and positive predictive value of 70.7%. The final model showed excellent calibration. The ICU risk evaluation for 30-day mortality (ICURE) prediction model performed equally well to the SAPS 3 score with only eight variables and improved further with the addition of more variables.

CONCLUSION

The ICURE prediction model predicts 30-day mortality rate at first-time ICU admission superiorly compared to the established SAPS 3 score.

Predicting type 2 diabetes via machine learning integration of multiple omics from human pancreatic islets

Type 2 diabetes (T2D) is the fastest growing non-infectious disease worldwide. Impaired insulin secretion from pancreatic beta-cells is a hallmark of T2D, but the mechanisms behind this defect are insufficiently characterized. Integrating multiple layers of biomedical information, such as different Omics, may allow more accurate understanding of complex diseases such as T2D. Our aim was to explore and use Machine Learning to integrate multiple sources of biological/molecular information (multiOmics), in our case RNA-sequening, DNA methylation, SNP and phenotypic data from islet donors with T2D and non-diabetic controls. We exploited Machine Learning to perform multiOmics integration of DNA methylation, expression, SNPs, and phenotypes from pancreatic islets of 110 individuals, with ~ 30% being T2D cases. DNA methylation was analyzed using Infinium MethylationEPIC array, expression was analyzed using RNA-sequencing, and SNPs were analyzed using HumanOmniExpress arrays. Supervised linear multiOmics integration via DIABLO based on Partial Least Squares (PLS) achieved an accuracy of 91 ± 15% of T2D prediction with an area under the curve of 0.96 ± 0.08 on the test dataset after cross-validation. Biomarkers identified by this multiOmics integration, including SACS and TXNIP DNA methylation, OPRD1 and RHOT1 expression and a SNP annotated to ANO1, provide novel insights into the interplay between different biological mechanisms contributing to T2D. This Machine Learning approach of multiOmics cross-sectional data from human pancreatic islets achieved a promising accuracy of T2D prediction, which may potentially find broad applications in clinical diagnostics. In addition, it delivered novel candidate biomarkers for T2D and links between them across the different Omics.

Use of Artificial Intelligence in Improving Outcomes in Heart Disease: A Scientific Statement From the American Heart Association

A major focus of academia, industry, and global governmental agencies is to develop and apply artificial intelligence and other advanced analytical tools to transform health care delivery. The American Heart Association supports the creation of tools and services that would further the science and practice of precision medicine by enabling more precise approaches to cardiovascular and stroke research, prevention, and care of individuals and populations. Nevertheless, several challenges exist, and few artificial intelligence tools have been shown to improve cardiovascular and stroke care sufficiently to be widely adopted. This scientific statement outlines the current state of the art on the use of artificial intelligence algorithms and data science in the diagnosis, classification, and treatment of cardiovascular disease. It also sets out to advance this mission, focusing on how digital tools and, in particular, artificial intelligence may provide clinical and mechanistic insights, address bias in clinical studies, and facilitate education and implementation science to improve cardiovascular and stroke outcomes. Last, a key objective of this scientific statement is to further the field by identifying best practices, gaps, and challenges for interested stakeholders.

Machine-learning models are superior to severity scoring systems for the prediction of the mortality of critically ill patients in a tertiary medical center

BACKGROUND

Intensive care unit (ICU) mortality prediction helps to guide therapeutic decision making for critically ill patients. Several scoring systems based on statistical techniques have been developed for this purpose. In this study, we developed a machine-learning model to predict patient mortality in the very early stage of ICU admission.

METHODS

This study was performed with data from all patients admitted to the intensive care units of a tertiary medical center in Taiwan from 2009 to 2018. The patients' comorbidities, co-medications, vital signs, and laboratory data on the day of ICU admission were obtained from electronic medical records. We constructed random forest and extreme gradient boosting (XGBoost) models to predict ICU mortality, and compared their performance with that of traditional scoring systems.

RESULTS

Data from 12,377 patients was allocated to training (n = 9901) and testing (n = 2476) datasets. The median patient age was 70.0 years; 9210 (74.41%) patients were under mechanical ventilation in the ICU. The areas under receiver operating characteristic curves for the random forest and XGBoost models (0.876 and 0.880, respectively) were larger than those for the Acute Physiology and Chronic Health Evaluation II score (0.738), Sequential Organ Failure Assessment score (0.747), and Simplified Acute Physiology Score II (0.743). The fraction of inspired oxygen on ICU admission was the most important predictive feature across all models.

CONCLUSION

The XGBoost model most accurately predicted ICU mortality and was superior to traditional scoring systems. Our results highlight the utility of machine learning for ICU mortality prediction in the Asian population.

A simple mortality prediction model for sepsis patients in intensive care

Background

Sepsis is common in the intensive care unit (ICU). Two of the ICU's most widely used mortality prediction models are the Simplified Acute Physiology Score 3 (SAPS-3) and the Sequential Organ Failure Assessment (SOFA) score. We aimed to assess the mortality prediction performance of SAPS-3 and SOFA upon ICU admission for sepsis and find a simpler mortality prediction model for these patients to be used in clinical practice and when conducting studies.

Methods

A retrospective study of adult patients fulfilling the Sepsis-3 criteria admitted to four general ICUs was performed. A simple prognostic model was created using backward stepwise multivariate logistic regression. The area under the curve (AUC) of SAPS-3, SOFA and the simple model was assessed.

Results

One thousand nine hundred eighty four admissions were included. A simple six-parameter model consisting of age, immunosuppression, Glasgow Coma Scale, body temperature, C-reactive protein and bilirubin had an AUC of 0.72 (95% confidence interval (CI) 0.69-0.75) for 30-day mortality, which was non-inferior to SAPS-3 (AUC 0.75, 95% CI 0.72-0.77) (p = 0.071). SOFA had an AUC of 0.67 (95% CI 0.64-0.70) and was inferior to SAPS-3 (p < 0.001) and our simple model (p = 0.0019).

Conclusion

SAPS-3 has a lower prognostic value in sepsis than in the general ICU population. SOFA performs less well than SAPS-3. Our simple six-parameter model predicts mortality just as well as SAPS-3 upon ICU admission for sepsis, allowing the design of simple studies and performance monitoring.

Predictive Modeling Using Artificial Intelligence and Machine Learning Algorithms on Electronic Health Record Data: Advantages and Challenges

The rapid adoption of electronic health record (EHR) systems in US hospitals from 2008 to 2014 produced novel data elements for analysis. Concurrent innovations in computing architecture and machine learning (ML) algorithms have made rapid consumption of health data feasible and a powerful engine for clinical innovation. In critical care research, the net convergence of these trends has resulted in an exponential increase in outcome prediction research. In the following article, we explore the history of outcome prediction in the intensive care unit (ICU), the growing use of EHR data, and the rise of artificial intelligence and ML (AI) in critical care.

Machine learning-based prediction of in-ICU mortality in pneumonia patients

Conventional severity-of-illness scoring systems have shown suboptimal performance for predicting in-intensive care unit (ICU) mortality in patients with severe pneumonia. This study aimed to develop and validate machine learning (ML) models for mortality prediction in patients with severe pneumonia. This retrospective study evaluated patients admitted to the ICU for severe pneumonia between January 2016 and December 2021. The predictive performance was analyzed by comparing the area under the receiver operating characteristic curve (AU-ROC) of ML models to that of conventional severity-of-illness scoring systems. Three ML models were evaluated: (1) logistic regression with L2 regularization, (2) gradient-boosted decision tree (LightGBM), and (3) multilayer perceptron (MLP). Among the 816 pneumonia patients included, 223 (27.3%) patients died. All ML models significantly outperformed the Simplified Acute Physiology Score II (AU-ROC: 0.650 [0.584-0.716] vs 0.820 [0.771-0.869] for logistic regression vs 0.827 [0.777-0.876] for LightGBM 0.838 [0.791-0.884] for MLP; P < 0.001). In the analysis for NRI, the LightGBM and MLP models showed superior reclassification compared with the logistic regression model in predicting in-ICU mortality in all length of stay in the ICU subgroups; all age subgroups; all subgroups with any APACHE II score, PaO₂/FiO₂ ratio < 200; all subgroups with or without history of respiratory disease; with or without history of CVA or dementia; treatment with mechanical ventilation, and use of inotropic agents. In conclusion, the ML models have excellent performance in predicting in-ICU mortality in patients with severe pneumonia. Moreover, this study highlights the potential advantages of selecting individual ML models for predicting in-ICU mortality in different subgroups.

Smart Greenhouse Based on ANN and IOT

The effective exploitation of smart technology in applications helps farmers make better decisions without increasing costs. Agricultural Research Centers (ARCs) are continually updating and producing new datasets from applied research, so the smart model should dynamically address all surrounding agricultural variables and improve its expertise from all available datasets. This research concentrates on sustainable agriculture using Adaptive Particle Swarm Optimization (PSO) and Artificial Neural Networks (ANNs). Therefore, if a new related dataset is created, this new incoming dataset is merged with the existing dataset. The proposed PSO then bypasses the summarization of the dataset. It deletes the least essential and speculative records and keeps the records that are the most influential in the classification process. The summarized dataset is interposed in the training process without re-establishing the system again by modifying the classical ANN. The proposed ANN comprises an adaptive input layer and an adaptive output layer to handle the process of continuously updating the datasets. A comparative study between the proposed adaptive PSO-ANN and other known and used methods on different datasets has been applied. The results prove the quality of the proposed Adaptive PSO-ANN from various standard measurements. The proposed PSO-ANN achieved an accuracy of 94.8%, precision of 91.15%, recall of 97.93%, and F1-score of 94.42%. The smart olive cultivation case study is accomplished with the proposed adaptive PSO-ANN and technological tools from the Internet of Things (IoT). The advanced tools from IoT technology are established and analyzed to control all the required procedures of olive cultivation. This case study addresses the necessary fertilizers and irrigation water to adapt to the changes in climate. Empirical results show that smart olive cultivation using the proposed adaptive PSO-ANN and IoT has high quality and efficiency. The quality and efficiency are measured by diversified metrics such as crop production and consumed water, which confirm the success of the proposed smart olive agriculture method.

Outcome Prediction in Critically-Ill Patients with Venous Thromboembolism and/or Cancer Using Machine Learning Algorithms: External Validation and Comparison with Scoring Systems

Intensive care unit (ICU) patients with venous thromboembolism (VTE) and/or cancer suffer from high mortality rates. Mortality prediction in the ICU has been a major medical challenge for which several scoring systems exist but lack in specificity. This study focuses on two target groups, namely patients with thrombosis or cancer. The main goal is to develop and validate interpretable machine learning (ML) models to predict early and late mortality, while exploiting all available data stored in the medical record. To this end, retrospective data from two freely accessible databases, MIMIC-III and eICU, were used. Well-established ML algorithms were implemented utilizing automated and purposely built ML frameworks for addressing class imbalance. Prediction of early mortality showed excellent performance in both disease categories, in terms of the area under the receiver operating characteristic curve (AUC–ROC): VTE-MIMIC-III 0.93, eICU 0.87, cancer-MIMIC-III 0.94. On the other hand, late mortality prediction showed lower performance, i.e., AUC–ROC: VTE 0.82, cancer 0.74–0.88. The predictive model of early mortality developed from 1651 VTE patients (MIMIC-III) ended up with a signature of 35 features and was externally validated in 2659 patients from the eICU dataset. Our model outperformed traditional scoring systems in predicting early as well as late mortality. Novel biomarkers, such as red cell distribution width, were identified.

Demystifying the Black Box: The Importance of Interpretability of Predictive Models in Neurocritical Care

Neurocritical care patients are a complex patient population, and to aid clinical decision-making, many models and scoring systems have previously been developed. More recently, techniques from the field of machine learning have been applied to neurocritical care patient data to develop models with high levels of predictive accuracy. However, although these recent models appear clinically promising, their interpretability has often not been considered and they tend to be black box models, making it extremely difficult to understand how the model came to its conclusion. Interpretable machine learning methods have the potential to provide the means to overcome some of these issues but are largely unexplored within the neurocritical care domain. This article examines existing models used in neurocritical care from the perspective of interpretability. Further, the use of interpretable machine learning will be explored, in particular the potential benefits and drawbacks that the techniques may have when applied to neurocritical care data. Finding a solution to the lack of model explanation, transparency, and accountability is important because these issues have the potential to contribute to model trust and clinical acceptance, and, increasingly, regulation is stipulating a right to explanation for decisions made by models and algorithms. To ensure that the prospective gains from sophisticated predictive models to neurocritical care provision can be realized, it is imperative that interpretability of these models is fully considered.

Advancements of Artificial Intelligence in Liver-Associated Diseases and Surgery

Background and Objectives: The advancement of artificial intelligence (AI) based technologies in medicine is progressing rapidly, but the majority of its real-world applications has not been implemented. The establishment of an accurate diagnosis with treatment has now transitioned into an artificial intelligence era, which has continued to provide an amplified understanding of liver cancer as a disease and helped to proceed better with the method of procurement. This article focuses on reviewing the AI in liver-associated diseases and surgical procedures, highlighting its development, use, and related counterparts. Materials and Methods: We searched for articles regarding AI in liver-related ailments and surgery, using the keywords (mentioned below) on PubMed, Google Scholar, Scopus, MEDLINE, and Cochrane Library. Choosing only the common studies suggested by these libraries, we segregated the matter based on disease. Finally, we compiled the essence of these articles under the various sub-headings. Results: After thorough review of articles, it was observed that there was a surge in the occurrence of liver-related surgeries, diagnoses, and treatments. Parallelly, advanced computer technologies governed by AI continue to prove their efficacy in the accurate screening, analysis, prediction, treatment, and recuperation of liver-related cases. Conclusions: The continual developments and high-order precision of AI is expanding its roots in all directions of applications. Despite being novel and lacking research, AI has shown its intrinsic worth for procedures in liver surgery while providing enhanced healing opportunities and personalized treatment for liver surgery patients.

Understanding basic principles of Artificial Intelligence: a practical guide for intensivists

BACKGROUND AND AIM

Artificial intelligence was born to allow computers to learn and control their environment, trying to imitate the human brain structure by simulating its biological evolution. Artificial intelligence makes it possible to analyze large amounts of data (big data) in real-time, providing forecasts that can support the clinician's decisions. This scenario can include diagnosis, prognosis, and treatment in anesthesiology, intensive care medicine, and pain medicine. Machine Learning is a subcategory of AI. It is based on algorithms trained for decisions making that automatically learn and recognize patterns from data. This article aims to offer an overview of the potential application of AI in anesthesiology and analyzes the operating principles of machine learning Every Machine Learning pathway starts from task definition and ends in model application.

CONCLUSIONS

High-performance characteristics and strict quality controls are needed during its progress. During this process, different measures can be identified (pre-processing, exploratory data analysis, model selection, model processing and evaluation). For inexperienced operators, the process can be facilitated by ad hoc tools for data engineering, machine learning, and analytics.

Artificial intelligence forecasting mortality at an intensive care unit and comparison to a logistic regression system

Objective

To explore an artificial intelligence approach based on gradient-boosted decision trees for prediction of all-cause mortality at an intensive care unit, comparing its performance to a recent logistic regression system in the literature, and a logistic regression model built on the same platform.

Methods

A gradient-boosted decision trees model and a logistic regression model were trained and tested with the Medical Information Mart for Intensive Care database. The 1-hour resolution physiological measurements of adult patients, collected during 5 hours in the intensive care unit, consisted of eight routine clinical parameters. The study addressed how the models learn to categorize patients to predict intensive care unit mortality or survival within 12 hours. The performance was evaluated with accuracy statistics and the area under the Receiver Operating Characteristic curve.

Results

The gradient-boosted trees yielded an area under the Receiver Operating Characteristic curve of 0.89, compared to 0.806 for the logistic regression. The accuracy was 0.814 for the gradient-boosted trees, compared to 0.782 for the logistic regression. The diagnostic odds ratio was 17.823 for the gradient-boosted trees, compared to 9.254 for the logistic regression. The Cohen’s kappa, F-measure, Matthews correlation coefficient, and markedness were higher for the gradient-boosted trees.

Conclusion

The discriminatory power of the gradient-boosted trees was excellent. The gradient-boosted trees outperformed the logistic regression regarding intensive care unit mortality prediction. The high diagnostic odds ratio and markedness values for the gradient-boosted trees are important in the context of the studied unbalanced dataset.

Machine Learning Models for Survival and Neurological Outcome Prediction of Out-of-Hospital Cardiac Arrest Patients

Background

Out-of-hospital cardiac arrest (OHCA) is a major health problem worldwide, and neurologic injury remains the leading cause of morbidity and mortality among survivors of OHCA. The purpose of this study was to investigate whether a machine learning algorithm could detect complex dependencies between clinical variables in emergency departments in OHCA survivors and perform reliable predictions of favorable neurologic outcomes.

Methods

This study included adults (≥18 years of age) with a sustained return of spontaneous circulation after successful resuscitation from OHCA between 1 January 2004 and 31 December 2014. We applied three machine learning algorithms, including logistic regression (LR), support vector machine (SVM), and extreme gradient boosting (XGB). The primary outcome was a favorable neurological outcome at hospital discharge, defined as a Glasgow-Pittsburgh cerebral performance category of 1 to 2. The secondary outcome was a 30-day survival rate and survival-to-discharge rate.

Results

The final analysis included 1071 participants from the study period. For neurologic outcome prediction, the area under the receiver operating curve (AUC) was 0.819, 0.771, and 0.956 in LR, SVM, and XGB, respectively. The sensitivity and specificity were 0.875 and 0.751 in LR, 0.687 and 0.793 in SVM, and 0.875 and 0.904 in XGB. The AUC was 0.766 and 0.732 in LR, 0.749 and 0.725 in SVM, and 0.866 and 0.831 in XGB, for survival-to-discharge and 30-day survival, respectively.

Conclusions

Prognostic models trained with ML technique showed appropriate calibration and high discrimination for survival and neurologic outcome of OHCA without using prehospital data, with XGB exhibiting the best performance.

Addition of admission lactate levels to Baux score improves mortality prediction in severe burns

Risk adjustment and mortality prediction models are central in optimising care and for benchmarking purposes. In the burn setting, the Baux score and its derivatives have been the mainstay for predictions of mortality from burns. Other well-known measures to predict mortality stem from the ICU setting, where, for example, the Simplified Acute Physiology Score (SAPS 3) models have been found to be instrumental. Other attempts to further improve the prediction of outcome have been based on the following variables at admission: Sequential Organ Failure Assessment (_aSOFA) score, determinations of _aLactate or Neutrophil to Lymphocyte Ratio (_aNLR). The aim of the present study was to examine if estimated mortality rate (EMR, SAPS 3), _aSOFA, _aLactate, and _aNLR can, either alone or in conjunction with the others, improve the mortality prediction beyond that of the effects of age and percentage total body surface area (TBSA%) burned among patients with severe burns who need critical care. This is a retrospective, explorative, single centre, registry study based on prospectively gathered data. The study included 222 patients with median (25th–75th centiles) age of 55.0 (38.0 to 69.0) years, TBSA% burned was 24.5 (13.0 to 37.2) and crude mortality was 17%. As anticipated highest predicting power was obtained with age and TBSA% with an AUC at 0.906 (95% CI 0.857 to 0.955) as compared with EMR, _aSOFA, _aLactate and _aNLR. The largest effect was seen thereafter by adding _aLactate to the model, increasing AUC to 0.938 (0.898 to 0.979) (p < 0.001). Whereafter, adding EMR, _aSOFA, and _aNLR, separately or in combinations, only marginally improved the prediction power. This study shows that the prediction model with age and TBSA% may be improved by adding _aLactate, despite the fact that _aLactate levels were only moderately increased. Thereafter, adding EMR, _aSOFA or _aNLR only marginally affected the mortality prediction.

Prediction of Hemorrhagic Transformation after Ischemic Stroke Using Machine Learning

Hemorrhagic transformation (HT) is one of the leading causes of a poor prognostic marker after acute ischemic stroke (AIS). We compared the performances of the several machine learning (ML) algorithms to predict HT after AIS using only structured data. A total of 2028 patients with AIS, who were admitted within seven days of symptoms onset, were included in this analysis. HT was defined based on the criteria of the European Co-operative Acute Stroke Study-II trial. The whole dataset was randomly divided into a training and a test dataset with a 7:3 ratio. Binary logistic regression, support vector machine, extreme gradient boosting, and artificial neural network (ANN) algorithms were used to assess the performance of predicting the HT occurrence after AIS. Five-fold cross validation and a grid search technique were used to optimize the hyperparameters of each ML model, which had its performance measured by the area under the receiver operating characteristic (AUROC) curve. Among the included AIS patients, the mean age and number of male subjects were 69.6 years and 1183 (58.3%), respectively. HT was observed in 318 subjects (15.7%). There were no significant differences in corresponding variables between the training and test dataset. Among all the ML algorithms, the ANN algorithm showed the best performance in terms of predicting the occurrence of HT in our dataset (0.844). Feature scaling including standardization and normalization, and the resampling strategy showed no additional improvement of the ANN's performance. The ANN-based prediction of HT after AIS showed better performance than the conventional ML algorithms. Deep learning may be used to predict important outcomes for structured data-based prediction.

Circulating dipeptidyl peptidase 3 on intensive care unit admission is a predictor of organ dysfunction and mortality

Background

Our aim was to investigate the prognostic potential of circulating dipeptidyl peptidase 3 (cDPP3) to predict mortality and development of organ dysfunction in a mixed intensive care unit (ICU) population, and for this reason, we analysed prospectively collected admission blood samples from adult ICU patients at four Swedish hospitals. Blood samples were stored in a biobank for later batch analysis. The association of cDPP3 levels with 30-day mortality and Sequential Organ Failure Assessment (SOFA) scores on day two was investigated before and after adjustment for the simplified acute physiology score III (SAPS-3), using multivariable (ordinal) logistic regression. The predictive power of cDPP3 was assessed using the area under the receiver operating characteristic curve (AUROC).

Results

Of 1978 included consecutive patients in 1 year (2016), 632 fulfilled the sepsis 3-criteria, 190 were admitted after cardiac arrest, and 157 because of trauma. Admission cDPP3 was independently (of SAPS-3) associated with 30-day mortality with odds ratios of 1.45 (95% confidence interval (CI) 1.28–1.64) in the entire ICU population, 1.30 (95% CI 1.08–1.57) in the sepsis subgroup and 2.28 (95% CI 1.50–3.62) in cardiac arrest. For trauma, there was no clear association. Circulating DPP3 alone was a moderate predictor of 30-day mortality with AUROCs of 0.68, 0.62, and 0.72 in the entire group, the sepsis subgroup, and the cardiac arrest subgroup, respectively. By adding cDPP3 to SAPS-3, AUROC improved for the entire group, the sepsis subgroup, and the cardiac arrest subgroup (p = 0.023).

Conclusion

Circulating DPP3 on admission is a SAPS-3 independent prognostic factor of day-two organ dysfunction and 30-day mortality in a mixed ICU population and needs further evaluation.

Supplementary Information

The online version contains supplementary material available at (10.1186/s40560-021-00561-9).

[Machine learning in anesthesiology]

The application of artificial intelligence (AI) is currently changing very different areas of life. Artificial intelligence involves the emulation of human behavior with the aid of methods from mathematics and informatics. Machine learning (ML) represents a subdivision of AI. Algorithms for ML have the potential to optimize patient care, in that they can be utilized in a supportive way in personalized medicine, decision making and risk prediction. Although the majority of the applications in medicine are still limited to data analysis and research, it is certain that ML will become increasingly more important in scientific and clinical aspects in this supportive function. Therefore, it is necessary for clinicians to have at least a basic understanding of the functional principles, strengths and weaknesses of ML.

Predicting neurological outcome after out-of-hospital cardiac arrest with cumulative information; development and internal validation of an artificial neural network algorithm

Background

Prognostication of neurological outcome in patients who remain comatose after cardiac arrest resuscitation is complex. Clinical variables, as well as biomarkers of brain injury, cardiac injury, and systemic inflammation, all yield some prognostic value. We hypothesised that cumulative information obtained during the first three days of intensive care could produce a reliable model for predicting neurological outcome following out-of-hospital cardiac arrest (OHCA) using artificial neural network (ANN) with and without biomarkers.

Methods

We performed a post hoc analysis of 932 patients from the Target Temperature Management trial. We focused on comatose patients at 24, 48, and 72 h post-cardiac arrest and excluded patients who were awake or deceased at these time points. 80% of the patients were allocated for model development (training set) and 20% for internal validation (test set). To investigate the prognostic potential of different levels of biomarkers (clinically available and research-grade), patients’ background information, and intensive care observation and treatment, we created three models for each time point: (1) clinical variables, (2) adding clinically accessible biomarkers, e.g., neuron-specific enolase (NSE) and (3) adding research-grade biomarkers, e.g., neurofilament light (NFL). Patient outcome was the dichotomised Cerebral Performance Category (CPC) at six months; a good outcome was defined as CPC 1–2 whilst a poor outcome was defined as CPC 3–5. The area under the receiver operating characteristic curve (AUROC) was calculated for all test sets.

Results

AUROC remained below 90% when using only clinical variables throughout the first three days in the ICU. Adding clinically accessible biomarkers such as NSE, AUROC increased from 82 to 94% (p < 0.01). The prognostic accuracy remained excellent from day 1 to day 3 with an AUROC at approximately 95% when adding research-grade biomarkers. The models which included NSE after 72 h and NFL on any of the three days had a low risk of false-positive predictions while retaining a low number of false-negative predictions.

Conclusions

In this exploratory study, ANNs provided good to excellent prognostic accuracy in predicting neurological outcome in comatose patients post OHCA. The models which included NSE after 72 h and NFL on all days showed promising prognostic performance.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-021-03505-9.

Development and validation of the predictive risk of death model for adult patients admitted to intensive care units in Japan: an approach to improve the accuracy of healthcare quality measures

BACKGROUND

The Acute Physiology and Chronic Health Evaluation (APACHE) III-j model is widely used to predict mortality in Japanese intensive care units (ICUs). Although the model's discrimination is excellent, its calibration is poor. APACHE III-j overestimates the risk of death, making its evaluation of healthcare quality inaccurate. This study aimed to improve the calibration of the model and develop a Japan Risk of Death (JROD) model for benchmarking purposes.

METHODS

A retrospective analysis was conducted using a national clinical registry of ICU patients in Japan. Adult patients admitted to an ICU between April 1, 2018, and March 31, 2019, were included. The APACHE III-j model was recalibrated with the following models: Model 1, predicting mortality with an offset variable for the linear predictor of the APACHE III-j model using a generalized linear model; model 2, predicting mortality with the linear predictor of the APACHE III-j model using a generalized linear model; and model 3, predicting mortality with the linear predictor of the APACHE III-j model using a hierarchical generalized additive model. Model performance was assessed with the area under the receiver operating characteristic curve (AUROC), the Brier score, and the modified Hosmer-Lemeshow test. To confirm model applicability to evaluating quality of care, funnel plots of the standardized mortality ratio and exponentially weighted moving average (EWMA) charts for mortality were drawn.

RESULTS

In total, 33,557 patients from 44 ICUs were included in the study population. ICU mortality was 3.8%, and hospital mortality was 8.1%. The AUROC, Brier score, and modified Hosmer-Lemeshow p value of the original model and models 1, 2, and 3 were 0.915, 0.062, and < .001; 0.915, 0.047, and < .001; 0.915, 0.047, and .002; and 0.917, 0.047, and .84, respectively. Except for model 3, the funnel plots showed overdispersion. The validity of the EWMA charts for the recalibrated models was determined by visual inspection.

CONCLUSIONS

Model 3 showed good performance and can be adopted as the JROD model for monitoring quality of care in an ICU, although further investigation of the clinical validity of outlier detection is required. This update method may also be useful in other settings.

Machine learning-based patient classification system for adult patients in intensive care units: A cross-sectional study

AIM

This study aimed to develop a patient classification system that stratifies patients admitted to the intensive care unit based on their disease severity and care needs.

BACKGROUND

Classifying patients into homogenous groups based on clinical characteristics can optimize nursing care. However, an objective method for determining such groups remains unclear.

METHODS

Predictors representing disease severity and nursing workload were considered. Patients were clustered into subgroups with different characteristics based on the results of a clustering algorithm. A patient classification system was developed using a partial least squares regression model.

RESULTS

Data of 300 patients were analysed. Cluster analysis identified three subgroups of critically patients with different levels of clinical trajectories. Except for blood potassium levels (p = .29), the subgroups were significantly different according to disease severity and nursing workload. The predicted value ranges of the regression model for Classes A, B and C were <1.44, 1.44-2.03 and >2.03. The model was shown to have good fit and satisfactory prediction efficiency using 200 permutation tests.

CONCLUSIONS

Classifying patients based on disease severity and care needs enables the development of tailored nursing management for each subgroup.

IMPLICATIONS FOR NURSING MANAGEMENT

The patient classification system can help nurse managers identify homogeneous patient groups and further improve the management of critically ill patients.

Development of a machine learning model for predicting pediatric mortality in the early stages of intensive care unit admission

The aim of this study was to develop a predictive model of pediatric mortality in the early stages of intensive care unit (ICU) admission using machine learning. Patients less than 18 years old who were admitted to ICUs at four tertiary referral hospitals were enrolled. Three hospitals were designated as the derivation cohort for machine learning model development and internal validation, and the other hospital was designated as the validation cohort for external validation. We developed a random forest (RF) model that predicts pediatric mortality within 72 h of ICU admission, evaluated its performance, and compared it with the Pediatric Index of Mortality 3 (PIM 3). The area under the receiver operating characteristic curve (AUROC) of RF model was 0.942 (95% confidence interval [CI] = 0.912–0.972) in the derivation cohort and 0.906 (95% CI = 0.900–0.912) in the validation cohort. In contrast, the AUROC of PIM 3 was 0.892 (95% CI = 0.878–0.906) in the derivation cohort and 0.845 (95% CI = 0.817–0.873) in the validation cohort. The RF model in our study showed improved predictive performance in terms of both internal and external validation and was superior even when compared to PIM 3.

Mortality risk stratification using artificial intelligence-augmented electrocardiogram in cardiac intensive care unit patients

AIMS

An artificial intelligence-augmented electrocardiogram (AI-ECG) algorithm can identify left ventricular systolic dysfunction (LVSD). We sought to determine whether this AI-ECG algorithm could stratify mortality risk in cardiac intensive care unit (CICU) patients, independent of the presence of LVSD by transthoracic echocardiography (TTE).

METHODS AND RESULTS

We included 11 266 unique Mayo Clinic CICU patients admitted from 2007 to 2018 who underwent AI-ECG after CICU admission. Left ventricular ejection fraction (LVEF) data were extracted for patients with a TTE during hospitalization. Hospital mortality was analysed using multivariable logistic regression. Mean age was 68 ± 15 years, including 37% females. Higher AI-ECG probability of LVSD remained associated with higher hospital mortality [adjusted odds ratio (OR) 1.05 per 0.1 higher, 95% confidence interval (CI) 1.02-1.08, P = 0.003] after adjustment for LVEF, which itself was inversely related with the risk of hospital mortality (adjusted OR 0.96 per 5% higher, 95% CI 0.93-0.99, P = 0.02). Patients with available LVEF data (n = 8242) were divided based on the presence of predicted (by AI-ECG) vs. observed (by TTE) LVSD (defined as LVEF ≤ 35%), using TTE as the gold standard. A stepwise increase in hospital mortality was observed for patients with a true negative, false positive, false negative, and true positive AI-ECG.

CONCLUSION

The AI-ECG prediction of LVSD is associated with hospital mortality in CICU patients, affording risk stratification in addition to that provided by echocardiographic LVEF. Our results emphasize the prognostic value of electrocardiographic patterns reflecting underlying myocardial disease that are recognized by the AI-ECG.

Applications of machine learning in acute care research

Artificial intelligence has been successfully applied to numerous health care and non-health care-related applications and its use in emergency medicine has been expanding. Among its advantages are its speed in decision making and the opportunity for rapid, actionable deduction from unstructured data with that increases with access to larger volumes of data. Artificial intelligence algorithms are currently being applied to enable faster prognosis and diagnosis of diseases and to improve patient outcomes.1,2 Despite the successful application of artificial intelligence, it is still fraught with limitations and "unknowns" pertaining to the fact that a model's accuracy is dependent on the amount of information available for training the model, and the understanding of the complexity presented by current artificial intelligence and machine learning algorithms is often limited in many individuals outside of those involved in the field. This paper reviews the applications of artificial intelligence and machine learning to acute care research and highlights commonly used machine learning techniques, limitations, and potential future applications.

Comparative analysis of three data mining techniques in diagnosis of lung cancer

There is a lot of abnormal information in the development of lung cancer, and how to extract useful knowledge is urgent from massive information. Data mining technology has become a popular tool for medical classification and prediction. However, each technology has its advantage and disadvantage, and several data mining methods will be applied to conduct the in-depth analysis step by step. And the prediction results of different models are compared. A total of 180 lung cancer patients and 243 lung benign individuals were collected from the First Affiliated Hospital of Zhengzhou University from October 2014 to March 2016, and the prediction models based on epidemiological data, clinical features and tumor markers were developed by artificial neural network (ANN), decision tree C5.0 and support vector machine (SVM). The results showed that there were significant differences between the lung cancer group and the lung benign group in terms of seven tumor markers and 10 epidemiological and clinical indicators. The accuracy rates of ANN, C5.0 and SVM were 76.47, 89.92 and 85.71%, respectively. The results of receiver operating characteristic curve (ROC) curve revealed the area under the ROC curve (AUC) of ANN was 0.811 (0.770-0.847), the AUC of C5.0 was 0.897 (0.864-0.924) and the AUC of SVM was 0.878 (0.843-0.908). It was shown that the decision tree C5.0 model has the least error rate and highest accuracy, and it could be used to diagnose lung cancer.

Artificial neural networks improve early outcome prediction and risk classification in out-of-hospital cardiac arrest patients admitted to intensive care

Background

Pre-hospital circumstances, cardiac arrest characteristics, comorbidities and clinical status on admission are strongly associated with outcome after out-of-hospital cardiac arrest (OHCA). Early prediction of outcome may inform prognosis, tailor therapy and help in interpreting the intervention effect in heterogenous clinical trials. This study aimed to create a model for early prediction of outcome by artificial neural networks (ANN) and use this model to investigate intervention effects on classes of illness severity in cardiac arrest patients treated with targeted temperature management (TTM).

Methods

Using the cohort of the TTM trial, we performed a post hoc analysis of 932 unconscious patients from 36 centres with OHCA of a presumed cardiac cause. The patient outcome was the functional outcome, including survival at 180 days follow-up using a dichotomised Cerebral Performance Category (CPC) scale with good functional outcome defined as CPC 1–2 and poor functional outcome defined as CPC 3–5. Outcome prediction and severity class assignment were performed using a supervised machine learning model based on ANN.

Results

The outcome was predicted with an area under the receiver operating characteristic curve (AUC) of 0.891 using 54 clinical variables available on admission to hospital, categorised as background, pre-hospital and admission data. Corresponding models using background, pre-hospital or admission variables separately had inferior prediction performance. When comparing the ANN model with a logistic regression-based model on the same cohort, the ANN model performed significantly better (p = 0.029). A simplified ANN model showed promising performance with an AUC above 0.852 when using three variables only: age, time to ROSC and first monitored rhythm. The ANN-stratified analyses showed similar intervention effect of TTM to 33 °C or 36 °C in predefined classes with different risk of a poor outcome.

Conclusion

A supervised machine learning model using ANN predicted neurological recovery, including survival excellently, and outperformed a conventional model based on logistic regression. Among the data available at the time of hospitalisation, factors related to the pre-hospital setting carried most information. ANN may be used to stratify a heterogenous trial population in risk classes and help determine intervention effects across subgroups.

Artificial Intelligence in the Intensive Care Unit

This article is one of ten reviews selected from the Annual Update in Intensive Care and Emergency Medicine 2020. Other selected articles can be found online at https://www.biomedcentral.com/collections/annualupdate2020. Further information about the Annual Update in Intensive Care and Emergency Medicine is available from http://www.springer.com/series/8901.

C-reactive protein as a prognostic factor in intensive care admissions for sepsis: A Swedish multicenter study

PURPOSE

C-reactive protein (CRP) is not included in the major intensive care unit (ICU) prognostic tools such as the Simplified Acute Physiology Score (SAPS). We assessed CRP on ICU admission as a SAPS-3 independent risk marker for short-term mortality and length of stay (LOS) in ICU patients with sepsis.

MATERIALS AND METHODS

Adult ICU admissions satisfying the Sepsis-3 criteria to four southern Swedish hospitals were retrospectively identified and divided into a low CRP group (<100 mg/L) and a high CRP group (>100 mg/L) based on the admission CRP level. The standardized mortality ratio (SMR) was calculated.

RESULTS

A total of 851 admissions were included. The SMR was higher in the high CRP group (0.85 vs. 0.67, P = .001 in the whole sepsis group and 0.85 vs. 0.59, P = .003 in the culture-positive subgroup). The CRP levels also correlated with ICU and hospital LOS in survivors (P < .001 and P = .002), again independent of SAPS-3.

CONCLUSION

An admission CRP level >100 mg/L is associated with an increased risk of ICU and 30-day mortality as well as prolonged LOS in survivors, irrespective of morbidity measured with SAPS-3. Thus, CRP may be a simple, early marker for prognosis in ICU admissions for sepsis.