Early prediction of circulatory failure in the intensive care unit using machine learning

Continuous prediction and clinical alarm management of late-onset sepsis in preterm infants using vital signs from a patient monitor

BACKGROUND AND OBJECTIVE

Continuous prediction of late-onset sepsis (LOS) could be helpful for improving clinical outcomes in neonatal intensive care units (NICU). This study aimed to develop an artificial intelligence (AI) model for assisting the bedside clinicians in successfully identifying infants at risk for LOS using non-invasive vital signs monitoring.

METHODS

In a retrospective study from the NICU of the Máxima Medical Center in Veldhoven, the Netherlands, a total of 492 preterm infants less than 32 weeks gestation were included between July 2016 and December 2018. Data on heart rate (HR), respiratory rate (RR), and oxygen saturation (SpO2) at 1 Hz were extracted from the patient monitor. We developed multiple AI models using 102 extracted features or raw time series to provide hourly LOS risk prediction. Shapley values were used to explain the model. For the best performing model, the effect of different vital signs and also the input type of signals on model performance was tested. To further assess the performance of applying the best performing model in a real-world clinical setting, we performed a simulation using four different alarm policies on continuous real-time predictions starting from three days after birth.

RESULTS

A total of 51 LOS patients and 68 controls were finally included according to the patient inclusion and exclusion criteria. When tested by seven-fold cross-validations, the mean (standard deviation) area under the receiver operating characteristic curve (AUC) six hours before CRASH was 0.875 (0.072) for the best performing model, compared to the other six models with AUC ranging from 0.782 (0.089) to 0.846 (0.083). The best performing model performed only slightly worse than the model learning from raw physiological waveforms (0.886 [0.068]), successfully detecting 96.1 % of LOS patients before CRASH. When setting the expected alarm window to 24 h and using a multi-threshold alarm policy, the sensitivity metric was 71.6 %, while the positive predictive value was 9.9 %, resulting in an average of 1.15 alarms per day per patient.

CONCLUSIONS

The proposed AI model, which learns from routinely collected vital signs, has the potential to assist clinicians in the early detection of LOS. Combined with interpretability and clinical alarm management, this model could be better translated into medical practice for future clinical implementation.

Tailoring Risk Prediction Models to Local Populations

Importance

Risk estimation is an integral part of cardiovascular care. Local recalibration of guideline-recommended models could address the limitations of existing tools.

Objective

To provide a machine learning (ML) approach to augment the performance of the American Heart Association's Predicting Risk of Cardiovascular Disease Events (AHA-PREVENT) equations when applied to a local population while preserving clinical interpretability.

Design, Setting, and Participants

This cohort study used a New England-based electronic health record cohort of patients without prior atherosclerotic cardiovascular disease (ASCVD) who had the data necessary to calculate the AHA-PREVENT 10-year risk of developing ASCVD in the event period (2007-2016). Patients with prior ASCVD events, death prior to 2007, or age 79 years or older in 2007 were subsequently excluded. The final study population of 95 326 patients was split into 3 nonoverlapping subsets for training, testing, and validation. The AHA-PREVENT model was adapted to this local population using the open-source ML model (MLM) Extreme Gradient Boosting model (XGBoost) with minimal predictor variables, including age, sex, and AHA-PREVENT. The MLM was monotonically constrained to preserve known associations between risk factors and ASCVD risk. Along with sex, race and ethnicity data from the electronic health record were collected to validate the performance of ASCVD risk prediction in subgroups. Data were analyzed from August 2021 to February 2024.

Main Outcomes and Measures

Consistent with the AHA-PREVENT model, ASCVD events were defined as the first occurrence of either nonfatal myocardial infarction, coronary artery disease, ischemic stroke, or cardiovascular death. Cardiovascular death was coded via government registries. Discrimination, calibration, and risk reclassification were assessed using the Harrell C index, a modified Hosmer-Lemeshow goodness-of-fit test and calibration curves, and reclassification tables, respectively.

Results

In the test set of 38 137 patients (mean [SD] age, 64.8 [6.9] years, 22 708 [59.5]% women and 15 429 [40.5%] men; 935 [2.5%] Asian, 2153 [5.6%] Black, 1414 [3.7%] Hispanic, 31 400 [82.3%] White, and 2235 [5.9%] other, including American Indian, multiple races, unspecified, and unrecorded, consolidated owing to small numbers), MLM-PREVENT had improved calibration (modified Hosmer-Lemeshow P > .05) compared to the AHA-PREVENT model across risk categories in the overall cohort (χ23 = 2.2; P = .53 vs χ23 > 16.3; P < .001) and sex subgroups (men: χ23 = 2.1; P = .55 vs χ23 > 16.3; P < .001; women: χ23 = 6.5; P = .09 vs. χ23 > 16.3; P < .001), while also surpassing a traditional recalibration approach. MLM-PREVENT maintained or improved AHA-PREVENT's calibration in Asian, Black, and White individuals. Both MLM-PREVENT and AHA-PREVENT performed equally well in discriminating risk (approximate ΔC index, ±0.01). Using a clinically significant 7.5% risk threshold, MLM-PREVENT reclassified a total of 11.5% of patients. We visualize the recalibration through MLM-PREVENT ASCVD risk charts that highlight preserved risk associations of the original AHA-PREVENT model.

Conclusions and Relevance

The interpretable ML approach presented in this article enhanced the accuracy of the AHA-PREVENT model when applied to a local population while still preserving the risk associations found by the original model. This method has the potential to recalibrate other established risk tools and is implementable in electronic health record systems for improved cardiovascular risk assessment.

Early Prediction of Cardiac Arrest in the Intensive Care Unit Using Explainable Machine Learning: Retrospective Study

BACKGROUND

Cardiac arrest (CA) is one of the leading causes of death among patients in the intensive care unit (ICU). Although many CA prediction models with high sensitivity have been developed to anticipate CA, their practical application has been challenging due to a lack of generalization and validation. Additionally, the heterogeneity among patients in different ICU subtypes has not been adequately addressed.

OBJECTIVE

This study aims to propose a clinically interpretable ensemble approach for the timely and accurate prediction of CA within 24 hours, regardless of patient heterogeneity, including variations across different populations and ICU subtypes. Additionally, we conducted patient-independent evaluations to emphasize the model's generalization performance and analyzed interpretable results that can be readily adopted by clinicians in real-time.

METHODS

Patients were retrospectively analyzed using data from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) and the eICU-Collaborative Research Database (eICU-CRD). To address the problem of underperformance, we constructed our framework using feature sets based on vital signs, multiresolution statistical analysis, and the Gini index, with a 12-hour window to capture the unique characteristics of CA. We extracted 3 types of features from each database to compare the performance of CA prediction between high-risk patient groups from MIMIC-IV and patients without CA from eICU-CRD. After feature extraction, we developed a tabular network (TabNet) model using feature screening with cost-sensitive learning. To assess real-time CA prediction performance, we used 10-fold leave-one-patient-out cross-validation and a cross-data set method. We evaluated MIMIC-IV and eICU-CRD across different cohort populations and subtypes of ICU within each database. Finally, external validation using the eICU-CRD and MIMIC-IV databases was conducted to assess the model's generalization ability. The decision mask of the proposed method was used to capture the interpretability of the model.

RESULTS

The proposed method outperformed conventional approaches across different cohort populations in both MIMIC-IV and eICU-CRD. Additionally, it achieved higher accuracy than baseline models for various ICU subtypes within both databases. The interpretable prediction results can enhance clinicians' understanding of CA prediction by serving as a statistical comparison between non-CA and CA groups. Next, we tested the eICU-CRD and MIMIC-IV data sets using models trained on MIMIC-IV and eICU-CRD, respectively, to evaluate generalization ability. The results demonstrated superior performance compared with baseline models.

CONCLUSIONS

Our novel framework for learning unique features provides stable predictive power across different ICU environments. Most of the interpretable global information reveals statistical differences between CA and non-CA groups, demonstrating its utility as an indicator for clinical decisions. Consequently, the proposed CA prediction system is a clinically validated algorithm that enables clinicians to intervene early based on CA prediction information and can be applied to clinical trials in digital health.

An open-source framework for end-to-end analysis of electronic health record data

With progressive digitalization of healthcare systems worldwide, large-scale collection of electronic health records (EHRs) has become commonplace. However, an extensible framework for comprehensive exploratory analysis that accounts for data heterogeneity is missing. Here we introduce ehrapy, a modular open-source Python framework designed for exploratory analysis of heterogeneous epidemiology and EHR data. ehrapy incorporates a series of analytical steps, from data extraction and quality control to the generation of low-dimensional representations. Complemented by rich statistical modules, ehrapy facilitates associating patients with disease states, differential comparison between patient clusters, survival analysis, trajectory inference, causal inference and more. Leveraging ontologies, ehrapy further enables data sharing and training EHR deep learning models, paving the way for foundational models in biomedical research. We demonstrate ehrapy's features in six distinct examples. We applied ehrapy to stratify patients affected by unspecified pneumonia into finer-grained phenotypes. Furthermore, we reveal biomarkers for significant differences in survival among these groups. Additionally, we quantify medication-class effects of pneumonia medications on length of stay. We further leveraged ehrapy to analyze cardiovascular risks across different data modalities. We reconstructed disease state trajectories in patients with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) based on imaging data. Finally, we conducted a case study to demonstrate how ehrapy can detect and mitigate biases in EHR data. ehrapy, thus, provides a framework that we envision will standardize analysis pipelines on EHR data and serve as a cornerstone for the community.

Transatlantic transferability and replicability of machine-learning algorithms to predict mental health crises

Transferring and replicating predictive algorithms across healthcare systems constitutes a unique yet crucial challenge that needs to be addressed to enable the widespread adoption of machine learning in healthcare. In this study, we explored the impact of important differences across healthcare systems and the associated Electronic Health Records (EHRs) on machine-learning algorithms to predict mental health crises, up to 28 days in advance. We evaluated both the transferability and replicability of such machine learning models, and for this purpose, we trained six models using features and methods developed on EHR data from the Birmingham and Solihull Mental Health NHS Foundation Trust in the UK. These machine learning models were then used to predict the mental health crises of 2907 patients seen at the Rush University System for Health in the US between 2018 and 2020. The best one was trained on a combination of US-specific structured features and frequency features from anonymized patient notes and achieved an AUROC of 0.837. A model with comparable performance, originally trained using UK structured data, was transferred and then tuned using US data, achieving an AUROC of 0.826. Our findings establish the feasibility of transferring and replicating machine learning models to predict mental health crises across diverse hospital systems.

Enhancing Patient Selection in Sepsis Clinical Trials Design Through an AI Enrichment Strategy: Algorithm Development and Validation

BACKGROUND

Sepsis is a heterogeneous syndrome, and enrollment of more homogeneous patients is essential to improve the efficiency of clinical trials. Artificial intelligence (AI) has facilitated the identification of homogeneous subgroups, but how to estimate the uncertainty of the model outputs when applying AI to clinical decision-making remains unknown.

OBJECTIVE

We aimed to design an AI-based model for purposeful patient enrollment, ensuring that a patient with sepsis recruited into a trial would still be persistently ill by the time the proposed therapy could impact patient outcome. We also expected that the model could provide interpretable factors and estimate the uncertainty of the model outputs at a customized confidence level.

METHODS

In this retrospective study, 9135 patients with sepsis requiring vasopressor treatment within 24 hours after sepsis onset were enrolled from Beth Israel Deaconess Medical Center. This cohort was used for model development, and 10-fold cross-validation with 50 repeats was used for internal validation. In total, 3743 patients with sepsis from the eICU Collaborative Research Database were used as the external validation cohort. All included patients with sepsis were stratified based on disease progression trajectories: rapid death, recovery, and persistent ill. A total of 148 variables were selected for predicting the 3 trajectories. Four machine learning algorithms with 3 different setups were used. We estimated the uncertainty of the model outputs using conformal prediction (CP). The Shapley Additive Explanations method was used to explain the model.

RESULTS

The multiclass gradient boosting machine was identified as the best-performing model with good discrimination and calibration performance in both validation cohorts. The mean area under the receiver operating characteristic curve with SD was 0.906 (0.018) for rapid death, 0.843 (0.008) for recovery, and 0.807 (0.010) for persistent ill in the internal validation cohort. In the external validation cohort, the mean area under the receiver operating characteristic curve (SD) was 0.878 (0.003) for rapid death, 0.764 (0.008) for recovery, and 0.696 (0.007) for persistent ill. The maximum norepinephrine equivalence, total urine output, Acute Physiology Score III, mean systolic blood pressure, and the coefficient of variation of oxygen saturation contributed the most. Compared to the model without CP, using the model with CP at a mixed confidence approach reduced overall prediction errors by 27.6% (n=62) and 30.7% (n=412) in the internal and external validation cohorts, respectively, as well as enabled the identification of more potentially persistent ill patients.

CONCLUSIONS

The implementation of our model has the potential to reduce heterogeneity and enroll more homogeneous patients in sepsis clinical trials. The use of CP for estimating the uncertainty of the model outputs allows for a more comprehensive understanding of the model's reliability and assists in making informed decisions based on the predicted outcomes.

Comparison of Differences in Cohort (Forward) and Case Control (Backward) Methodologic Approaches for Validation of the Hypotension Prediction Index

BACKGROUND

The Hypotension Prediction Index (the index) software is a machine learning algorithm that detects physiologic changes that may lead to hypotension. The original validation used a case control (backward) analysis that has been suggested to be biased. This study therefore conducted a cohort (forward) analysis and compared this to the original validation technique.

METHODS

A retrospective analysis of data from previously reported studies was conducted. All data were analyzed identically with two different methodologies, and receiver operating characteristic curves were constructed. Both backward and forward analyses were performed to examine differences in area under the receiver operating characteristic curves for the Hypotension Prediction Index and other hemodynamic variables to predict a mean arterial pressure (MAP) less than 65 mmHg for at least 1 min 5, 10, and 15 min in advance.

RESULTS

The analysis included 2,022 patients, yielding 4,152,124 measurements taken at 20-s intervals. The area under the curve for the index predicting hypotension analyzed by backward and forward methodologies respectively was 0.957 (95% CI, 0.947 to 0.964) versus 0.923 (95% CI, 0.912 to 0.933) 5 min in advance, 0.933 (95% CI, 0.924 to 0.942) versus 0.923 (95% CI, 0.911 to 0.933) 10 min in advance, and 0.929 (95% CI, 0.918 to 0.938) versus 0.926 (95% CI, 0.914 to 0.937) 15 min in advance. No variable other than MAP had an area under the curve greater than 0.7. The areas under the curve using forward analysis for MAP predicting hypotension 5, 10, and 15 min in advance were 0.932 (95% CI, 0.920 to 0.940), 0.929 (95% CI, 0.918 to 0.938), and 0.932 (95% CI, 0.921 to 0.940), respectively. The R2 for the variation in the index due to MAP was 0.77.

CONCLUSIONS

Using an updated methodology, the study found that the utility of the Hypotension Prediction Index to predict future hypotensive events is high, with an area under the receiver operating characteristics curve similar to that of the original validation method.

EDITOR’S PERSPECTIVE

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.

Applying AI to Safely and Effectively Scale Care to Address Chronic MSK Conditions

Background/Objectives: The rising prevalence of musculoskeletal (MSK) conditions has not been balanced by a sufficient increase in healthcare providers. Scalability challenges are being addressed through the use of artificial intelligence (AI) in some healthcare sectors, with this showing potential to also improve MSK care. Digital care programs (DCP) generate automatically collected data, thus making them ideal candidates for AI implementation into workflows, with the potential to unlock care scalability. In this study, we aimed to assess the impact of scaling care through AI in patient outcomes, engagement, satisfaction, and adverse events. Methods: Post hoc analysis of a prospective, pre-post cohort study assessing the impact on outcomes after a 2.3-fold increase in PT-to-patient ratio, supported by the implementation of a machine learning-based tool to assist physical therapists (PTs) in patient care management. The intervention group (IG) consisted of a DCP supported by an AI tool, while the comparison group (CG) consisted of the DCP alone. The primary outcome concerned the pain response rate (reaching a minimal clinically important change of 30%). Other outcomes included mental health, program engagement, satisfaction, and the adverse event rate. Results: Similar improvements in pain response were observed, regardless of the group (response rate: 64% vs. 63%; p = 0.399). Equivalent recoveries were also reported in mental health outcomes, specifically in anxiety (p = 0.928) and depression (p = 0.187). Higher completion rates were observed in the IG (79.9% (N = 19,252) vs. CG 70.1% (N = 8489); p < 0.001). Patient engagement remained consistent in both groups, as well as high satisfaction (IG: 8.76/10, SD 1.75 vs. CG: 8.60/10, SD 1.76; p = 0.021). Intervention-related adverse events were rare and even across groups (IG: 0.58% and CG 0.69%; p = 0.231). Conclusions: The study underscores the potential of scaling MSK care that is supported by AI without compromising patient outcomes, despite the increase in PT-to-patient ratios.

The Impact of Multi-Institution Datasets on the Generalizability of Machine Learning Prediction Models in the ICU

OBJECTIVES

To evaluate the transferability of deep learning (DL) models for the early detection of adverse events to previously unseen hospitals.

DESIGN

Retrospective observational cohort study utilizing harmonized intensive care data from four public datasets.

SETTING

ICUs across Europe and the United States.

PATIENTS

Adult patients admitted to the ICU for at least 6 hours who had good data quality.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Using carefully harmonized data from a total of 334,812 ICU stays, we systematically assessed the transferability of DL models for three common adverse events: death, acute kidney injury (AKI), and sepsis. We tested whether using more than one data source and/or algorithmically optimizing for generalizability during training improves model performance at new hospitals. We found that models achieved high area under the receiver operating characteristic (AUROC) for mortality (0.838-0.869), AKI (0.823-0.866), and sepsis (0.749-0.824) at the training hospital. As expected, AUROC dropped when models were applied at other hospitals, sometimes by as much as -0.200. Using more than one dataset for training mitigated the performance drop, with multicenter models performing roughly on par with the best single-center model. Dedicated methods promoting generalizability did not noticeably improve performance in our experiments.

CONCLUSIONS

Our results emphasize the importance of diverse training data for DL-based risk prediction. They suggest that as data from more hospitals become available for training, models may become increasingly generalizable. Even so, good performance at a new hospital still depended on the inclusion of compatible hospitals during training.

Recent Advances in Digital Technology in Implant Dentistry

Digital technology has emerged as a transformative tool in dental implantation, profoundly enhancing accuracy and effectiveness across multiple facets, such as diagnosis, preoperative treatment planning, surgical procedures, and restoration delivery. The multiple integration of radiographic data and intraoral data, sometimes with facial scan data or electronic facebow through virtual planning software, enables comprehensive 3-dimensional visualization of the hard and soft tissue and the position of future restoration, resulting in heightened diagnostic precision. In virtual surgery design, the incorporation of both prosthetic arrangement and individual anatomical details enables the virtual execution of critical procedures (e.g., implant placement, extended applications, etc.) through analysis of cross-sectional images and the reconstruction of 3-dimensional surface models. After verification, the utilization of digital technology including templates, navigation, combined techniques, and implant robots achieved seamless transfer of the virtual treatment plan to the actual surgical sites, ultimately leading to enhanced surgical outcomes with highly improved accuracy. In restoration delivery, digital techniques for impression, shade matching, and prosthesis fabrication have advanced, enabling seamless digital data conversion and efficient communication among clinicians and technicians. Compared with clinical medicine, artificial intelligence (AI) technology in dental implantology primarily focuses on diagnosis and prediction. AI-supported preoperative planning and surgery remain in developmental phases, impeded by the complexity of clinical cases and ethical considerations, thereby constraining widespread adoption.

A common data model for the standardization of intensive care unit medication features

Objective

Common data models provide a standard means of describing data for artificial intelligence (AI) applications, but this process has never been undertaken for medications used in the intensive care unit (ICU). We sought to develop a common data model (CDM) for ICU medications to standardize the medication features needed to support future ICU AI efforts.

Materials and Methods

A 9-member, multi-professional team of ICU clinicians and AI experts conducted a 5-round modified Delphi process employing conference calls, web-based communication, and electronic surveys to define the most important medication features for AI efforts. Candidate ICU medication features were generated through group discussion and then independently scored by each team member based on relevance to ICU clinical decision-making and feasibility for collection and coding. A key consideration was to ensure the final ontology both distinguished unique medications and met Findable, Accessible, Interoperable, and Reusable (FAIR) guiding principles.

Results

Using a list of 889 ICU medications, the team initially generated 106 different medication features, and 71 were ranked as being core features for the CDM. Through this process, 106 medication features were assigned to 2 key feature domains: drug product-related (n = 43) and clinical practice-related (n = 63). Each feature included a standardized definition and suggested response values housed in the electronic data library. This CDM for ICU medications is available online.

Conclusion

The CDM for ICU medications represents an important first step for the research community focused on exploring how AI can improve patient outcomes and will require ongoing engagement and refinement.

Early Warning Systems for Critical Illness Outside the Intensive Care Unit

Early warning systems (EWSs) are designed and deployed to create a rapid assessment and response for patients with clinical deterioration outside the intensive care unit (ICU). These models incorporate patient-level data such as vital signs and laboratory values to detect or prevent adverse clinical events, such as vital signs and laboratories to allow detection and prevention of adverse clinical events such as cardiac arrest, intensive care transfer, or sepsis. The applicability, development, clinical utility, and general perception of EWS in clinical practice vary widely. Here, we review the field as it has grown from early vital sign-based scoring systems to contemporary multidimensional algorithms and predictive technologies for clinical decompensation outside the ICU.

An empirical study on KDIGO-defined acute kidney injury prediction in the intensive care unit

MOTIVATION

Acute kidney injury (AKI) is a syndrome that affects a large fraction of all critically ill patients, and early diagnosis to receive adequate treatment is as imperative as it is challenging to make early. Consequently, machine learning approaches have been developed to predict AKI ahead of time. However, the prevalence of AKI is often underestimated in state-of-the-art approaches, as they rely on an AKI event annotation solely based on creatinine, ignoring urine output.

We construct and evaluate early warning systems for AKI in a multi-disciplinary ICU setting, using the complete KDIGO definition of AKI. We propose several variants of gradient-boosted decision tree (GBDT)-based models, including a novel time-stacking based approach. A state-of-the-art LSTM-based model previously proposed for AKI prediction is used as a comparison, which was not specifically evaluated in ICU settings yet.

RESULTS

We find that optimal performance is achieved by using GBDT with the time-based stacking technique (AUPRC = 65.7%, compared with the LSTM-based model's AUPRC = 62.6%), which is motivated by the high relevance of time since ICU admission for this task. Both models show mildly reduced performance in the limited training data setting, perform fairly across different subcohorts, and exhibit no issues in gender transfer.

Following the official KDIGO definition substantially increases the number of annotated AKI events. In our study GBDTs outperform LSTM models for AKI prediction. Generally, we find that both model types are robust in a variety of challenging settings arising for ICU data.

AVAILABILITY AND IMPLEMENTATION

The code to reproduce the findings of our manuscript can be found at: https://github.com/ratschlab/AKI-EWS.

Enhancing the diagnosis of functionally relevant coronary artery disease with machine learning

Functionally relevant coronary artery disease (fCAD) can result in premature death or nonfatal acute myocardial infarction. Its early detection is a fundamentally important task in medicine. Classical detection approaches suffer from limited diagnostic accuracy or expose patients to possibly harmful radiation. Here we show how machine learning (ML) can outperform cardiologists in predicting the presence of stress-induced fCAD in terms of area under the receiver operating characteristic (AUROC: 0.71 vs. 0.64, p = 4.0E-13). We present two ML approaches, the first using eight static clinical variables, whereas the second leverages electrocardiogram signals from exercise stress testing. At a target post-test probability for fCAD of <15%, ML facilitates a potential reduction of imaging procedures by 15-17% compared to the cardiologist's judgement. Predictive performance is validated on an internal temporal data split as well as externally. We also show that combining clinical judgement with conventional ML and deep learning using logistic regression results in a mean AUROC of 0.74.

Intelligent alert system for predicting invasive mechanical ventilation needs via noninvasive parameters: employing an integrated machine learning method with integration of multicenter databases

The use of invasive mechanical ventilation (IMV) is crucial in rescuing patients with respiratory dysfunction. Accurately predicting the demand for IMV is vital for clinical decision-making. However, current techniques are invasive and challenging to implement in pre-hospital and emergency rescue settings. To address this issue, a real-time prediction method utilizing only non-invasive parameters was developed to forecast IMV demand in this study. The model introduced the concept of real-time warning and leveraged the advantages of machine learning and integrated methods, achieving an AUC value of 0.935 (95% CI 0.933-0.937). The AUC value for the multi-center validation using the AmsterdamUMCdb database was 0.727, surpassing the performance of traditional risk adjustment algorithms (OSI(oxygenation saturation index): 0.608, P/F(oxygenation index): 0.558). Feature weight analysis demonstrated that BMI, Gcsverbal, and age significantly contributed to the model's decision-making. These findings highlight the substantial potential of a machine learning real-time dynamic warning model that solely relies on non-invasive parameters to predict IMV demand. Such a model can provide technical support for predicting the need for IMV in pre-hospital and disaster scenarios.

Predictive approach for liberation from acute dialysis in ICU patients using interpretable machine learning

Renal recovery following dialysis-requiring acute kidney injury (AKI-D) is a vital clinical outcome in critical care, yet it remains an understudied area. This retrospective cohort study, conducted in a medical center in Taiwan from 2015 to 2020, enrolled patients with AKI-D during intensive care unit stays. We aimed to develop and temporally test models for predicting dialysis liberation before hospital discharge using machine learning algorithms and explore early predictors. The dataset comprised 90 routinely collected variables within the first three days of dialysis initiation. Out of 1,381 patients who received acute dialysis, 27.3% experienced renal recovery. The cohort was divided into the training group (N = 1135) and temporal testing group (N = 251). The models demonstrated good performance, with an area under the receiver operating characteristic curve of 0.85 (95% CI, 0.81-0.88) and an area under the precision-recall curve of 0.69 (95% CI, 0.62-0.76) for the XGBoost model. Key predictors included urine volume, Charlson comorbidity index, vital sign derivatives (trend of respiratory rate and SpO2), and lactate levels. We successfully developed early prediction models for renal recovery by integrating early changes in vital signs and inputs/outputs, which have the potential to aid clinical decision-making in the ICU.

[Data-driven intensive care: a lack of comprehensive datasets]

Intensive care units provide a data-rich environment with the potential to generate datasets in the realm of big data, which could be utilized to train powerful machine learning (ML) models. However, the currently available datasets are too small and exhibit too little diversity due to their limitation to individual hospitals. This lack of extensive and varied datasets is a primary reason for the limited generalizability and resulting low clinical utility of current ML models. Often, these models are based on data from single centers and suffer from poor external validity. There is an urgent need for the development of large-scale, multicentric, and multinational datasets. Ensuring data protection and minimizing re-identification risks pose central challenges in this process. The "Amsterdam University Medical Center database (AmsterdamUMCdb)" and the "Salzburg Intensive Care database (SICdb)" demonstrate that open access datasets are possible in Europe while complying with the data protection regulations of the General Data Protection Regulation (GDPR). Another challenge in building intensive care datasets is the absence of semantic definitions in the source data and the heterogeneity of data formats. Establishing binding industry standards for the semantic definition is crucial to ensure seamless semantic interoperability between datasets.

Using routinely collected clinical data for circadian medicine: A review of opportunities and challenges

A wealth of data is available from electronic health records (EHR) that are collected as part of routine clinical care in hospitals worldwide. These rich, longitudinal data offer an attractive object of study for the field of circadian medicine, which aims to translate knowledge of circadian rhythms to improve patient health. This narrative review aims to discuss opportunities for EHR in studies of circadian medicine, highlight the methodological challenges, and provide recommendations for using these data to advance the field. In the existing literature, we find that data collected in real-world clinical settings have the potential to shed light on key questions in circadian medicine, including how 24-hour rhythms in clinical features are associated with-or even predictive of-health outcomes, whether the effect of medication or other clinical activities depend on time of day, and how circadian rhythms in physiology may influence clinical reference ranges or sampling protocols. However, optimal use of EHR to advance circadian medicine requires careful consideration of the limitations and sources of bias that are inherent to these data sources. In particular, time of day influences almost every interaction between a patient and the healthcare system, creating operational 24-hour patterns in the data that have little or nothing to do with biology. Addressing these challenges could help to expand the evidence base for the use of EHR in the field of circadian medicine.

Experts vs. machine - comparison of machine learning to expert-informed prediction of outcome after major liver surgery

BACKGROUND

Machine learning (ML) has been successfully implemented for classification tasks (e.g., cancer diagnosis). ML performance for more challenging predictions is largely unexplored. This study's objective was to compare machine learning vs. expert-informed predictions for surgical outcome in patients undergoing major liver surgery.

METHODS

Single tertiary center data on preoperative parameters and postoperative complications for elective hepatic surgery patients were included (2008-2021). Expert-informed prediction models were established on 14 parameters identified by two expert liver surgeons to impact on postoperative outcome. ML models used all available preoperative patient variables (n = 62). Model performance was compared for predicting 3-month postoperative overall morbidity. Temporal validation and additional analysis in major liver resection patients were conducted.

RESULTS

889 patients included. Expert-informed models showed low average bias (2-5 CCI points) with high over/underprediction. ML models performed similarly: average prediction 5-10 points higher than observed CCI values with high variability (95% CI -30 to 50). No performance improvement for major liver surgery patients.

CONCLUSION

No clinical relevance in the application of ML for predicting postoperative overall morbidity was found. Despite being a novel hype, ML has the potential for application in clinical practice. However, at this stage it does not replace established approaches of prediction modelling.

Simplifying Data Analysis in Biomedical Research: An Automated, User-Friendly Tool

Robust data normalization and analysis are pivotal in biomedical research to ensure that observed differences in populations are directly attributable to the target variable, rather than disparities between control and study groups. ArsHive addresses this challenge using advanced algorithms to normalize populations (e.g., control and study groups) and perform statistical evaluations between demographic, clinical, and other variables within biomedical datasets, resulting in more balanced and unbiased analyses. The tool's functionality extends to comprehensive data reporting, which elucidates the effects of data processing, while maintaining dataset integrity. Additionally, ArsHive is complemented by A.D.A. (Autonomous Digital Assistant), which employs OpenAI's GPT-4 model to assist researchers with inquiries, enhancing the decision-making process. In this proof-of-concept study, we tested ArsHive on three different datasets derived from proprietary data, demonstrating its effectiveness in managing complex clinical and therapeutic information and highlighting its versatility for diverse research fields.

Development and validation of a machine learning model for prediction of type 2 diabetes in patients with mental illness

BACKGROUND

Type 2 diabetes (T2D) is approximately twice as common among individuals with mental illness compared with the background population, but may be prevented by early intervention on lifestyle, diet, or pharmacologically. Such prevention relies on identification of those at elevated risk (prediction). The aim of this study was to develop and validate a machine learning model for prediction of T2D among patients with mental illness.

METHODS

The study was based on routine clinical data from electronic health records from the psychiatric services of the Central Denmark Region. A total of 74,880 patients with 1.59 million psychiatric service contacts were included in the analyses. We created 1343 potential predictors from 51 source variables, covering patient-level information on demographics, diagnoses, pharmacological treatment, and laboratory results. T2D was operationalised as HbA1c ≥48 mmol/mol, fasting plasma glucose ≥7.0 mmol/mol, oral glucose tolerance test ≥11.1 mmol/mol or random plasma glucose ≥11.1 mmol/mol. Two machine learning models (XGBoost and regularised logistic regression) were trained to predict T2D based on 85% of the included contacts. The predictive performance of the best performing model was tested on the remaining 15% of the contacts.

RESULTS

The XGBoost model detected patients at high risk 2.7 years before T2D, achieving an area under the receiver operating characteristic curve of 0.84. Of the 996 patients developing T2D in the test set, the model issued at least one positive prediction for 305 (31%).

CONCLUSION

A machine learning model can accurately predict development of T2D among patients with mental illness based on routine clinical data from electronic health records. A decision support system based on such a model may inform measures to prevent development of T2D in this high-risk population.

[Artificial intelligence in intensive care medicine]

The integration of artificial intelligence (AI) into intensive care medicine has made considerable progress in recent studies, particularly in the areas of predictive analytics, early detection of complications, and the development of decision support systems. The main challenges remain availability and quality of data, reduction of bias and the need for explainable results from algorithms and models. Methods to explain these systems are essential to increase trust, understanding, and ethical considerations among healthcare professionals and patients. Proper training of healthcare professionals in AI principles, terminology, ethical considerations, and practical application is crucial for the successful use of AI. Careful assessment of the impact of AI on patient autonomy and data protection is essential for its responsible use in intensive care medicine. A balance between ethical and practical considerations must be maintained to ensure patient-centered care while complying with data protection regulations. Synergistic collaboration between clinicians, AI engineers, and regulators is critical to realizing the full potential of AI in intensive care medicine and maximizing its positive impact on patient care. Future research and development efforts should focus on improving AI models for real-time predictions, increasing the accuracy and utility of AI-based closed-loop systems, and overcoming ethical, technical, and regulatory challenges, especially in generative AI systems.

ARTEMIS: An alarm threshold and policy mining system for the intensive care unit

BACKGROUND

Alarm fatigue is a major technology-induced hazard for patients and staff in intensive care units. Too many - mostly unnecessary - alarms cause desensitisation and lack of response in medical staff. Unsuitable alarm policies are one reason for alarm fatigue. But changing alarm policies is a delicate issue since it concerns patient safety.

OBJECTIVE

We present ARTEMIS, a novel, computer-aided clinical decision support system for policy makers that can help to considerably improve alarm policies using data from hospital information systems.

METHODS

Policy makers can use different policy components from ARTEMIS' internal library to assemble tailor-made alarm policies for their intensive care units. Alternatively, policy makers can provide even more highly customised policy components as Python functions using data the hospital information systems. This can even include machine learning models - for example for setting alarm thresholds. Finally, policy makers can evaluate their system of policies and compare the resulting alarm loads.

RESULTS

ARTEMIS reports and compares numbers of alarms caused by different alarm policies for an easily adaptable target population. ARTEMIS can compare policies side-by-side and provides grid comparisons and heat maps for parameter optimisation. For example, we found that the utility of alarm delays varies based on target population. Furthermore, policy makers can introduce virtual parameters that are not in the original data by providing a formula to compute them. Virtual parameters help measuring and alarming on the right metric, even if the patient monitors do not directly measure this metric.

CONCLUSION

ARTEMIS does not release the policy maker from assessing the policy from a medical standpoint. But as a knowledge discovery and clinical decision support system, it provides a strong quantitative foundation for medical decisions. At comparatively low cost of implementation, ARTEMIS can have a substantial impact on patients and staff alike - with organisational, economic, and clinical benefits for the implementing hospital.

Unsupervised machine learning analysis to identify patterns of ICU medication use for fluid overload prediction

INTRODUCTION

Intravenous (IV) medications are a fundamental cause of fluid overload (FO) in the intensive care unit (ICU); however, the association between IV medication use (including volume), administration timing, and FO occurrence remains unclear.

METHODS

This retrospective cohort study included consecutive adults admitted to an ICU ≥72 hours with available fluid balance data. FO was defined as a positive fluid balance ≥7% of admission body weight within 72 hours of ICU admission. After reviewing medication administration record (MAR) data in three-hour periods, IV medication exposure was categorized into clusters using principal component analysis (PCA) and Restricted Boltzmann Machine (RBM). Medication regimens of patients with and without FO were compared within clusters to assess for temporal clusters associated with FO using the Wilcoxon rank sum test. Exploratory analyses of the medication cluster most associated with FO for medications frequently appearing and used in the first 24 hours was conducted.

RESULTS

FO occurred in 127/927 (13.7%) of the patients enrolled. Patients received a median (IQR) of 31 (13-65) discrete IV medication administrations over the 72-hour period. Across all 47,803 IV medication administrations, ten unique IV medication clusters were identified with 121-130 medications in each cluster. Among the ten clusters, cluster 7 had the greatest association with FO; the mean number of cluster 7 medications received was significantly greater in patients in the FO cohort compared to patients who did not experience FO (25.6 vs.10.9. p<0.0001). 51 of the 127 medications in cluster 7 (40.2%) appeared in > 5 separate 3-hour periods during the 72-hour study window. The most common cluster 7 medications included continuous infusions, antibiotics, and sedatives/analgesics. Addition of cluster 7 medications to a prediction model with APACHE II score and receipt of diuretics improved the ability for the model to predict fluid overload (AUROC 5.65, p =0.0004).

CONCLUSIONS

Using ML approaches, a unique IV medication cluster was strongly associated with FO. Incorporation of this cluster improved the ability to predict development of fluid overload in ICU patients compared with traditional prediction models. This method may be further developed into real-time clinical applications to improve early detection of adverse outcomes.

Early prediction of ventricular peritoneal shunt dependency in aneurysmal subarachnoid haemorrhage patients by recurrent neural network-based machine learning using routine intensive care unit data

Aneurysmal subarachnoid haemorrhage (aSAH) can lead to complications such as acute hydrocephalic congestion. Treatment of this acute condition often includes establishing an external ventricular drainage (EVD). However, chronic hydrocephalus develops in some patients, who then require placement of a permanent ventriculoperitoneal (VP) shunt. The aim of this study was to employ recurrent neural network (RNN)-based machine learning techniques to identify patients who require VP shunt placement at an early stage. This retrospective single-centre study included all patients who were diagnosed with aSAH and treated in the intensive care unit (ICU) between November 2010 and May 2020 (n = 602). More than 120 parameters were analysed, including routine neurocritical care data, vital signs and blood gas analyses. Various machine learning techniques, including RNNs and gradient boosting machines, were evaluated for their ability to predict VP shunt dependency. VP-shunt dependency could be predicted using an RNN after just one day of ICU stay, with an AUC-ROC of 0.77 (CI: 0.75-0.79). The accuracy of the prediction improved after four days of observation (Day 4: AUC-ROC 0.81, CI: 0.79-0.84). At that point, the accuracy of the prediction was 76% (CI: 75.98-83.09%), with a sensitivity of 85% (CI: 83-88%) and a specificity of 74% (CI: 71-78%). RNN-based machine learning has the potential to predict VP shunt dependency on Day 4 after ictus in aSAH patients using routine data collected in the ICU. The use of machine learning may allow early identification of patients with specific therapeutic needs and accelerate the execution of required procedures.

External Validation of Deep Learning-Based Cardiac Arrest Risk Management System for Predicting In-Hospital Cardiac Arrest in Patients Admitted to General Wards Based on Rapid Response System Operating and Nonoperating Periods: A Single-Center Study

OBJECTIVES

The limitations of current early warning scores have prompted the development of deep learning-based systems, such as deep learning-based cardiac arrest risk management systems (DeepCARS). Unfortunately, in South Korea, only two institutions operate 24-hour Rapid Response System (RRS), whereas most hospitals have part-time or no RRS coverage at all. This study validated the predictive performance of DeepCARS during RRS operation and nonoperation periods and explored its potential beyond RRS operating hours.

DESIGN

Retrospective cohort study.

SETTING

In this 1-year retrospective study conducted at Yonsei University Health System Severance Hospital in South Korea, DeepCARS was compared with conventional early warning systems for predicting in-hospital cardiac arrest (IHCA). The study focused on adult patients admitted to the general ward, with the primary outcome being IHCA-prediction performance within 24 hours of the alarm.

PATIENTS

We analyzed the data records of adult patients admitted to a general ward from September 1, 2019, to August 31, 2020.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Performance evaluation was conducted separately for the operational and nonoperational periods of the RRS, using the area under the receiver operating characteristic curve (AUROC) as the metric. DeepCARS demonstrated a superior AUROC as compared with the Modified Early Warning Score (MEWS) and the National Early Warning Score (NEWS), both during RRS operating and nonoperating hours. Although the MEWS and NEWS exhibited varying performance across the two periods, DeepCARS showed consistent performance.

CONCLUSIONS

The accuracy and efficiency for predicting IHCA of DeepCARS were superior to that of conventional methods, regardless of whether the RRS was in operation. These findings emphasize that DeepCARS is an effective screening tool suitable for hospitals with full-time RRS, part-time RRS, and even those without any RRS.

Explainable Machine Learning Models for Rapid Risk Stratification in the Emergency Department: A Multicenter Study

BACKGROUND

Risk stratification of patients presenting to the emergency department (ED) is important for appropriate triage. Diagnostic laboratory tests are an essential part of the workup and risk stratification of these patients. Using machine learning, the prognostic power and clinical value of these tests can be amplified greatly. In this study, we applied machine learning to develop an accurate and explainable clinical decision support tool model that predicts the likelihood of 31-day mortality in ED patients (the RISKINDEX). This tool was developed and evaluated in four Dutch hospitals.

METHODS

Machine learning models included patient characteristics and available laboratory data collected within the first 2 h after ED presentation, and were trained using 5 years of data from consecutive ED patients from the Maastricht University Medical Center (Maastricht), Meander Medical Center (Amersfoort), and Zuyderland Medical Center (Sittard and Heerlen). A sixth year of data was used to evaluate the models using area under the receiver-operating-characteristic curve (AUROC) and calibration curves. The Shapley additive explanations (SHAP) algorithm was used to obtain explainable machine learning models.

RESULTS

The present study included 266 327 patients with 7.1 million laboratory results available. Models show high diagnostic performance with AUROCs of 0.94, 0.98, 0.88, and 0.90 for Maastricht, Amersfoort, Sittard and Heerlen, respectively. The SHAP algorithm was utilized to visualize patient characteristics and laboratory data patterns that underlie individual RISKINDEX predictions.

CONCLUSIONS

Our clinical decision support tool has excellent diagnostic performance in predicting 31-day mortality in ED patients. Follow-up studies will assess whether implementation of these algorithms can improve clinically relevant end points.

The SingHealth Perioperative and Anesthesia Subject Area Registry (PASAR), a large-scale perioperative data mart and registry

BACKGROUND

To enhance perioperative outcomes, a perioperative registry that integrates high-quality real-world data throughout the perioperative period is essential. Singapore General Hospital established the Perioperative and Anesthesia Subject Area Registry (PASAR) to unify data from the preoperative, intraoperative, and postoperative stages. This study presents the methodology employed to create this database.

METHODS

Since 2016, data from surgical patients have been collected from the hospital electronic medical record systems, de-identified, and stored securely in compliance with privacy and data protection laws. As a representative sample, data from initiation in 2016 to December 2022 were collected.

RESULTS

As of December 2022, PASAR data comprise 26 tables, encompassing 153,312 patient admissions and 168,977 operation sessions. For this period, the median age of the patients was 60.0 years, sex distribution was balanced, and the majority were Chinese. Hypertension and cardiovascular comorbidities were also prevalent. Information including operation type and time, intensive care unit (ICU) length of stay, and 30-day and 1-year mortality rates were collected. Emergency surgeries resulted in longer ICU stays, but shorter operation times than elective surgeries.

CONCLUSIONS

The PASAR provides a comprehensive and automated approach to gathering high-quality perioperative patient data.

Heart rate complexity helps mortality prediction in the intensive care unit: A pilot study using artificial intelligence

BACKGROUND

In intensive care units (ICUs), accurate mortality prediction is crucial for effective patient management and resource allocation. The Simplified Acute Physiology Score II (SAPS-2), though commonly used, relies heavily on comprehensive clinical data and blood samples. This study sought to develop an artificial intelligence (AI) model utilizing key hemodynamic parameters to predict ICU mortality within the first 24 h and assess its performance relative to SAPS-2.

METHODS

We conducted an analysis of select hemodynamic parameters and the structure of heart rate curves to identify potential predictors of ICU mortality. A machine-learning model was subsequently trained and validated on distinct patient cohorts. The AI algorithm's performance was then compared to the SAPS-2, focusing on classification accuracy, calibration, and generalizability.

MEASUREMENTS AND MAIN RESULTS

The study included 1298 ICU admissions from March 27th, 2015, to March 27th, 2017. An additional cohort from 2022 to 2023 comprised 590 patients, resulting in a total dataset of 1888 patients. The observed mortality rate stood at 24.0%. Key determinants of mortality were the Glasgow Coma Scale score, heart rate complexity, patient age, duration of diastolic blood pressure below 50 mmHg, heart rate variability, and specific mean and systolic blood pressure thresholds. The AI model, informed by these determinants, exhibited a performance profile in predicting mortality that was comparable, if not superior, to the SAPS-2.

CONCLUSIONS

The AI model, which integrates heart rate and blood pressure curve analyses with basic clinical parameters, provides a methodological approach to predict in-hospital mortality in ICU patients. This model offers an alternative to existing tools that depend on extensive clinical data and laboratory inputs. Its potential integration into ICU monitoring systems may facilitate more streamlined mortality prediction processes.

Identifying the need for infection-related consultations in intensive care patients using machine learning models

Infection-related consultations on intensive care units (ICU) have a positive impact on quality of care and clinical outcome. However, timing of these consultations is essential and to date they are typically event-triggered and reactive. Here, we investigate a proactive approach to identify patients in need for infection-related consultations by machine learning models using routine electronic health records. Data was retrieved from a mixed ICU at a large academic tertiary care hospital including 9684 admissions. Infection-related consultations were predicted using logistic regression, random forest, gradient boosting machines, and long short-term memory neural networks (LSTM). Overall, 7.8% of admitted patients received an infection-related consultation. Time-sensitive modelling approaches performed better than static approaches. Using LSTM resulted in the prediction of infection-related consultations in the next clinical shift (up to eight hours in advance) with an area under the receiver operating curve (AUROC) of 0.921 and an area under the precision recall curve (AUPRC) of 0.541. The successful prediction of infection-related consultations for ICU patients was done without the use of classical triggers, such as (interim) microbiology reports. Predicting this key event can potentially streamline ICU and consultant workflows and improve care as well as outcome for critically ill patients with (suspected) infections.

MHC-I upregulation safeguards neoplastic T cells in the skin against NK cell-mediated eradication in mycosis fungoides

Cancer-associated immune dysfunction is a major challenge for effective therapies. The emergence of antibodies targeting tumor cell-surface antigens led to advancements in the treatment of hematopoietic malignancies, particularly blood cancers. Yet their impact is constrained against tumors of hematopoietic origin manifesting in the skin. In this study, we employ a clonality-supervised deep learning methodology to dissect key pathological features implicated in mycosis fungoides, the most common cutaneous T-cell lymphoma. Our investigations unveil the prominence of the IL-32β-major histocompatibility complex (MHC)-I axis as a critical determinant in tumor T-cell immune evasion within the skin microenvironment. In patients' skin, we find MHC-I to detrimentally impact the functionality of natural killer (NK) cells, diminishing antibody-dependent cellular cytotoxicity and promoting resistance of tumor skin T-cells to cell-surface targeting therapies. Through murine experiments in female mice, we demonstrate that disruption of the MHC-I interaction with NK cell inhibitory Ly49 receptors restores NK cell anti-tumor activity and targeted T-cell lymphoma elimination in vivo. These findings underscore the significance of attenuating the MHC-I-dependent immunosuppressive networks within skin tumors. Overall, our study introduces a strategy to reinvigorate NK cell-mediated anti-tumor responses to overcome treatment resistance to existing cell-surface targeted therapies for skin lymphoma.

Chinese experts' consensus on the application of intensive care big data

The development of intensive care medicine is inseparable from the diversified monitoring data. Intensive care medicine has been closely integrated with data since its birth. Critical care research requires an integrative approach that embraces the complexity of critical illness and the computational technology and algorithms that can make it possible. Considering the need of standardization of application of big data in intensive care, Intensive Care Medicine Branch of China Health Information and Health Care Big Data Society, Standard Committee has convened expert group, secretary group and the external audit expert group to formulate Chinese Experts' Consensus on the Application of Intensive Care Big Data (2022). This consensus makes 29 recommendations on the following five parts: Concept of intensive care big data, Important scientific issues, Standards and principles of database, Methodology in solving big data problems, Clinical application and safety consideration of intensive care big data. The consensus group believes this consensus is the starting step of application big data in the field of intensive care. More explorations and big data based retrospective research should be carried out in order to enhance safety and reliability of big data based models of critical care field.

A locally optimised machine learning approach to early prognostication of long-term neurological outcomes after out-of-hospital cardiac arrest

Background

Out-of-hospital cardiac arrest (OHCA) represents a major burden for society and health care, with an average incidence in adults of 67 to 170 cases per 100,000 person-years in Europe and in-hospital survival rates of less than 10%. Patients and practitioners would benefit from a prognostication tool for long-term good neurological outcomes.

Objective

We aim to develop a machine learning (ML) pipeline on a local database to classify patients according to their neurological outcomes and identify prognostic features.

Methods

We collected clinical and biological data consecutively from 595 patients who presented OHCA and were routed to a single regional cardiac arrest centre in the south of France. We applied recursive feature elimination and ML analyses to identify the main features associated with a good neurological outcome, defined as a Cerebral Performance Category score less than or equal to 2 at six months post-OHCA.

Results

We identified 12 variables 24 h after admission, capable of predicting a six-month good neurological outcome. The best model (extreme gradient boosting) achieved an AUC of 0.96 and an accuracy of 0.92 in the test cohort.

Conclusion

We demonstrated that it is possible to build accurate, locally optimised prediction and prognostication scores using datasets of limited size and breadth. We proposed and shared a generic machine-learning pipeline which allows external teams to replicate the approach locally.

Role of artificial intelligence in haemodynamic monitoring

This narrative review explores the evolving role of artificial intelligence (AI) in haemodynamic monitoring, emphasising its potential to revolutionise patient care. The historical reliance on invasive procedures for haemodynamic assessments is contrasted with the emerging non-invasive AI-driven approaches that address limitations and risks associated with traditional methods. Developing the hypotension prediction index and introducing CircEWSTM and CircEWS-lite TM showcase AI's effectiveness in predicting and managing circulatory failure. The crucial aspects include the balance between AI and healthcare professionals, ethical considerations, and the need for regulatory frameworks. The use of AI in haemodynamic monitoring will keep growing with ongoing research, better technology, and teamwork. As we navigate these advancements, it is crucial to balance AI's power and healthcare professionals' essential role. Clinicians must continue to use their clinical acumen to ensure that patient outliers or system problems do not compromise the treatment of the condition and patient safety.

Optimizing artificial intelligence in sepsis management: Opportunities in the present and looking closely to the future

Sepsis remains a major challenge internationally for healthcare systems. Its incidence is rising due to poor public awareness and delays in its recognition and subsequent management. In sepsis, mortality increases with every hour left untreated. Artificial intelligence (AI) is transforming worldwide healthcare delivery at present. This review has outlined how AI can augment strategies to address this global disease burden. AI and machine learning (ML) algorithms can analyze vast quantities of increasingly complex clinical datasets from electronic medical records to assist clinicians in diagnosing and treating sepsis earlier than traditional methods. Our review highlights how these models can predict the risk of sepsis and organ failure even before it occurs. This gives providers additional time to plan and execute treatment plans, thereby avoiding increasing complications associated with delayed diagnosis of sepsis. The potential for cost savings with AI implementation is also discussed, including improving workflow efficiencies, reducing administrative costs, and improving healthcare outcomes. Despite these advantages, clinicians have been slow to adopt AI into clinical practice. Some of the limitations posed by AI solutions include the lack of diverse data sets for model building so that they are widely applicable for routine clinical use. Furthermore, the subsequent algorithms are often based on complex mathematics leading to clinician hesitancy to embrace such technologies. Finally, we highlight the need for robust political and regulatory frameworks in this area to achieve the trust and approval of clinicians and patients to implement this transformational technology.

Improving cardiovascular risk prediction through machine learning modelling of irregularly repeated electronic health records

Aims

Existing electronic health records (EHRs) often consist of abundant but irregular longitudinal measurements of risk factors. In this study, we aim to leverage such data to improve the risk prediction of atherosclerotic cardiovascular disease (ASCVD) by applying machine learning (ML) algorithms, which can allow automatic screening of the population.

Methods and results

A total of 215 744 Chinese adults aged between 40 and 79 without a history of cardiovascular disease were included (6081 cases) from an EHR-based longitudinal cohort study. To allow interpretability of the model, the predictors of demographic characteristics, medication treatment, and repeatedly measured records of lipids, glycaemia, obesity, blood pressure, and renal function were used. The primary outcome was ASCVD, defined as non-fatal acute myocardial infarction, coronary heart disease death, or fatal and non-fatal stroke. The eXtreme Gradient boosting (XGBoost) algorithm and Least Absolute Shrinkage and Selection Operator (LASSO) regression models were derived to predict the 5-year ASCVD risk. In the validation set, compared with the refitted Chinese guideline-recommended Cox model (i.e. the China-PAR), the XGBoost model had a significantly higher C-statistic of 0.792, (the differences in the C-statistics: 0.011, 0.006-0.017, P < 0.001), with similar results reported for LASSO regression (the differences in the C-statistics: 0.008, 0.005-0.011, P < 0.001). The XGBoost model demonstrated the best calibration performance (men: Dx = 0.598, P = 0.75; women: Dx = 1.867, P = 0.08). Moreover, the risk distribution of the ML algorithms differed from that of the conventional model. The net reclassification improvement rates of XGBoost and LASSO over the Cox model were 3.9% (1.4-6.4%) and 2.8% (0.7-4.9%), respectively.

Conclusion

Machine learning algorithms with irregular, repeated real-world data could improve cardiovascular risk prediction. They demonstrated significantly better performance for reclassification to identify the high-risk population correctly.

Mortality prediction with adaptive feature importance recalibration for peritoneal dialysis patients

The study aims to develop AICare, an interpretable mortality prediction model, using electronic medical records (EMR) from follow-up visits for end-stage renal disease (ESRD) patients. AICare includes a multichannel feature extraction module and an adaptive feature importance recalibration module. It integrates dynamic records and static features to perform personalized health context representation learning. The dataset encompasses 13,091 visits and demographic data of 656 peritoneal dialysis (PD) patients spanning 12 years. An additional public dataset of 4,789 visits from 1,363 hemodialysis (HD) patients is also considered. AICare outperforms traditional deep learning models in mortality prediction while retaining interpretability. It uncovers mortality-feature relationships and variations in feature importance and provides reference values. An AI-doctor interaction system is developed for visualizing patients' health trajectories and risk indicators.

Machine learning for dynamic and early prediction of acute kidney injury after cardiac surgery

OBJECTIVE

Acute kidney injury after cardiac surgery increases morbidity and mortality. Diagnosis relies on oliguria or increased serum creatinine, which develop 48 to 72 hours after injury. We hypothesized machine learning incorporating preoperative, operative, and intensive care unit data could dynamically predict acute kidney injury before conventional identification.

METHODS

Cardiac surgery patients at a tertiary hospital (2008-2019) were identified using electronic medical records in the Medical Information Mart for Intensive Care IV database. Preoperative and intraoperative parameters included demographics, Charlson Comorbidity subcategories, and operative details. Intensive care unit data included hemodynamics, medications, fluid intake/output, and laboratory results. Kidney Disease: Improving Global Outcomes creatinine criteria were used for acute kidney injury diagnosis. An ensemble machine learning model was trained for hourly predictions of future acute kidney injury within 48 hours. Performance was evaluated by area under the receiver operating characteristic curve and balanced accuracy.

RESULTS

Within the cohort (n = 4267), there were approximately 7 million data points. Median baseline creatinine was 1.0 g/dL (interquartile range, 0.8-1.2), with 17% (735/4267) of patients having chronic kidney disease. Postoperative stage 1 acute kidney injury occurred in 50% (2129/4267), stage 2 occurred in 8% (324/4267), and stage 3 occurred in 4% (183/4267). For hourly prediction of any acute kidney injury over the next 48 hours, area under the receiver operating characteristic curve was 0.82, and balanced accuracy was 75%. For hourly prediction of stage 2 or greater acute kidney injury over the next 48 hours, area under the receiver operating characteristic curve was 0.95 and balanced accuracy was 86%. The model predicted acute kidney injury before clinical detection in 89% of cases.

CONCLUSIONS

Ensemble machine learning models using electronic medical records data can dynamically predict acute kidney injury risk after cardiac surgery. Continuous postoperative risk assessment could facilitate interventions to limit or prevent renal injury.

Advancing prognostic precision in pulmonary embolism: A clinical and laboratory-based artificial intelligence approach for enhanced early mortality risk stratification

BACKGROUND

Acute pulmonary embolism (PE) is a critical medical emergency that necessitates prompt identification and intervention. Accurate prognostication of early mortality is vital for recognizing patients at elevated risk for unfavourable outcomes and administering suitable therapy. Machine learning (ML) algorithms hold promise for enhancing the precision of early mortality prediction in PE patients.

OBJECTIVE

To devise an ML algorithm for early mortality prediction in PE patients by employing clinical and laboratory variables.

METHODS

This study utilized diverse oversampling techniques to improve the performance of various machine learning models including ANN, SVM, DT, RF, and AdaBoost for early mortality prediction. Appropriate oversampling methods were chosen for each model based on algorithm characteristics and dataset properties. Predictor variables included four lab tests, eight physiological time series indicators, and two general descriptors. Evaluation used metrics like accuracy, F1_score, precision, recall, Area Under the Curve (AUC) and Receiver Operating Characteristic (ROC) curves, providing a comprehensive view of models' predictive abilities.

RESULTS

The findings indicated that the RF model with random oversampling exhibited superior performance among the five models assessed, achieving elevated accuracy and precision alongside high recall for predicting the death class. The oversampling approaches effectively equalized the sample distribution among the classes and enhanced the models' performance.

CONCLUSIONS

The suggested ML technique can efficiently prognosticate mortality in patients afflicted with acute PE. The RF model with random oversampling can aid healthcare professionals in making well-informed decisions regarding the treatment of patients with acute PE. The study underscores the significance of oversampling methods in managing imbalanced data and emphasizes the potential of ML algorithms in refining early mortality prediction for PE patients.

Artificial Intelligence in the Early Prediction of Cardiogenic Shock in Acute Heart Failure or Myocardial Infarction Patients: A Systematic Review and Meta-Analysis

Cardiogenic shock (CS) may have a negative impact on mortality in patients with heart failure (HF) or acute myocardial infarction (AMI). Early prediction of CS can result in improved survival. Artificial intelligence (AI) through machine learning (ML) models have shown promise in predictive medicine. Here, we conduct a systematic review and meta-analysis to assess the effectiveness of these models in the early prediction of CS. A thorough search of the PubMed, Web of Science, Cochrane, and Scopus databases was conducted from the time of inception until November 2, 2023, to find relevant studies. Our outcomes were area under the curve (AUC), the sensitivity and specificity of the ML model, the accuracy of the ML model, and the predictor variables that had the most impact in predicting CS. Comprehensive Meta-Analysis (CMA) Version 3.0 was used to conduct the meta-analysis. Six studies were considered in our study. The pooled mean AUC was 0.808 (95% confidence interval: 0.727, 0.890). The AUC in the included studies ranged from 0.77 to 0.91. ML models performed well, with accuracy ranging from 0.88 to 0.93 and sensitivity and specificity of 58%-78% and 88%-93%, respectively. Age, blood pressure, heart rate, oxygen saturation, and blood glucose were the most significant variables required by ML models to acquire their outputs. In conclusion, AI has the potential for early prediction of CS, which may lead to a decrease in the high mortality rate associated with it. Future studies are needed to confirm the results.

Establishment of a model for predicting preterm birth based on the machine learning algorithm

BACKGROUND

The purpose of this study was to construct a preterm birth prediction model based on electronic health records and to provide a reference for preterm birth prediction in the future.

METHODS

This was a cross-sectional design. The risk factors for the outcomes of preterm birth were assessed by multifactor logistic regression analysis. In this study, a logical regression model, decision tree, Naive Bayes, support vector machine, and AdaBoost are used to construct the prediction model. Accuracy, recall, precision, F1 value, and receiver operating characteristic curve, were used to evaluate the prediction performance of the model, and the clinical application of the model was verified.

RESULTS

A total of 5411 participants were included and were used for model construction. AdaBoost model has the best prediction ability among the five models. The accuracy of the model for the prediction of "non-preterm birth" was the highest, reaching 100%, and that of "preterm birth" was 72.73%.

CONCLUSIONS

By constructing a preterm birth prediction model based on electronic health records, we believe that machine algorithms have great potential for preterm birth identification. However, more relevant studies are needed before its application in the clinic.

Machine learning vs. traditional regression analysis for fluid overload prediction in the ICU

Fluid overload, while common in the ICU and associated with serious sequelae, is hard to predict and may be influenced by ICU medication use. Machine learning (ML) approaches may offer advantages over traditional regression techniques to predict it. We compared the ability of traditional regression techniques and different ML-based modeling approaches to identify clinically meaningful fluid overload predictors. This was a retrospective, observational cohort study of adult patients admitted to an ICU ≥ 72 h between 10/1/2015 and 10/31/2020 with available fluid balance data. Models to predict fluid overload (a positive fluid balance ≥ 10% of the admission body weight) in the 48-72 h after ICU admission were created. Potential patient and medication fluid overload predictor variables (n = 28) were collected at either baseline or 24 h after ICU admission. The optimal traditional logistic regression model was created using backward selection. Supervised, classification-based ML models were trained and optimized, including a meta-modeling approach. Area under the receiver operating characteristic (AUROC), positive predictive value (PPV), and negative predictive value (NPV) were compared between the traditional and ML fluid prediction models. A total of 49 of the 391 (12.5%) patients developed fluid overload. Among the ML models, the XGBoost model had the highest performance (AUROC 0.78, PPV 0.27, NPV 0.94) for fluid overload prediction. The XGBoost model performed similarly to the final traditional logistic regression model (AUROC 0.70; PPV 0.20, NPV 0.94). Feature importance analysis revealed severity of illness scores and medication-related data were the most important predictors of fluid overload. In the context of our study, ML and traditional models appear to perform similarly to predict fluid overload in the ICU. Baseline severity of illness and ICU medication regimen complexity are important predictors of fluid overload.

Application of extended criteria donor grafts in liver transplantation for acute-on-chronic liver failure: A retrospective cohort study

BACKGROUND

There is no consensus on the usage of extended criteria donor (ECD) grafts in liver transplantation (LT) for acute-on-chronic liver failure (ACLF) patients.

AIM

To summarize the experience of using ECD livers in ACLF-LT.

METHODS

A retrospective cohort study was conducted, enrolling patients who underwent LT at the First Affiliated Hospital of Sun Yat-Sen University from January 2015 to November 2021. The patients were divided into ECD and non-ECD groups for analysis.

RESULTS

A total of 145 recipients were enrolled in this study, of which ECD and non-ECD recipients accounted for 53.8% and 46.2%, respectively. Donation after cardiac death (DCD) recipients accounted for the minority compared with donation after brain death (DBD) recipients (16.6% vs 83.4%). Neither overall survival nor graft survival significantly differed between ECD and non-ECD and DCD and DBD recipients. ECD grafts were associated with a significantly higher incidence of early allograft dysfunction (EAD) than non-ECD grafts (67.9% vs 41.8%, P = 0.002). Postoperative outcomes between DCD and DBD recipients were comparable (P > 0.05). ECD graft (P = 0.009), anhepatic phase (P = 0.034) and recipient gamma glutamyltransferase (P = 0.016) were independent risk factors for EAD. Recipient preoperative number of extrahepatic organ failures > 2 (P = 0.015) and intraoperative blood loss (P = 0.000) were independent predictors of poor post-LT survival.

CONCLUSION

Although related to a higher risk of EAD, ECD grafts can be safely used in ACLF-LT. The main factors affecting post-LT survival in ACLF patients are their own severe preoperative disease and intraoperative blood loss.

Machine learning applications and challenges in graft-versus-host disease: a scoping review

PURPOSE OF REVIEW

This review delves into the potential of artificial intelligence (AI), particularly machine learning (ML), in enhancing graft-versus-host disease (GVHD) risk assessment, diagnosis, and personalized treatment.

RECENT FINDINGS

Recent studies have demonstrated the superiority of ML algorithms over traditional multivariate statistical models in donor selection for allogeneic hematopoietic stem cell transplantation. ML has recently enabled dynamic risk assessment by modeling time-series data, an upgrade from the static, "snapshot" assessment of patients that conventional statistical models and older ML algorithms offer. Regarding diagnosis, a deep learning model, a subset of ML, can accurately identify skin segments affected with chronic GVHD with satisfactory results. ML methods such as Q-learning and deep reinforcement learning have been utilized to develop adaptive treatment strategies (ATS) for the personalized prevention and treatment of acute and chronic GVHD.

SUMMARY

To capitalize on these promising advancements, there is a need for large-scale, multicenter collaborations to develop generalizable ML models. Furthermore, addressing pertinent issues such as the implementation of stringent ethical guidelines is crucial before the widespread introduction of AI into GVHD care.

Early prediction of need for invasive mechanical ventilation in the neonatal intensive care unit using artificial intelligence and electronic health records: a clinical study

BACKGROUND

Respiratory support is crucial for newborns with underdeveloped lung. The clinical outcomes of patients depend on the clinician's ability to recognize the status underlying the presented symptoms and signs. With the increasing number of high-risk infants, artificial intelligence (AI) should be considered as a tool for personalized neonatal care. Continuous monitoring of vital signs is essential in cardiorespiratory care. In this study, we developed deep learning (DL) prediction models for rapid and accurate detection of mechanical ventilation requirements in neonates using electronic health records (EHR).

METHODS

We utilized data from the neonatal intensive care unit in a single center, collected between March 3, 2012, and March 4, 2022, including 1,394 patient records used for model development, consisting of 505 and 889 patients with and without invasive mechanical ventilation (IMV) support, respectively. The proposed model architecture includes feature embedding using feature-wise fully connected (FC) layers, followed by three bidirectional long short-term memory (LSTM) layers.

RESULTS

A mean gestational age (GA) was 36.61 ± 3.25 weeks, and the mean birth weight was 2,734.01 ± 784.98 g. The IMV group had lower GA, birth weight, and longer hospitalization duration than the non-IMV group (P < 0.05). Our proposed model, tested on a dataset from March 4, 2019, to March 4, 2022. The mean AUROC of our proposed model for IMV support prediction performance demonstrated 0.861 (95%CI, 0.853-0.869). It is superior to conventional approaches, such as newborn early warning score systems (NEWS), Random Forest, and eXtreme gradient boosting (XGBoost) with 0.611 (95%CI, 0.600-0.622), 0.837 (95%CI, 0.828-0.845), and 0.0.831 (95%CI, 0.821-0.845), respectively. The highest AUPRC value is shown in the proposed model at 0.327 (95%CI, 0.308-0.347). The proposed model performed more accurate predictions as gestational age decreased. Additionally, the model exhibited the lowest alarm rate while maintaining the same sensitivity level.

CONCLUSION

Deep learning approaches can help accurately standardize the prediction of invasive mechanical ventilation for neonatal patients and facilitate advanced neonatal care. The results of predictive, recall, and alarm performances of the proposed model outperformed the other models.

Multi-omics indicators of long-term survival benefits after immune checkpoint inhibitor therapy

Molecular indicators of long-term survival (LTS) in response to immune-checkpoint inhibitor (ICI) treatment have the potential to provide both mechanistic and therapeutic insights. In this study, we construct predictive models of LTS following ICI therapy based on data from 158 clinical trials involving 21,023 patients of 25 cancer types with available 1-year overall survival (OS) rates. We present evidence for the use of 1-year OS rate as a surrogate for LTS. Based on these and corresponding TCGA multi-omics data, total neoantigen, metabolism score, CD8+ T cell, and MHC_score were identified as predictive biomarkers. These were integrated into a Gaussian process regression model that estimates "long-term survival predictive score of immunotherapy" (iLSPS). We found that iLSPS outperformed the predictive capabilities of individual biomarkers and successfully predicted LTS of patient groups with melanoma and lung cancer. Our study explores the feasibility of modeling LTS based on multi-omics indicators and machine-learning methods.