Machine Learning and Decision Support in Critical Care

Identification of data elements for blood gas analysis dataset: a base for developing registries and artificial intelligence-based systems

Background

One of the challenging decision-making tasks in healthcare centers is the interpretation of blood gas tests. One of the most effective assisting approaches for the interpretation of blood gas analysis (BGA) can be artificial intelligence (AI)-based decision support systems. A primary step to develop intelligent systems is to determine information requirements and automated data input for the secondary analyses. Datasets can help the automated data input from dispersed information systems. Therefore, the current study aimed to identify the data elements required for supporting BGA as a dataset.

Materials and methods

This cross-sectional descriptive study was conducted in Nemazee Hospital, Shiraz, Iran. A combination of literature review, experts’ consensus, and the Delphi technique was used to develop the dataset. A review of the literature was performed on electronic databases to find the dataset for BGA. An expert panel was formed to discuss on, add, or remove the data elements extracted through searching the literature. Delphi technique was used to reach consensus and validate the draft dataset.

Results

The data elements of the BGA dataset were categorized into ten categories, namely personal information, admission details, present illnesses, past medical history, social status, physical examination, paraclinical investigation, blood gas parameter, sequential organ failure assessment (SOFA) score, and sampling technique errors. Overall, 313 data elements, including 172 mandatory and 141 optional data elements were confirmed by the experts for being included in the dataset.

Conclusions

We proposed a dataset as a base for registries and AI-based systems to assist BGA. It helps the storage of accurate and comprehensive data, as well as integrating them with other information systems. As a result, high-quality care is provided and clinical decision-making is improved.

Developing, implementing and governing artificial intelligence in medicine: a step-by-step approach to prevent an artificial intelligence winter

Objective

Although the role of artificial intelligence (AI) in medicine is increasingly studied, most patients do not benefit because the majority of AI models remain in the testing and prototyping environment. The development and implementation trajectory of clinical AI models are complex and a structured overview is missing. We therefore propose a step-by-step overview to enhance clinicians’ understanding and to promote quality of medical AI research.

Methods

We summarised key elements (such as current guidelines, challenges, regulatory documents and good practices) that are needed to develop and safely implement AI in medicine.

Conclusion

This overview complements other frameworks in a way that it is accessible to stakeholders without prior AI knowledge and as such provides a step-by-step approach incorporating all the key elements and current guidelines that are essential for implementation, and can thereby help to move AI from bytes to bedside.

Using Machine Learning to Predict Complications in Pregnancy: A Systematic Review

Introduction: Artificial intelligence is widely used in medical field, and machine learning has been increasingly used in health care, prediction, and diagnosis and as a method of determining priority. Machine learning methods have been features of several tools in the fields of obstetrics and childcare. This present review aims to summarize the machine learning techniques to predict perinatal complications.

Objective: To identify the applicability and performance of machine learning methods used to identify pregnancy complications.

Methods: A total of 98 articles were obtained with the keywords “machine learning,” “deep learning,” “artificial intelligence,” and accordingly as they related to perinatal complications (“complications in pregnancy,” “pregnancy complications”) from three scientific databases: PubMed, Scopus, and Web of Science. These were managed on the Mendeley platform and classified using the PRISMA method.

Results: A total of 31 articles were selected after elimination according to inclusion and exclusion criteria. The features used to predict perinatal complications were primarily electronic medical records (48%), medical images (29%), and biological markers (19%), while 4% were based on other types of features, such as sensors and fetal heart rate. The main perinatal complications considered in the application of machine learning thus far are pre-eclampsia and prematurity. In the 31 studies, a total of sixteen complications were predicted. The main precision metric used is the AUC. The machine learning methods with the best results were the prediction of prematurity from medical images using the support vector machine technique, with an accuracy of 95.7%, and the prediction of neonatal mortality with the XGBoost technique, with 99.7% accuracy.

Conclusion: It is important to continue promoting this area of research and promote solutions with multicenter clinical applicability through machine learning to reduce perinatal complications. This systematic review contributes significantly to the specialized literature on artificial intelligence and women’s health.

Prediction model of acute kidney injury induced by cisplatin in older adults using a machine learning algorithm

BACKGROUND

Early detection and prediction of cisplatin-induced acute kidney injury (Cis-AKI) are essential for the management of patients on chemotherapy with cisplatin. This study aimed to evaluate the performance of a prediction model for Cis-AKI.

METHODS

Japanese patients, who received cisplatin as the first-line chemotherapy at Fujita Health University Hospital, were enrolled in the study. The main metrics for evaluating the machine learning model were the area under the curve (AUC), accuracy, precision, recall, and F-measure. In addition, the rank of contribution as a predictive factor of Cis-AKI was determined by machine learning.

RESULTS

A total of 1,014 and 226 patients were assigned to the development and validation data groups, respectively. The current prediction model showed the highest performance in patients 65 years old and above (AUC: 0.78, accuracy: 0.77, precision: 0.38, recall: 0.70, F-measure: 0.49). The maximum daily cisplatin dose and serum albumin levels contributed the most to the prediction of Cis-AKI.

CONCLUSION

Our prediction model for Cis-AKI performed effectively in older patients.

Optimising Energy Management in Hybrid Microgrids

This article deals with the optimization of the operation of hybrid microgrids. Both the problem of controlling the management of load sharing between the different generators and energy storage and possible solutions for the integration of the microgrid into the electricity market will be discussed. Solar and wind energy as well as hybrid storage with hydrogen, as renewable sources, will be considered, which allows management of the energy balance on different time scales. The Machine Learning method of Decision Trees, combined with ensemble methods, will also be introduced to study the optimization of microgrids. The conclusions obtained indicate that the development of suitable controllers can facilitate a competitive participation of renewable energies and the integration of microgrids in the electricity system.

Precision Medicine for Hypertension Patients with Type 2 Diabetes via Reinforcement Learning

Precision medicine is a new approach to understanding health and disease based on patient-specific data such as medical diagnoses; clinical phenotype; biologic investigations such as laboratory studies and imaging; and environmental, demographic, and lifestyle factors. The importance of machine learning techniques in healthcare has expanded quickly in the last decade owing to the rising availability of vast multi-modality data and developed computational models and algorithms. Reinforcement learning is an appealing method for developing efficient policies in various healthcare areas where the decision-making process is typically defined by a long period or a sequential process. In our research, we leverage the power of reinforcement learning and electronic health records of South Koreans to dynamically recommend treatment prescriptions, which are personalized based on patient information of hypertension. Our proposed reinforcement learning-based treatment recommendation system decides whether to use mono, dual, or triple therapy according to the state of the hypertension patients. We evaluated the performance of our personalized treatment recommendation model by lowering the occurrence of hypertension-related complications and blood pressure levels of patients who followed our model’s recommendation. With our findings, we believe that our proposed hypertension treatment recommendation model could assist doctors in prescribing appropriate antihypertensive medications.

Machine Learning Models for Predicting Short-Long Length of Stay of COVID-19 Patients

During 2020 and 2021, managing limited healthcare resources and hospital beds has been a fundamental aspect of the fight against the COVID-19 pandemic. Predicting in advance the length of stay, and in particular identifying whether a patient is going to stay in the hospital longer or less than a week, can provide important support in handling resources allocation. However, there have been significant changes in terms of containment measures, virus diffusion, new treatments, vaccines, and new variants of SARS-CoV-2 during the last period. These changes pose several conceptual drift issues that can limit the usefulness of machine learning in this context. In this work, we present a machine learning system trained and tested using data from more than 6000 hospitalised patients in northern Italy, distributed over almost two years of pandemic. We show how machine learning can be effective even by analysing data over this long period of time, also exploiting a model that predicts the patient's outcome in terms of discharge or death. Furthermore, learning from data that also consider deceased patients is a common issue in predicting the length of stay because they have severe conditions similar to patients with a long stay period, but may actually have a very short duration of hospitalisation. For this purpose, we present a method for handling data from alive and deceased patients, exploiting more patient records, increasing the robustness of the model and its performance in this task. Finally, we investigate the features that are most relevant to the prediction of the simplified length of stay.

Wearable multimodal sensors for the detection of behavioral and psychological symptoms of dementia using personalized machine learning models

Introduction

Behavioral and psychological symptoms of dementia (BPSD) signal distress or unmet needs and present a risk to people with dementia and their caregivers. Variability in the expression of these symptoms is a barrier to the performance of digital biomarkers. The aim of this study was to use wearable multimodal sensors to develop personalized machine learning models capable of detecting individual patterns of BPSD.

Methods

Older adults with dementia and BPSD (n = 17) on a dementia care unit wore a wristband during waking hours for up to 8 weeks. The wristband captured motion (accelerometer) and physiological indicators (blood volume pulse, electrodermal activity, and skin temperature). Agitation or aggression events were tracked, and research staff reviewed videos to precisely annotate the sensor data. Personalized machine learning models were developed using 1‐minute intervals and classifying the presence of behavioral symptoms, and behavioral symptoms by type (motor agitation, verbal aggression, or physical aggression).

Results

Behavioral events were rare, representing 3.4% of the total data. Personalized models classified behavioral symptoms with a median area under the receiver operating curve (AUC) of 0.87 (range 0.64–0.95). The relative importance of the different sensor features to the predictive models varied both by individual and behavior type.

Discussion

Patterns of sensor data associated with BPSD are highly individualized, and future studies of the digital phenotyping of these behaviors would benefit from personalization.

A Machine Learning Model for Predicting Unscheduled 72 h Return Visits to the Emergency Department by Patients with Abdominal Pain

Seventy-two-hour unscheduled return visits (URVs) by emergency department patients are a key clinical index for evaluating the quality of care in emergency departments (EDs). This study aimed to develop a machine learning model to predict 72 h URVs for ED patients with abdominal pain. Electronic health records data were collected from the Chang Gung Research Database (CGRD) for 25,151 ED visits by patients with abdominal pain and a total of 617 features were used for analysis. We used supervised machine learning models, namely logistic regression (LR), support vector machine (SVM), random forest (RF), extreme gradient boosting (XGB), and voting classifier (VC), to predict URVs. The VC model achieved more favorable overall performance than other models (AUROC: 0.74; 95% confidence interval (CI), 0.69–0.76; sensitivity, 0.39; specificity, 0.89; F1 score, 0.25). The reduced VC model achieved comparable performance (AUROC: 0.72; 95% CI, 0.69–0.74) to the full models using all clinical features. The VC model exhibited the most favorable performance in predicting 72 h URVs for patients with abdominal pain, both for all-features and reduced-features models. Application of the VC model in the clinical setting after validation may help physicians to make accurate decisions and decrease URVs.

On Missingness Features in Machine Learning Models for Critical Care: Observational Study

Background

Missing data in electronic health records is inevitable and considered to be nonrandom. Several studies have found that features indicating missing patterns (missingness) encode useful information about a patient’s health and advocate for their inclusion in clinical prediction models. But their effectiveness has not been comprehensively evaluated.

Objective

The goal of the research is to study the effect of including informative missingness features in machine learning models for various clinically relevant outcomes and explore robustness of these features across patient subgroups and task settings.

Methods

A total of 48,336 electronic health records from the 2012 and 2019 PhysioNet Challenges were used, and mortality, length of stay, and sepsis outcomes were chosen. The latter dataset was multicenter, allowing external validation. Gated recurrent units were used to learn sequential patterns in the data and classify or predict labels of interest. Models were evaluated on various criteria and across population subgroups evaluating discriminative ability and calibration.

Results

Generally improved model performance in retrospective tasks was observed on including missingness features. Extent of improvement depended on the outcome of interest (area under the curve of the receiver operating characteristic [AUROC] improved from 1.2% to 7.7%) and even patient subgroup. However, missingness features did not display utility in a simulated prospective setting, being outperformed (0.9% difference in AUROC) by the model relying only on pathological features. This was despite leading to earlier detection of disease (true positives), since including these features led to a concomitant rise in false positive detections.

Conclusions

This study comprehensively evaluated effectiveness of missingness features on machine learning models. A detailed understanding of how these features affect model performance may lead to their informed use in clinical settings especially for administrative tasks like length of stay prediction where they present the greatest benefit. While missingness features, representative of health care processes, vary greatly due to intra- and interhospital factors, they may still be used in prediction models for clinically relevant outcomes. However, their use in prospective models producing frequent predictions needs to be explored further.

Towards machine learning discovery of dual antibacterial drug-nanoparticle systems

Artificial Intelligence/Machine Learning (AI/ML) algorithms may speed up the design of DADNP systems formed by Antibacterial Drugs (AD) and Nanoparticles (NP). In this work, we used IFPTML = Information Fusion (IF) + Perturbation-Theory (PT) + Machine Learning (ML) algorithm for the first time to study of a large dataset of putative DADNP systems composed by >165 000 ChEMBL AD assays and 300 NP assays vs. multiple bacteria species. We trained alternative models with Linear Discriminant Analysis (LDA), Artificial Neural Networks (ANN), Bayesian Networks (BNN), K-Nearest Neighbour (KNN) and other algorithms. IFPTML-LDA model was simpler with values of Sp ≈ 90% and Sn ≈ 74% in both training (>124 K cases) and validation (>41 K cases) series. IFPTML-ANN and KNN models are notably more complicated even when they are more balanced Sn ≈ Sp ≈ 88.5%-99.0% and AUROC ≈ 0.94-0.99 in both series. We also carried out a simulation (>1900 calculations) of the expected behavior for putative DADNPs in 72 different biological assays. The putative DADNPs studied are formed by 27 different drugs with multiple classes of NP and types of coats. In addition, we tested the validity of our additive model with 80 DADNP complexes experimentally synthetized and biologically tested (reported in >45 papers). All these DADNPs show values of MIC < 50 μg mL^-1 (cutoff used) better that MIC of AD and NP alone (synergistic or additive effect). The assays involve DADNP complexes with 10 types of NP, 6 coating materials, NP size range 5-100 nm vs. 15 different antibiotics, and 12 bacteria species. The IFPTML-LDA model classified correctly 100% (80 out of 80) DADNP complexes as biologically active. IFPMTL additive strategy may become a useful tool to assist the design of DADNP systems for antibacterial therapy taking into consideration only information about AD and NP components by separate.

Systematic review of machine learning models for personalised dosing of heparin

AIM

To identify and critically appraise studies of prediction models, developed using machine learning (ML) methods, for determining the optimal dosing of unfractionated heparin (UFH).

METHODS

Embase, PubMed, CINAHL, Web of Science, International Pharmaceutical Abstracts and IEEE Xplore databases were searched from inception to 31 January 2020 to identify relevant studies using key search terms synonymous with artificial intelligence or ML, 'prediction', 'dose', 'activated partial thromboplastin time (aPTT)' and 'UFH.' Studies had to have used ML methods for developing models that predicted optimal dose of UFH or target therapeutic aPTT levels in the hospital setting. The CHARMS Checklist was used to assess quality and risk of bias of included studies.

RESULTS

Of 8393 retrieved abstracts, 61 underwent full text review and eight studies met inclusion criteria. Four studies described models for predicting aPTT, three studies described models predicting optimal dose of heparin during dialysis and one study described a model that used surrogate outcomes of clotting and bleeding to predict a therapeutic aPTT. Studies varied widely in reporting of study participants, feature characterisation and selection, handling of missing data, sample size calculations and the intended clinical application of the model. Only one study conducted an external validation and no studies evaluated model impacts in clinical practice.

CONCLUSION

Studies of ML models for UFH dosing are few and none report a model ready for routine clinical use. Existing studies are limited by low methodological quality, inadequate reporting of study factors and absence of external validation and impact analysis.

Machine Learning Prediction Models for Mortality in Intensive Care Unit Patients with Lactic Acidosis

Background: Lactic acidosis is the most common cause of anion gap metabolic acidosis in the intensive care unit (ICU), associated with poor outcomes including mortality. We sought to compare machine learning (ML) approaches versus logistic regression analysis for prediction of mortality in lactic acidosis patients admitted to the ICU. Methods: We used the Medical Information Mart for Intensive Care (MIMIC-III) database to identify ICU adult patients with lactic acidosis (serum lactate ≥4 mmol/L). The outcome of interest was hospital mortality. We developed prediction models using four ML approaches consisting of random forest (RF), decision tree (DT), extreme gradient boosting (XGBoost), artificial neural network (ANN), and statistical modeling with forward stepwise logistic regression using the testing dataset. We then assessed model performance using area under the receiver operating characteristic curve (AUROC), accuracy, precision, error rate, Matthews correlation coefficient (MCC), F1 score, and assessed model calibration using the Brier score, in the independent testing dataset. Results: Of 1919 lactic acidosis ICU patients, 1535 and 384 were included in the training and testing dataset, respectively. Hospital mortality was 30%. RF had the highest AUROC at 0.83, followed by logistic regression 0.81, XGBoost 0.81, ANN 0.79, and DT 0.71. In addition, RF also had the highest accuracy (0.79), MCC (0.45), F1 score (0.56), and lowest error rate (21.4%). The RF model was the most well-calibrated. The Brier score for RF, DT, XGBoost, ANN, and multivariable logistic regression was 0.15, 0.19, 0.18, 0.19, and 0.16, respectively. The RF model outperformed multivariable logistic regression model, SOFA score (AUROC 0.74), SAP II score (AUROC 0.77), and Charlson score (AUROC 0.69). Conclusion: The ML prediction model using RF algorithm provided the highest predictive performance for hospital mortality among ICU patient with lactic acidosis.

Artificial Intelligence in Brain Tumour Surgery-An Emerging Paradigm

Artificial intelligence (AI) platforms have the potential to cause a paradigm shift in brain tumour surgery. Brain tumour surgery augmented with AI can result in safer and more effective treatment. In this review article, we explore the current and future role of AI in patients undergoing brain tumour surgery, including aiding diagnosis, optimising the surgical plan, providing support during the operation, and better predicting the prognosis. Finally, we discuss barriers to the successful clinical implementation, the ethical concerns, and we provide our perspective on how the field could be advanced.

Estimated Pao2: A Continuous and Noninvasive Method to Estimate Pao2 and Oxygenation Index

Supplemental Digital Content is available in the text.

Pao₂ is the gold standard to assess acute hypoxic respiratory failure, but it is only routinely available by intermittent spot checks, precluding any automatic continuous analysis for bedside tools.

Organizational readiness for artificial intelligence in health care: insights for decision-making and practice

PURPOSE

Artificial intelligence (AI) raises many expectations regarding its ability to profoundly transform health care delivery. There is an abundant literature on the technical performance of AI applications in many clinical fields (e.g. radiology, ophthalmology). This article aims to bring forward the importance of studying organizational readiness to integrate AI into health care delivery.

DESIGN/METHODOLOGY/APPROACH

The reflection is based on our experience in digital health technologies, diffusion of innovations and healthcare organizations and systems. It provides insights into why and how organizational readiness should be carefully considered.

FINDINGS

As an important step to ensure successful integration of AI and avoid unnecessary investments and costly failures, better consideration should be given to: (1) Needs and added-value assessment; (2) Workplace readiness: stakeholder acceptance and engagement; (3) Technology-organization alignment assessment and (4) Business plan: financing and investments. In summary, decision-makers and technology promoters should better address the complexity of AI and understand the systemic challenges raised by its implementation in healthcare organizations and systems.

ORIGINALITY/VALUE

Few studies have focused on the organizational issues raised by the integration of AI into clinical routine. The current context is marked by a perplexing gap between the willingness of decision-makers and technology promoters to capitalize on AI applications to improve health care delivery and the reality on the ground, where it is difficult to initiate the changes needed to realize their full benefits while avoiding their negative impacts.

A Novel Switching State-Space Model for Post-ICU Mortality Prediction and Survival Analysis

Predicting mortality risk in patients accurately during and after intensive care unit (ICU) stay is an essential component for supporting critical care decision-making. To date, various scoring systems have been designed for survival analysis and mortality prediction by providing risk scores based on patient's vital signs and lab results. However, it is challenging using these universal scores to represent the overall severity level of illness and to look into patient's deterioration leading to high mortality risk during ICU stay. Thus, a close monitoring of the severity level over time during ICU stay is more preferable. In this study, we design a new switching state-space model by correlating patient's condition dynamics in last hours of ICU stay to the risk probabilities in a short time period (1-6 days) after ICU discharge. More specifically, we propose to integrate a cumulative hazard function estimating survival probability into the autoregressive hidden Markov model using time-interval sequential SAPS II scores as features. We demonstrate the significant improvement of mortality prediction comparing to SAPS I, SAPS II, and SOFA scoring systems for the PhysioNet MIMIC II Challenge data.

An Artificial Neural Network-Based Pediatric Mortality Risk Score: Development and Performance Evaluation Using Data From a Large North American Registry

Background

In the pediatric intensive care unit (PICU), quantifying illness severity can be guided by risk models to enable timely identification and appropriate intervention. Logistic regression models, including the pediatric index of mortality 2 (PIM-2) and pediatric risk of mortality III (PRISM-III), produce a mortality risk score using data that are routinely available at PICU admission. Artificial neural networks (ANNs) outperform regression models in some medical fields.

Objective

In light of this potential, we aim to examine ANN performance, compared to that of logistic regression, for mortality risk estimation in the PICU.

Methods

The analyzed data set included patients from North American PICUs whose discharge diagnostic codes indicated evidence of infection and included the data used for the PIM-2 and PRISM-III calculations and their corresponding scores. We stratified the data set into training and test sets, with approximately equal mortality rates, in an effort to replicate real-world data. Data preprocessing included imputing missing data through simple substitution and normalizing data into binary variables using PRISM-III thresholds. A 2-layer ANN model was built to predict pediatric mortality, along with a simple logistic regression model for comparison. Both models used the same features required by PIM-2 and PRISM-III. Alternative ANN models using single-layer or unnormalized data were also evaluated. Model performance was compared using the area under the receiver operating characteristic curve (AUROC) and the area under the precision recall curve (AUPRC) and their empirical 95% CIs.

Results

Data from 102,945 patients (including 4068 deaths) were included in the analysis. The highest performing ANN (AUROC 0.871, 95% CI 0.862-0.880; AUPRC 0.372, 95% CI 0.345-0.396) that used normalized data performed better than PIM-2 (AUROC 0.805, 95% CI 0.801-0.816; AUPRC 0.234, 95% CI 0.213-0.255) and PRISM-III (AUROC 0.844, 95% CI 0.841-0.855; AUPRC 0.348, 95% CI 0.322-0.367). The performance of this ANN was also significantly better than that of the logistic regression model (AUROC 0.862, 95% CI 0.852-0.872; AUPRC 0.329, 95% CI 0.304-0.351). The performance of the ANN that used unnormalized data (AUROC 0.865, 95% CI 0.856-0.874) was slightly inferior to our highest performing ANN; the single-layer ANN architecture performed poorly and was not investigated further.

Conclusions

A simple ANN model performed slightly better than the benchmark PIM-2 and PRISM-III scores and a traditional logistic regression model trained on the same data set. The small performance gains achieved by this two-layer ANN model may not offer clinically significant improvement; however, further research with other or more sophisticated model designs and better imputation of missing data may be warranted.

Candidemia Risk Prediction (CanDETEC) Model for Patients With Malignancy: Model Development and Validation in a Single-Center Retrospective Study

BACKGROUND

Appropriate empirical treatment for candidemia is associated with reduced mortality; however, the timely diagnosis of candidemia in patients with sepsis remains poor.

OBJECTIVE

We aimed to use machine learning algorithms to develop and validate a candidemia prediction model for patients with cancer.

METHODS

We conducted a single-center retrospective study using the cancer registry of a tertiary academic hospital. Adult patients diagnosed with malignancies between January 2010 and December 2018 were included. Our study outcome was the prediction of candidemia events. A stratified undersampling method was used to extract control data for algorithm learning. Multiple models were developed-a combination of 4 variable groups and 5 algorithms (auto-machine learning, deep neural network, gradient boosting, logistic regression, and random forest). The model with the largest area under the receiver operating characteristic curve (AUROC) was selected as the Candida detection (CanDETEC) model after comparing its performance indexes with those of the Candida Score Model.

RESULTS

From a total of 273,380 blood cultures from 186,404 registered patients with cancer, we extracted 501 records of candidemia events and 2000 records as control data. Performance among the different models varied (AUROC 0.771- 0.889), with all models demonstrating superior performance to that of the Candida Score (AUROC 0.677). The random forest model performed the best (AUROC 0.889, 95% CI 0.888-0.889); therefore, it was selected as the CanDETEC model.

CONCLUSIONS

The CanDETEC model predicted candidemia in patients with cancer with high discriminative power. This algorithm could be used for the timely diagnosis and appropriate empirical treatment of candidemia.

Non-occlusive mesenteric ischemia: Diagnostic challenges and perspectives in the era of artificial intelligence

Acute mesenteric ischemia (AMI) is a severe condition associated with poor prognosis, ultimately leading to death due to multiorgan failure. Several mechanisms may lead to AMI, and non-occlusive mesenteric ischemia (NOMI) represents a particular form of AMI. NOMI is prevalent in intensive care units in critically ill patients. In NOMI management, promptness and accuracy of diagnosis are paramount to achieve decisive treatment, but the last decades have been marked by failure to improve NOMI prognosis, due to lack of tools to detect this condition. While real-life diagnostic management relies on a combination of physical examination, several biomarkers, imaging, and endoscopy to detect the possibility of several grades of NOMI, research studies only focus on a few elements at a time. In the era of artificial intelligence (AI), which can aggregate thousands of variables in complex longitudinal models, the prospect of achieving accurate diagnosis through machine-learning-based algorithms may be sought. In the following work, we bring you a state-of-the-art literature review regarding NOMI, its presentation, its mechanics, and the pitfalls of routine work-up diagnostic exams including biomarkers, imaging, and endoscopy, we raise the perspectives of new biomarker exams, and finally we discuss what AI may add to the field, after summarizing what this technique encompasses.

Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit

PURPOSE

Due to the increasing demand for intensive care unit (ICU) treatment, and to improve quality and efficiency of care, there is a need for adequate and efficient clinical decision-making. The advancement of artificial intelligence (AI) technologies has resulted in the development of prediction models, which might aid clinical decision-making. This systematic review seeks to give a contemporary overview of the current maturity of AI in the ICU, the research methods behind these studies, and the risk of bias in these studies.

METHODS

A systematic search was conducted in Embase, Medline, Web of Science Core Collection and Cochrane Central Register of Controlled Trials databases to identify eligible studies. Studies using AI to analyze ICU data were considered eligible. Specifically, the study design, study aim, dataset size, level of validation, level of readiness, and the outcomes of clinical trials were extracted. Risk of bias in individual studies was evaluated by the Prediction model Risk Of Bias ASsessment Tool (PROBAST).

RESULTS

Out of 6455 studies identified through literature search, 494 were included. The most common study design was retrospective [476 studies (96.4% of all studies)] followed by prospective observational [8 (1.6%)] and clinical [10 (2%)] trials. 378 (80.9%) retrospective studies were classified as high risk of bias. No studies were identified that reported on the outcome evaluation of an AI model integrated in routine clinical practice.

CONCLUSION

The vast majority of developed ICU-AI models remain within the testing and prototyping environment; only a handful were actually evaluated in clinical practice. A uniform and structured approach can support the development, safe delivery, and implementation of AI to determine clinical benefit in the ICU.

Machine Learning Analysis to Identify Digital Behavioral Phenotypes for Engagement and Health Outcome Efficacy of an mHealth Intervention for Obesity: Randomized Controlled Trial

BACKGROUND

The digital health care community has been urged to enhance engagement and clinical outcomes by analyzing multidimensional digital phenotypes.

OBJECTIVE

This study aims to use a machine learning approach to investigate the performance of multivariate phenotypes in predicting the engagement rate and health outcomes of digital cognitive behavioral therapy.

METHODS

We leveraged both conventional phenotypes assessed by validated psychological questionnaires and multidimensional digital phenotypes within time-series data from a mobile app of 45 participants undergoing digital cognitive behavioral therapy for 8 weeks. We conducted a machine learning analysis to discriminate the important characteristics.

RESULTS

A higher engagement rate was associated with higher weight loss at 8 weeks (r=-0.59; P<.001) and 24 weeks (r=-0.52; P=.001). Applying the machine learning approach, lower self-esteem on the conventional phenotype and higher in-app motivational measures on digital phenotypes commonly accounted for both engagement and health outcomes. In addition, 16 types of digital phenotypes (ie, lower intake of high-calorie food and evening snacks and higher interaction frequency with mentors) predicted engagement rates (mean R² 0.416, SD 0.006). The prediction of short-term weight change (mean R² 0.382, SD 0.015) was associated with 13 different digital phenotypes (ie, lower intake of high-calorie food and carbohydrate and higher intake of low-calorie food). Finally, 8 measures of digital phenotypes (ie, lower intake of carbohydrate and evening snacks and higher motivation) were associated with a long-term weight change (mean R² 0.590, SD 0.011).

CONCLUSIONS

Our findings successfully demonstrated how multiple psychological constructs, such as emotional, cognitive, behavioral, and motivational phenotypes, elucidate the mechanisms and clinical efficacy of a digital intervention using the machine learning method. Accordingly, our study designed an interpretable digital phenotype model, including multiple aspects of motivation before and during the intervention, predicting both engagement and clinical efficacy. This line of research may shed light on the development of advanced prevention and personalized digital therapeutics.

TRIAL REGISTRATION

ClinicalTrials.gov NCT03465306; https://clinicaltrials.gov/ct2/show/NCT03465306.

White Blood Cell and Platelet Dynamics Define Human Inflammatory Recovery

Inflammation is the physiologic reaction to cellular and tissue damage caused by pathologic processes including trauma, infection, and ischemia 1 . Effective inflammatory responses integrate molecular and cellular functions to prevent further tissue damage, initiate repair, and restore homeostasis, while futile or dysfunctional responses allow escalating injury, delay recovery, and may hasten death 2 . Elevation of white blood cell count (WBC) and altered levels of other acute phase reactants are cardinal signs of inflammation, but the dynamics of these changes and their resolution are not established 3,4 . Patient responses appear to vary dramatically with no clearly defined signs of good prognosis, leaving physicians reliant on qualitative interpretations of laboratory trends 4,5 . We retrospectively, observationally studied the human acute inflammatory response to trauma, ischemia, and infection by tracking the longitudinal dynamics of cellular and serum markers in hospitalized patients. Unexpectedly, we identified a conserved pattern of recovery defined by co-regulation of WBC and platelet (PLT) populations. Across all inflammatory conditions studied, recovering patients followed a consistent WBC-PLT trajectory shape that is well-approximated by exponential WBC decay and delayed linear PLT growth. This recovery trajectory shape may represent a fundamental archetype of human physiologic response at the cellular population scale, and provides a generic approach for identifying high-risk patients: 32x relative risk of adverse outcomes for cardiac surgery patients, 9x relative risk of death for COVID-19, and 5x relative risk of death for myocardial infarction.

Machine learning-based prediction of outcomes of the endoscopic endonasal approach in Cushing disease: is the future coming?

OBJECTIVE

Machine learning (ML) is an innovative method to analyze large and complex data sets. The aim of this study was to evaluate the use of ML to identify predictors of early postsurgical and long-term outcomes in patients treated for Cushing disease (CD).

METHODS

All consecutive patients in our center who underwent surgery for CD through the endoscopic endonasal approach were retrospectively reviewed. Study endpoints were gross-tumor removal (GTR), postsurgical remission, and long-term control of disease. Several demographic, radiological, and histological factors were assessed as potential predictors. For ML-based modeling, data were randomly divided into 2 sets with an 80% to 20% ratio for bootstrapped training and testing, respectively. Several algorithms were tested and tuned for the area under the curve (AUC).

RESULTS

The study included 151 patients. GTR was achieved in 137 patients (91%), and postsurgical hypersecretion remission was achieved in 133 patients (88%). At last follow-up, 116 patients (77%) were still in remission after surgery and in 21 patients (14%), CD was controlled with complementary treatment (overall, of 131 cases, 87% were under control at follow-up). At internal validation, the endpoints were predicted with AUCs of 0.81-1.00, accuracy of 81%-100%, and Brier scores of 0.035-0.151. Tumor size and invasiveness and histological confirmation of adrenocorticotropic hormone (ACTH)-secreting cells were the main predictors for the 3 endpoints of interest.

CONCLUSIONS

ML algorithms were used to train and internally validate robust models for all the endpoints, giving accurate outcome predictions in CD cases. This analytical method seems promising for potentially improving future patient care and counseling; however, careful clinical interpretation of the results remains necessary before any clinical adoption of ML. Moreover, further studies and increased sample sizes are definitely required before the widespread adoption of ML to the study of CD.

Predictors of death and new disability after critical illness: a multicentre prospective cohort study

PURPOSE

This study aimed to determine the prevalence and predictors of death or new disability following critical illness.

METHODS

Prospective, multicentre cohort study conducted in six metropolitan intensive care units (ICU). Participants were adults admitted to the ICU who received more than 24 h of mechanical ventilation. The primary outcome was death or new disability at 6 months, with new disability defined by a 10% increase in the WHODAS 2.0.

RESULTS

Of 628 patients with the primary outcome available (median age of 62 [49-71] years, 379 [61.0%] had a medical admission and 370 (58.9%) died or developed new disability by 6 months. Independent predictors of death or new disability included age [OR 1.02 (1.01-1.03), P = 0.001], higher severity of illness (APACHE III) [OR 1.02 (1.01-1.03), P < 0.001] and admission diagnosis. Compared to patients with a surgical admission diagnosis, patients with a cardiac arrest [OR (95% CI) 4.06 (1.89-8.68), P < 0.001], sepsis [OR (95% CI) 2.43 (1.32-4.47), P = 0.004], or trauma [OR (95% CI) 6.24 (3.07-12.71), P < 0.001] diagnosis had higher odds of death or new disability, while patients with a lung transplant [OR (95% CI) 0.21 (0.07-0.58), P = 0.003] diagnosis had lower odds. A model including these three variables had good calibration (Brier score 0.20) and acceptable discriminative power with an area under the receiver operating characteristic curve of 0.76 (95% CI 0.72-0.80).

CONCLUSION

Less than half of all patients mechanically ventilated for more than 24 h were alive and free of new disability at 6 months after admission to ICU. A model including age, illness severity and admission diagnosis has acceptable discriminative ability to predict death or new disability at 6 months.

Predicting need for hospital-specific interventional care after surgery using electronic health record data

BACKGROUND

A significant proportion of surgical inpatients is often admitted longer than necessary. Early identification of patients who do not need care that is strictly provided within hospitals would allow timely discharge of patients to a postoperative nursing home for further recovery. We aimed to develop a model to predict whether a patient needs hospital-specific interventional care beyond the second postoperative day.

METHODS

This study included all adult patients discharged from surgical care in the surgical oncology department from June 2017 to February 2020. The primary outcome was to predict whether a patient still needs hospital-specific interventional care beyond the second postoperative day. Hospital-specific care was defined as unplanned reoperations, radiological interventions, and intravenous antibiotics administration. Different analytical methods were compared with respect to the area under the receiver-operating characteristics curve, sensitivity, specificity, positive predictive value, and negative predictive value.

RESULTS

Each model was trained on 1,174 episodes. In total, 847 (50.5%) patients required an intervention during postoperative admission. A random forest model performed best with an area under the receiver-operating characteristics curve of 0.88 (95% confidence interval 0.83-0.93), sensitivity of 79.1% (95% confidence interval 0.67-0.92), specificity of 80.0% (0.73-0.87), positive predictive value of 57.6% (0.45-0.70) and negative predictive value of 91.7% (0.87-0.97).

CONCLUSION

This proof-of-concept study found that a random forest model could successfully predict whether a patient could be safely discharged to a nursing home and does not need hospital care anymore. Such a model could aid hospitals in addressing capacity challenges and improve patient flow, allowing for timely surgical care.

Predicting the 9-year course of mood and anxiety disorders with automated machine learning: A comparison between auto-sklearn, naïve Bayes classifier, and traditional logistic regression

BACKGROUND

Predicting the onset and course of mood and anxiety disorders is of clinical importance but remains difficult. We compared the predictive performances of traditional logistic regression, basic probabilistic machine learning (ML) methods, and automated ML (Auto-sklearn).

METHODS

Data were derived from the Netherlands Study of Depression and Anxiety. We compared how well multinomial logistic regression, a naïve Bayes classifier, and Auto-sklearn predicted depression and anxiety diagnoses at a 2-, 4-, 6-, and 9-year follow up, operationalized as binary or categorical variables. Predictor sets included demographic and self-report data, which can be easily collected in clinical practice at two initial time points (baseline and 1-year follow up).

RESULTS

At baseline, participants were 42.2 years old, 66.5% were women, and 53.6% had a current mood or anxiety disorder. The three methods were similarly successful in predicting (mental) health status, with correct predictions for up to 79% (95% CI 75-81%). However, Auto-sklearn was superior when assessing a more complex dataset with individual item scores.

CONCLUSIONS

Automated ML methods added only limited value, compared to traditional data modelling when predicting the onset and course of depression and anxiety. However, they hold potential for automatization and may be better suited for complex datasets.

Incorporating real-world evidence into the development of patient blood glucose prediction algorithms for the ICU

Objective

Glycemic control is an important component of critical care. We present a data-driven method for predicting intensive care unit (ICU) patient response to glycemic control protocols while accounting for patient heterogeneity and variations in care.

Materials and Methods

Using electronic medical records (EMRs) of 18 961 ICU admissions from the MIMIC-III dataset, including 318 574 blood glucose measurements, we train and validate a gradient boosted tree machine learning (ML) algorithm to forecast patient blood glucose and a 95% prediction interval at 2-hour intervals. The model uses as inputs irregular multivariate time series data relating to recent in-patient medical history and glycemic control, including previous blood glucose, nutrition, and insulin dosing.

Results

Our forecasting model using routinely collected EMRs achieves performance comparable to previous models developed in planned research studies using continuous blood glucose monitoring. Model error, expressed as mean absolute percentage error is 16.5%–16.8%, with Clarke error grid analysis demonstrating that 97% of predictions would be clinically acceptable. The 95% prediction intervals achieve near intended coverage at 93%–94%.

Discussion

ML algorithms built on observational data sources, such as EMRs, present a promising approach for personalization and automation of glycemic control in critical care. Future research may benefit from applying a combination of methodologies and data sources to develop robust methodologies that account for the variations seen in ICU patients and difficultly in detecting the extremes of observed blood glucose values.

Conclusion

We demonstrate that EMRs can be used to train ML algorithms that may be suitable for incorporation into ICU decision support systems.

Convergence of Precision Medicine and Public Health Into Precision Public Health: Toward a Big Data Perspective

The field of precision medicine explores disease treatments by looking at genetic, socio-environmental, and clinical factors, thus trying to provide a holistic view of a person's health. Public health, on the other hand, is focused on improving the health of populations through preventive strategies and timely interventions. With recent advances in technology, we are able to collect, analyze and store for the first-time large volumes of real-time, diverse and continuous health data. Typically, the field of precision medicine deals with a huge amount of data from few individuals; public health, on the other hand, deals with limited data from a population. With the coming of Big Data, the fields of precision medicine and public health are converging into precision public health, the study of biological and genetic factors supported by large amounts of population data. In this paper, we explore through a comprehensive review the data types and use cases found in precision medicine and public health. We also discuss how these data types and use cases can converge toward precision public health, as well as challenges and opportunities provided by research and analyses of health data.

Prediction of Mortality in Surgical Intensive Care Unit Patients Using Machine Learning Algorithms

Objective: Predicting prognosis of in-hospital patients is critical. However, it is challenging to accurately predict the life and death of certain patients at certain period. To determine whether machine learning algorithms could predict in-hospital death of critically ill patients with considerable accuracy and identify factors contributing to the prediction power.

Materials and Methods: Using medical data of 1,384 patients admitted to the Surgical Intensive Care Unit (SICU) of our institution, we investigated whether machine learning algorithms could predict in-hospital death using demographic, laboratory, and other disease-related variables, and compared predictions using three different algorithmic methods. The outcome measurement was the incidence of unexpected postoperative mortality which was defined as mortality without pre-existing not-for-resuscitation order that occurred within 30 days of the surgery or within the same hospital stay as the surgery.

Results: Machine learning algorithms trained with 43 variables successfully classified dead and live patients with very high accuracy. Most notably, the decision tree showed the higher classification results (Area Under the Receiver Operating Curve, AUC = 0.96) than the neural network classifier (AUC = 0.80). Further analysis provided the insight that serum albumin concentration, total prenatal nutritional intake, and peak dose of dopamine drug played an important role in predicting the mortality of SICU patients.

Conclusion: Our results suggest that machine learning algorithms, especially the decision tree method, can provide information on structured and explainable decision flow and accurately predict hospital mortality in SICU hospitalized patients.

Outcomes prediction in longitudinal data: Study designs evaluation, use case in ICU acquired sepsis

Outcomes' prediction in Electronic Health Records (EHR) and specifically in Critical Care is increasingly attracting more exploration and research. In this study, we used clinical data from the Intensive Care Unit (ICU), focusing on ICU acquired sepsis. Looking at the current literature, several evaluation approaches are reported, inspired by epidemiological designs, in which some do not always reflect real-life application's conditions. This problem seems relevant generally to outcomes' prediction in longitudinal EHR data, or generally longitudinal data, while in this study we focused on ICU data. Unlike in most previous studies that investigated all sepsis admissions, we focused specifically on ICU-Acquired Sepsis. Due to the sparse nature of the longitudinal data, we employed the use of Temporal Abstraction and Time Interval-Related Patterns discovery, which are further used as classification features. Two experiments were designed using three different outcomes prediction study designs from the literature, implementing various levels of real-life conditions to evaluate the prediction models. The first experiment focused on predicting whether a patient would suffer from ICU-acquired sepsis and when during her admission, given a sliding observation time window, and the comparison of the three study designs behavior. The second experiment focused only on predicting whether the patient will suffer from ICU-acquired sepsis, based on data taken relatively to his admission start time. Our results show that using Temporal Discretization for Classification (TD4C) led to better performance than using the Equal-Width Discretization, Knowledge-Based, or SAX. Also, using two states abstraction was better than three or four. Using the default Binary TIRP representation method performed better than Mean Duration, Horizontal Support, and horizontally normalized horizontal support. Using XGBoost as a classifier performed better than Logistic Regression, Neural Net, or Random Forest. Additionally, it is demonstrated why the use of case-crossover-control is most appropriate for real life application conditions evaluation, unlike other incomplete designs that may even result in "better performance".

PREDICTING LAND PRICES AND MEASURING UNCERTAINTY BY COMBINING SUPERVISED AND UNSUPERVISED LEARNING

Despite the popularity deep learning has been gaining, measuring the uncertainty within the result has not met expectations in many deep learning applications and this includes property valuation. In real-world tasks, however, rather than simply requiring predictions, assurance of the certainty of the predictions is also demanded. In this study, supervised learning is combined with unsupervised learning to bridge this gap. A method based on principal component analysis, a popular tool of unsupervised learning, was developed and used to represent the uncertainty in property valuation. Then, a neural network, a representative algorithm to implement supervised learning, was constructed, and trained to predict land prices. Finally, the uncertainty that was measured using principal component analysis was incorporated into the price predicted by the neural network. This hybrid approach is shown to be likely to improve the credibility of the valuation work. The findings of this study are expected to generate interest in the integration of the two learning approaches, thereby promoting the rapid adoption of deep learning tools in the property valuation industry.

Dynamic Detection of Delayed Cerebral Ischemia: A Study in 3 Centers

BACKGROUND AND PURPOSE

Delayed cerebral ischemia (DCI) after aneurysmal subarachnoid hemorrhage negatively impacts long-term recovery but is often detected too late to prevent damage. We aim to develop hourly risk scores using routinely collected clinical data to detect DCI.

METHODS

A DCI classification model was trained using vital sign measurements (heart rate, blood pressure, respiratory rate, and oxygen saturation) and demographics routinely collected for clinical care. Twenty-two time-varying physiological measures were computed including mean, SD, and cross-correlation of heart rate time series with each of the other vitals. Classification was achieved using an ensemble approach with L2-regularized logistic regression, random forest, and support vector machines models. Classifier performance was determined by area under the receiver operating characteristic curves and confusion matrices. Hourly DCI risk scores were generated as the posterior probability at time t using the Ensemble classifier on cohorts recruited at 2 external institutions (n=38 and 40).

RESULTS

Three hundred ten patients were included in the training model (median, 54 years old [interquartile range, 45-65]; 80.2% women, 28.4% Hunt and Hess scale 4-5, 38.7% Modified Fisher Scale 3-4); 101 (33%) developed DCI with a median onset day 6 (interquartile range, 5-8). Classification accuracy before DCI onset was 0.83 (interquartile range, 0.76-0.83) area under the receiver operating characteristic curve. Risk scores applied to external institution datasets correctly predicted 64% and 91% of DCI events as early as 12 hours before clinical detection, with 2.7 and 1.6 true alerts for every false alert.

CONCLUSIONS

An hourly risk score for DCI derived from routine vital signs may have the potential to alert clinicians to DCI, which could reduce neurological injury.

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.

Intensive care digital health response to emerging infectious disease outbreaks such as COVID-19

The COVID-19 pandemic has required intensive care units to rapidly adjust and adapt their existing practices. Although there has a focus on expanding critical care infrastructure, equipment and workforce, plans have not emphasised the need to increase digital capabilities. The objective of this report was to recognise key areas of digital health related to the COVID-19 response. We identified and explored six focus areas relevant to intensive care, including using digital solutions to increase critical care capacity, developing surge capacity within an electronic health record, maintenance and downtime planning, training considerations and the role of data analytics. This article forms the basis of a framework for the intensive care digital health response to COVID-19 and other emerging infectious disease outbreaks.

An artificial intelligence–based decision support and resource management system for COVID-19 pandemic

COVID-19 crisis has shown that the World is not ready for such a rapid spread of a virus resulting in a catastrophic pandemic. Effective use of information technologies is one of the key aspects in reducing the adverse effects of any epidemic or pandemic. Existing management systems have failed to fulfill requirements for curbing the rapid spread of the virus. This chapter firstly describes the current solutions by giving real-world examples. Then, we propose an epidemic management system (EMS) that relies on unimpeded and timely information flow between nations and organizations to ensure resources are distributed effectively. This system will use mobile technology, blockchain, epidemic modeling, and artificial intelligence technologies. We used the Multiplatform Interoperable Scalable Architecture (MPISA) model that allows the integration of multiple platforms and provides a solution for scalability and interoperability problems. Open data repositories and the MiPasa blockchain are also described. These relevant data can be used to predict the potential future spread of the epidemic. Selecting the correct methods for epidemic modeling is discussed as well. Another challenge is deciding on allocating resources where they are most necessary; we propose deploying automated machine learning and stochastic epidemic model-based decision support systems for such purposes. Citizens should not have privacy concerns about the information systems. These trust issues and privacy concerns can be solved by using decentralized identity and zero-knowledge proof-based mechanisms. These mechanisms will ensure that users are in control of their data. In this chapter, we also discuss choosing the right machine learning method, privacy measures, and how the performance challenges can be addressed. This chapter concludes on a discussion of how we can design and deploy better EMSs and possible future studies.

Dynamic survival prediction in intensive care units from heterogeneous time series without the need for variable selection or curation

Extensive monitoring in intensive care units (ICUs) generates large quantities of data which contain numerous trends that are difficult for clinicians to systematically evaluate. Current approaches to such heterogeneity in electronic health records (EHRs) discard pertinent information. We present a deep learning pipeline that uses all uncurated chart, lab, and output events for prediction of in-hospital mortality without variable selection. Over 21,000 ICU patients and tens of thousands of variables derived from the MIMIC-III database were used to train and validate our model. Recordings in the first few hours of a patient's stay were found to be strongly predictive of mortality, outperforming models using SAPS II and OASIS scores, AUROC 0.72 and 0.76 at 24 h respectively, within just 12 h of ICU admission. Our model achieves a very strong predictive performance of AUROC 0.85 (95% CI 0.83-0.86) after 48 h. Predictive performance increases over the first 48 h, but suffers from diminishing returns, providing rationale for time-limited trials of critical care and suggesting that the timing of decision making can be optimised and individualised.

Deep-learning-based real-time prediction of acute kidney injury outperforms human predictive performance

Acute kidney injury (AKI) is a major complication after cardiothoracic surgery. Early prediction of AKI could prompt preventive measures, but is challenging in the clinical routine. One important reason is that the amount of postoperative data is too massive and too high-dimensional to be effectively processed by the human operator. We therefore sought to develop a deep-learning-based algorithm that is able to predict postoperative AKI prior to the onset of symptoms and complications. Based on 96 routinely collected parameters we built a recurrent neural network (RNN) for real-time prediction of AKI after cardiothoracic surgery. From the data of 15,564 admissions we constructed a balanced training set (2224 admissions) for the development of the RNN. The model was then evaluated on an independent test set (350 admissions) and yielded an area under curve (AUC) (95% confidence interval) of 0.893 (0.862-0.924). We compared the performance of our model against that of experienced clinicians. The RNN significantly outperformed clinicians (AUC = 0.901 vs. 0.745, p < 0.001) and was overall well calibrated. This was not the case for the physicians, who systematically underestimated the risk (p < 0.001). In conclusion, the RNN was superior to physicians in the prediction of AKI after cardiothoracic surgery. It could potentially be integrated into hospitals' electronic health records for real-time patient monitoring and may help to detect early AKI and hence modify the treatment in perioperative care.

Predicting thromboembolic complications in COVID-19 ICU patients using machine learning

BACKGROUND

The coronavirus disease 2019 (COVID-19) pandemic is a challenge for intensive care units (ICU) in part due to the failure to identify risks for patients early and the inability to render an accurate prognosis. Previous reports suggest a strong association between hypercoagulability and poor outcome. Factors related to hemostasis may, therefore, serve as tools to improve the management of COVID-19 patients.

AIM

The purpose of this report is to develop a model to determine whether it is possible to early identify COVID-19 patients at risk for thromboembolic complications (TCs).

METHODS

We analyzed electronic health record data of 108 consecutive COVID-19 patients admitted to the adult ICU of the Erasmus University Medical Center between February 27 and May 20, 2020. By training a decision tree classifier on 66% of the available data, a model for the prediction of TCs was developed.

RESULTS

The median (interquartile range) age was 62 (53-70) years and 73% were male. Forty-three patients (40%) developed a TC during their ICU stay. Mortality was higher for patients in the TCs group compared to the control group (26% vs. 8%, P=0.03). Lactate dehydrogenase, standardized bicarbonate, albumin, and leukocytes were identified by the Decision Tree classifier as the most powerful predictors for TCs 2 days before the onset of the TC, with a sensitivity of 73% and a positive likelihood ratio of 2.7 on the test dataset.

CONCLUSIONS

Clinically relevant TCs frequently occur in critically ill COVID-19 patients. These can successfully be predicted using a decision tree model. Although this model could be of special importance to aid clinical decision making, its generalizability and clinical impact should be determined in a larger population.

RELEVANCE FOR PATIENTS

Recently, severe TCs were observed in COVID-19 patients with progressive respiratory failure warranting ICU treatment. Timely identification of patients at risk of developing TCs is critical inasmuch as it would enable clinicians to initiate potentially salvaging therapeutic anticoagulation.

Sampling methods and feature selection for mortality prediction with neural networks

Along with digitization, automatic data-driven decision support systems become increasingly popular. Mortality prediction is a vital part of that decision process. With more data available, sophisticated machine learning models like (Artificial) Neural Networks (NNs) can be applied and promise favorable performance. We evaluate the reproducibility of a published mortality prediction approach using NNs along with the possibility to generalize it to a bigger and more generic dataset. We describe an extensive preprocessing pipeline, as well as the evaluation of different sampling techniques and NN architectures. Through training on a loss function that optimizes both, precision and recall, in combination with a good set of hyperparameters and a set of new features, we use a NN to predict in-hospital mortality with accuracy, sensitivity, and area under the receiver operating characteristic score of greater than 0.8.

Exploration of critical care data by using unsupervised machine learning

BACKGROUND AND OBJECTIVE

Identification of subgroups may be useful to understand the clinical characteristics of ICU patients. The purposes of this study were to apply an unsupervised machine learning method to ICU patient data to discover subgroups among them; and to examine their clinical characteristics, therapeutic procedures conducted during the ICU stay, and discharge dispositions.

METHODS

K-means clustering method was used with 1503 observations and 9 types of laboratory test results as features.

RESULTS

Three clusters were identified from this specific population. Blood urea nitrogen, creatinine, potassium, hemoglobin, and red blood cell were distinctive between the clusters. Cluster Three presented the highest blood products transfusion rate (19.8%), followed by Cluster One (15.5%) and cluster Two (9.3%), which was significantly different. Hemodialysis was more frequently provided to Cluster Three while bronchoscopy was done to Cluster One and Two. Cluster Three showed the highest mortality (30.4%), which was more than two-fold compared to Cluster One (14.1%) and Two (12.2%).

CONCLUSION

Three subgroups were identified and their clinical characteristics were compared. These findings may be useful to anticipate treatment strategies and probable outcomes of ICU patients. Unsupervised machine learning may enable ICU multi-dimensional data to be organized and to make sense of the data.

From Clinic to Computer and Back Again: Practical Considerations When Designing and Implementing Machine Learning Solutions for Pediatrics

Purpose of review

Machine learning (ML), a branch of artificial intelligence, is influencing all fields in medicine, with an abundance of work describing its application to adult practice. ML in pediatrics is distinctly unique with clinical, technical, and ethical nuances limiting the direct translation of ML tools developed for adults to pediatric populations. To our knowledge, no work has yet focused on outlining the unique considerations that need to be taken into account when designing and implementing ML in pediatrics.

Recent findings

The nature of varying developmental stages and the prominence of family-centered care lead to vastly different data-generating processes in pediatrics. Data heterogeneity and a lack of high-quality pediatric databases further complicate ML research. In order to address some of these nuances, we provide a common pipeline for clinicians and computer scientists to use as a foundation for structuring ML projects, and a framework for the translation of a developed model into clinical practice in pediatrics. Throughout these pathways, we also highlight ethical and legal considerations that must be taken into account when working with pediatric populations and data.

Summary

Here, we describe a comprehensive outline of special considerations required of ML in pediatrics from project ideation to implementation. We hope this review can serve as a high-level guideline for ML scientists and clinicians alike to identify applications in the pediatric setting, generate effective ML solutions, and subsequently deliver them to patients, families, and providers.

Machine Learning Predicts Prolonged Acute Hypoxemic Respiratory Failure in Pediatric Severe Influenza

Supplemental Digital Content is available in the text.

Influenza virus is a major cause of acute hypoxemic respiratory failure. Early identification of patients who will suffer severe complications can help stratify patients for clinical trials and plan for resource use in case of pandemic.

Prediction of Cardiac Arrest in the Emergency Department Based on Machine Learning and Sequential Characteristics: Model Development and Retrospective Clinical Validation Study

Background

The development and application of clinical prediction models using machine learning in clinical decision support systems is attracting increasing attention.

Objective

The aims of this study were to develop a prediction model for cardiac arrest in the emergency department (ED) using machine learning and sequential characteristics and to validate its clinical usefulness.

Methods

This retrospective study was conducted with ED patients at a tertiary academic hospital who suffered cardiac arrest. To resolve the class imbalance problem, sampling was performed using propensity score matching. The data set was chronologically allocated to a development cohort (years 2013 to 2016) and a validation cohort (year 2017). We trained three machine learning algorithms with repeated 10-fold cross-validation.

Results

The main performance parameters were the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC). The random forest algorithm (AUROC 0.97; AUPRC 0.86) outperformed the recurrent neural network (AUROC 0.95; AUPRC 0.82) and the logistic regression algorithm (AUROC 0.92; AUPRC=0.72). The performance of the model was maintained over time, with the AUROC remaining at least 80% across the monitored time points during the 24 hours before event occurrence.

Conclusions

We developed a prediction model of cardiac arrest in the ED using machine learning and sequential characteristics. The model was validated for clinical usefulness by chronological visualization focused on clinical usability.

Enabling individualised health in learning healthcare systems

The rising burden of healthcare costs suggests that the healthcare system could benefit from novel methods that allow for continuous learning to provide more data-driven, individualised care at lower costs and with improved outcomes. Here, we present our synergistic Learning approach for Prediction, Interpretation/Inference and Communication (Learning PIC) framework to address the challenges hindering the successful implementation of learning healthcare systems and to enable the effective delivery of evidence-based medicine.