A Data-Driven Approach to Predicting Septic Shock in the Intensive Care Unit

Machine Learning-Based Early Prediction of Sepsis Using Electronic Health Records: A Systematic Review

BACKGROUND

Sepsis, a life-threatening infection-induced inflammatory condition, has significant global health impacts. Timely detection is crucial for improving patient outcomes as sepsis can rapidly progress to severe forms. The application of machine learning (ML) and deep learning (DL) to predict sepsis using electronic health records (EHRs) has gained considerable attention for timely intervention.

METHODS

PubMed, IEEE Xplore, Google Scholar, and Scopus were searched for relevant studies. All studies that used ML/DL to detect or early-predict the onset of sepsis in the adult population using EHRs were considered. Data were extracted and analyzed from all studies that met the criteria and were also evaluated for their quality.

RESULTS

This systematic review examined 1942 articles, selecting 42 studies while adhering to strict criteria. The chosen studies were predominantly retrospective (n = 38) and spanned diverse geographic settings, with a focus on the United States. Different datasets, sepsis definitions, and prevalence rates were employed, necessitating data augmentation. Heterogeneous parameter utilization, diverse model distribution, and varying quality assessments were observed. Longitudinal data enabled early sepsis prediction, and quality criteria fulfillment varied, with inconsistent funding-article quality correlation.

CONCLUSIONS

This systematic review underscores the significance of ML/DL methods for sepsis detection and early prediction through EHR data.

Performance effectiveness of vital parameter combinations for early warning of sepsis-an exhaustive study using machine learning

Objective

To carry out exhaustive data-driven computations for the performance of noninvasive vital signs heart rate (HR), respiratory rate (RR), peripheral oxygen saturation (SpO2), and temperature (Temp), considered both independently and in all possible combinations, for early detection of sepsis.

Materials and methods

By extracting features interpretable by clinicians, we applied Gradient Boosted Decision Tree machine learning on a dataset of 2630 patients to build 240 models. Validation was performed on a geographically distinct dataset. Relative to onset, predictions were clocked as per 16 pairs of monitoring intervals and prediction times, and the outcomes were ranked.

Results

The combination of HR and Temp was found to be a minimal feature set yielding maximal predictability with area under receiver operating curve 0.94, sensitivity of 0.85, and specificity of 0.90. Whereas HR and RR each directly enhance prediction, the effects of SpO2 and Temp are significant only when combined with HR or RR. In benchmarking relative to standard methods Systemic Inflammatory Response Syndrome (SIRS), National Early Warning Score (NEWS), and quick-Sequential Organ Failure Assessment (qSOFA), Vital-SEP outperformed all 3 of them.

Conclusion

It can be concluded that using intensive care unit data even 2 vital signs are adequate to predict sepsis upto 6 h in advance with promising accuracy comparable to standard scoring methods and other sepsis predictive tools reported in literature. Vital-SEP can be used for fast-track prediction especially in limited resource hospital settings where laboratory based hematologic or biochemical assays may be unavailable, inaccurate, or entail clinically inordinate delays. A prospective study is essential to determine the clinical impact of the proposed sepsis prediction model and evaluate other outcomes such as mortality and duration of hospital stay.

Comparative Analyses of the Impact of Different Criteria for Sepsis Diagnosis on Outcome in Patients with Spontaneous Subarachnoid Hemorrhage

Data on sepsis in patients with a subarachnoid hemorrhage (SAH) are scarce. We assessed the impact of different sepsis criteria on the outcome in an SAH cohort. Adult patients admitted to our ICU with a spontaneous SAH between 11/2014 and 11/2018 were retrospectively included. In patients developing an infection, different criteria for sepsis diagnosis (Sepsis-1, Sepsis-3_original, Sepsis-3_modified accounting for SAH-specific therapy, alternative sepsis criteria compiled of consensus conferences) were applied and their impact on functional outcome using the modified Rankin Scale (mRS) on hospital discharge and in-hospital mortality was evaluated. Of 270 SAH patients, 129 (48%) developed an infection. Depending on the underlying criteria, the incidence of sepsis and septic shock ranged between 21–46% and 9–39%. In multivariate logistic regression, the Sepsis-1 criteria were not associated with the outcome. The Sepsis-3 criteria were not associated with the functional outcome, but in shock with mortality. Alternative sepsis criteria were associated with mortality for sepsis and in shock with mortality and the functional outcome. While Sepsis-1 criteria were irrelevant for the outcome in SAH patients, septic shock, according to the Sepsis-3 criteria, adversely impacted survival. This impact was higher for the modified Sepsis-3 criteria, accounting for SAH-specific treatment. Modified Sepsis-3 and alternative sepsis criteria diagnosed septic conditions of a higher relevance for outcomes in patients with an SAH.

Beyond technology: Can artificial intelligence support clinical decisions in the prediction of sepsis?

ABSTRACT Objective: To analyze the critical alarms predictors of clinical deterioration/sepsis for clinical decision making in patients admitted to a reference hospital complex. Methods: An observational retrospective cohort study. The Machine Learning (ML) tool, Robot Laura®, scores changes in vital parameters and lab tests, classifying them by severity. Inpatients and patients over 18 years of age were included. Results: A total of 122,703 alarms were extracted from the platform, classified as 2 to 9. The pre-selection of critical alarms (6 to 9) indicated 263 urgent alerts (0.2%), from which, after filtering exclusion criteria, 254 alerts were delimited for 61 inpatients. Patient mortality from sepsis was 75%, of which 52% was due to sepsis related to the new coronavirus. After the alarms were answered, 82% of the patients remained in the sectors. Conclusions: Far beyond technology, ML models can speed up assertive clinical decisions by nurses, optimizing time and specialized human resources.

Intelligent patient monitoring for proactive alerting of key personnel in intensive care: A single-center study

A code blue event is an emergency code to indicate when a patient goes into cardiac arrest and needs resuscitation. In this paper, we model the binary response of a intensive care unit (ICU) patients experiencing a code-blue event, starting with vital time-series data of patients in 12 ICU beds. Our study introduces day-of and day-ahead risk scoring models trained against ground truth information on per-patient-per-day code-blue events, starting with multi-variate vital-time-series-sequences of varying durations with a plurality of engineered features capturing temporal variations of these signals. Actionable events, including code-blue events, aggregated by patient by day were predicted on the day-of or day-ahead with an overall accuracy of over 80% in our best models. Such models have potential to improve healthcare delivery by providing just-in-time alerting, enabling proactive and preventative clinical interventions, through continuous patient monitoring.

Predicting in-hospital mortality in adult non-traumatic emergency department patients: a retrospective comparison of the Modified Early Warning Score (MEWS) and machine learning approach

Background

A feasible and accurate risk prediction systems for emergency department (ED) patients is urgently required. The Modified Early Warning Score (MEWS) is a wide-used tool to predict clinical outcomes in ED. Literatures showed that machine learning (ML) had better predictability in specific patient population than traditional scoring system. By analyzing a large multicenter dataset, we aim to develop a ML model to predict in-hospital morality of the adult non traumatic ED patients for different time stages, and comparing performance with other ML models and MEWS.

Methods

A retrospective observational cohort study was conducted in five Taiwan EDs including two tertiary medical centers and three regional hospitals. All consecutively adult (>17 years old) non-traumatic patients admit to ED during a 9-year period (January first, 2008 to December 31th, 2016) were included. Exclusion criteria including patients with (1) out-of-hospital cardiac arrest and (2) discharge against medical advice and transferred to other hospital (3) missing collect variables. The primary outcome was in-hospital mortality and were categorized into 6, 24, 72, 168 hours mortality. MEWS was calculated by systolic blood pressure, pulse rate, respiratory rate, body temperature, and level of consciousness. An ensemble supervised stacking ML model was developed and compared to sensitive and unsensitive Xgboost, Random Forest, and Adaboost. We conducted a performance test and examine both the area under the receiver operating characteristic (AUROC) and the area under the precision and recall curve (AUPRC) as the comparative measures.

Result

After excluding 182,001 visits (7.46%), study group was consisted of 24,37,326 ED visits. The dataset was split into 67% training data and 33% test data for ML model development. There was no statistically difference found in the characteristics between two groups. For the prediction of 6, 24, 72, 168 hours in-hospital mortality, the AUROC of MEW and ML mode was 0.897, 0.865, 0.841, 0.816 and 0.939, 0.928, 0.913, 0.902 respectively. The stacking ML model outperform other ML model as well. For the prediction of in-hospital mortality over 48-hours, AUPRC performance of MEWS drop below 0.1, while the AUPRC of ML mode was 0.317 in 6 hours and 0.2150 in 168 hours. For each time frame, ML model achieved statistically significant higher AUROC and AUPRC than MEWS (all P < 0.001). Both models showed decreasing prediction ability as time elapse, but there was a trend that the gap of AUROC values between two model increases gradually (P < 0.001). Three MEWS thresholds (score >3, >4, and >5) were determined as baselines for comparison, ML mode consistently showed improved or equally performance in sensitivity, PPV, NPV, but not in specific.

Conclusion

Stacking ML methods improve predicted in-hospital mortality than MEWS in adult non-traumatic ED patients, especially in the prediction of delayed mortality.

A predictive algorithm using clinical and laboratory parameters may assist in ruling out and in diagnosing MDS

We present a noninvasive Web-based app to help exclude or diagnose myelodysplastic syndrome (MDS), a bone marrow (BM) disorder with cytopenias and leukemic risk, diagnosed by BM examination. A sample of 502 MDS patients from the European MDS (EUMDS) registry (n > 2600) was combined with 502 controls (all BM proven). Gradient-boosted models (GBMs) were used to predict/exclude MDS using demographic, clinical, and laboratory variables. Area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were used to evaluate the models, and performance was validated using 100 times fivefold cross-validation. Model stability was assessed by repeating its fit using different randomly chosen groups of 502 EUMDS cases. AUC was 0.96 (95% confidence interval, 0.95-0.97). MDS is predicted/excluded accurately in 86% of patients with unexplained anemia. A GBM score (range, 0-1) of less than 0.68 (GBM < 0.68) resulted in a negative predictive value of 0.94, that is, MDS was excluded. GBM ≥ 0.82 provided a positive predictive value of 0.88, that is, MDS. The diagnosis of the remaining patients (0.68 ≤ GBM < 0.82) is indeterminate. The discriminating variables: age, sex, hemoglobin, white blood cells, platelets, mean corpuscular volume, neutrophils, monocytes, glucose, and creatinine. A Web-based app was developed; physicians could use it to exclude or predict MDS noninvasively in most patients without a BM examination. Future work will add peripheral blood cytogenetics/genetics, EUMDS-based prospective validation, and prognostication.

A Locally Optimized Data-Driven Tool to Predict Sepsis-Associated Vasopressor Use in the ICU

OBJECTIVES

To train a model to predict vasopressor use in ICU patients with sepsis and optimize external performance across hospital systems using domain adaptation, a transfer learning approach.

DESIGN

Observational cohort study.

SETTING

Two academic medical centers from January 2014 to June 2017.

PATIENTS

Data were analyzed from 14,512 patients (9,423 at the development site and 5,089 at the validation site) who were admitted to an ICU and met Center for Medicare and Medicaid Services definition of severe sepsis either before or during the ICU stay. Patients were excluded if they never developed sepsis, if the ICU length of stay was less than 8 hours or more than 20 days or if they developed shock up to the first 4 hours of ICU admission.

MEASUREMENTS AND MAIN RESULTS

Forty retrospectively collected features from the electronic medical records of adult ICU patients at the development site (four hospitals) were used as inputs for a neural network Weibull-Cox survival model to derive a prediction tool for future need of vasopressors. Domain adaptation updated parameters to optimize model performance in the validation site (two hospitals), a different healthcare system over 2,000 miles away. The cohorts at both sites were randomly split into training and testing sets (80% and 20%, respectively). When applied to the test set in the development site, the model predicted vasopressor use 4-24 hours in advance with an area under the receiver operator characteristic curve, specificity, and positive predictive value ranging from 0.80 to 0.81, 56.2% to 61.8%, and 5.6% to 12.1%, respectively. Domain adaptation improved performance of the model to predict vasopressor use within 4 hours at the validation site (area under the receiver operator characteristic curve 0.81 [CI, 0.80-0.81] from 0.77 [CI, 0.76-0.77], p < 0.01; specificity 59.7% [CI, 58.9-62.5%] from 49.9% [CI, 49.5-50.7%], p < 0.01; positive predictive value 8.9% [CI, 8.5-9.4%] from 7.3 [7.1-7.4%], p < 0.01).

CONCLUSIONS

Domain adaptation improved performance of a model predicting sepsis-associated vasopressor use during external validation.

Artificial Intelligence for Clinical Decision Support in Sepsis

Sepsis is one of the main causes of death in critically ill patients. Despite the continuous development of medical technology in recent years, its morbidity and mortality are still high. This is mainly related to the delay in starting treatment and non-adherence of clinical guidelines. Artificial intelligence (AI) is an evolving field in medicine, which has been used to develop a variety of innovative Clinical Decision Support Systems. It has shown great potential in predicting the clinical condition of patients and assisting in clinical decision-making. AI-derived algorithms can be applied to multiple stages of sepsis, such as early prediction, prognosis assessment, mortality prediction, and optimal management. This review describes the latest literature on AI for clinical decision support in sepsis, and outlines the application of AI in the prediction, diagnosis, subphenotyping, prognosis assessment, and clinical management of sepsis. In addition, we discussed the challenges of implementing and accepting this non-traditional methodology for clinical purposes.

Data Science Methods for Nursing-Relevant Patient Outcomes and Clinical Processes: The 2019 Literature Year in Review

Data science continues to be recognized and used within healthcare due to the increased availability of large data sets and advanced analytics. It can be challenging for nurse leaders to remain apprised of this rapidly changing landscape. In this article, we describe our findings from a scoping literature review of papers published in 2019 that use data science to explore, explain, and/or predict 15 phenomena of interest to nurses. Fourteen of the 15 phenomena were associated with at least one paper published in 2019. We identified the use of many contemporary data science methods (eg, natural language processing, neural networks) for many of the outcomes. We found many studies exploring Readmissions and Pressure Injuries. The topics of Artificial Intelligence/Machine Learning Acceptance, Burnout, Patient Safety, and Unit Culture were poorly represented. We hope that the studies described in this article help readers: (1) understand the breadth and depth of data science's ability to improve clinical processes and patient outcomes that are relevant to nurses and (2) identify gaps in the literature that are in need of exploration.

A machine learning approach to predict healthcare-associated infections at intensive care unit admission: findings from the SPIN-UTI project

BACKGROUND

Identifying patients at higher risk of healthcare-associated infections (HAIs) in intensive care units (ICUs) represents a major challenge for public health. Machine learning could improve patient risk stratification and lead to targeted infection prevention and control interventions.

AIM

To evaluate the performance of the Simplified Acute Physiology Score (SAPS) II for HAI risk prediction in ICUs, using both traditional statistical and machine learning approaches.

METHODS

Data for 7827 patients from the 'Italian Nosocomial Infections Surveillance in Intensive Care Units' project were used in this study. The Support Vector Machines (SVM) algorithm was applied to classify patients according to sex, patient origin, non-surgical treatment for acute coronary disease, surgical intervention, SAPS II at admission, presence of invasive devices, trauma, impaired immunity, and antibiotic therapy in 48 h preceding ICU admission.

FINDINGS

The performance of SAPS II for predicting HAI risk provides a receiver operating characteristic curve with an area under the curve of 0.612 (P<0.001) and accuracy of 56%. Considering SAPS II along with other characteristics at ICU admission, the SVM classifier was found to have accuracy of 88% and an AUC of 0.90 (P<0.001) for the test set. The predictive ability was lower when considering the same SVM model but with the SAPS II variable removed (accuracy 78%, AUC 0.66).

CONCLUSIONS

This study suggested that the SVM model is a useful tool for early prediction of patients at higher risk of HAIs at ICU admission.

Early Detection of Sepsis With Machine Learning Techniques: A Brief Clinical Perspective

Sepsis is a major cause of death worldwide. Over the past years, prediction of clinically relevant events through machine learning models has gained particular attention. In the present perspective, we provide a brief, clinician-oriented vision on the following relevant aspects concerning the use of machine learning predictive models for the early detection of sepsis in the daily practice: (i) the controversy of sepsis definition and its influence on the development of prediction models; (ii) the choice and availability of input features; (iii) the measure of the model performance, the output, and their usefulness in the clinical practice. The increasing involvement of artificial intelligence and machine learning in health care cannot be disregarded, despite important pitfalls that should be always carefully taken into consideration. In the long run, a rigorous multidisciplinary approach to enrich our understanding in the application of machine learning techniques for the early recognition of sepsis may show potential to augment medical decision-making when facing this heterogeneous and complex syndrome.

Rethinking animal models of sepsis - working towards improved clinical translation whilst integrating the 3Rs

Sepsis is a major worldwide healthcare issue with unmet clinical need. Despite extensive animal research in this area, successful clinical translation has been largely unsuccessful.

We propose one reason for this is that, sometimes, the experimental question is misdirected or unrealistic expectations are being made of the animal model.

As sepsis models can lead to a rapid and substantial suffering – it is essential that we continually review experimental approaches and undertake a full harm:benefit impact assessment for each study. In some instances, this may require refinement of existing sepsis models. In other cases, it may be replacement to a different experimental system altogether, answering a mechanistic question whilst aligning with the principles of reduction, refinement and replacement (3Rs).

We discuss making better use of patient data to identify potentially useful therapeutic targets which can subsequently be validated in preclinical systems. This may be achieved through greater use of construct validity models, from which mechanistic conclusions are drawn. We argue that such models could provide equally useful scientific data as face validity models, but with an improved 3Rs impact. Indeed, construct validity models may not require sepsis to be modelled, per se. We propose that approaches that could support and refine clinical translation of research findings, whilst reducing the overall welfare burden on research animals.

Continuous and automatic mortality risk prediction using vital signs in the intensive care unit: a hybrid neural network approach

Mortality risk prediction can greatly improve the utilization of resources in intensive care units (ICUs). Existing schemes in ICUs today require laborious manual input of many complex parameters. In this work, we present a scheme that uses variations in vital signs over a 24-h period to make mortality risk assessments for 3-day, 7-day, and 14-day windows. We develop a hybrid neural network model that combines convolutional (CNN) layers with bidirectional long short-term memory (BiLSTM) to predict mortality from statistics describing the variation of heart rate, blood pressure, respiratory rate, blood oxygen levels, and temperature. Our scheme performs strongly compared to state-of-the-art schemes in the literature for mortality prediction, with our highest-performing model achieving an area under the receiver-operator curve of 0.884. We conclude that the use of a hybrid CNN-BiLSTM network is highly effective in determining mortality risk for the 3, 7, and 14 day windows from vital signs. As vital signs are routinely recorded, in many cases automatically, our scheme could be implemented such that highly accurate mortality risk could be predicted continuously and automatically, reducing the burden on healthcare providers and improving patient outcomes.

Neural Networks to Predict Radiographic Brain Injury in Pediatric Patients Treated with Extracorporeal Membrane Oxygenation

Brain injury is a significant source of morbidity and mortality for pediatric patients treated with Extracorporeal Membrane Oxygenation (ECMO). Our objective was to utilize neural networks to predict radiographic evidence of brain injury in pediatric ECMO-supported patients and identify specific variables that can be explored for future research. Data from 174 ECMO-supported patients were collected up to 24 h prior to, and for the duration of, the ECMO course. Thirty-five variables were collected, including physiological data, markers of end-organ perfusion, acid-base homeostasis, vasoactive infusions, markers of coagulation, and ECMO-machine factors. The primary outcome was the presence of radiologic evidence of moderate to severe brain injury as established by brain CT or MRI. This information was analyzed by a neural network, and results were compared to a logistic regression model as well as clinician judgement. The neural network model was able to predict brain injury with an Area Under the Curve (AUC) of 0.76, 73% sensitivity, and 80% specificity. Logistic regression had 62% sensitivity and 61% specificity. Clinician judgment had 39% sensitivity and 69% specificity. Sequential feature group masking demonstrated a relatively greater contribution of physiological data and minor contribution of coagulation factors to the model's performance. These findings lay the foundation for further areas of research directions.