Machine Learning Methods to Predict Acute Respiratory Failure and Acute Respiratory Distress Syndrome

A risk prediction model for efficient intubation in the emergency department: A 4-year single-center retrospective analysis

Objective

To analyze the risk factors associated with intubated critically ill patients in the emergency department (ED) and develop a prediction model by machine learning algorithms.

Methods

This study was conducted in an academic tertiary hospital in Hangzhou, China. Critically ill patients admitted to the ED were retrospectively analyzed from May 2018 to July 2022. The demographic characteristics, distribution of organ dysfunction, parameters for different organs' examination, and status of mechanical ventilation were recorded. These patients were assigned to the intubation and non-intubation groups according to ventilation support. We used the eXtreme Gradient Boosting (XGBoost) algorithm to develop the prediction model and compared it with other algorithms, such as logistic regression, artificial neural network, and random forest. SHapley Additive exPlanations was used to analyze the risk factors of intubated critically ill patients in the ED.

Results

Of 14,589 critically ill patients, 10,212 comprised the training group and 4377 comprised the test group; 2289 intubated patients were obtained from the electronic medical records. The mean age, mean scores of vital signs, parameters of different organs, and blood oxygen examination results differed significantly between the two groups (p < 0.05). The white blood cell count, international normalized ratio, respiratory rate, and pH are the top four risk factors for intubation in critically ill patients. Based on the risk factors in different predictive models, the XGBoost model showed the highest area under the receiver operating characteristic curve (0.84) for predicting ED intubation.

Conclusions

For critically ill patients in the ED, the proposed model can predict potential intubation based on the risk factors in the clinically predictive model.

Acute Respiratory Distress Syndrome: Definition, Diagnosis, and Routine Management

Acute respiratory distress syndrome (ARDS) is an acute inflammatory lung injury characterized by severe hypoxemic respiratory failure, bilateral opacities on chest imaging, and low lung compliance. ARDS is a heterogeneous syndrome that is the common end point of a wide variety of predisposing conditions, with complex pathophysiology and underlying mechanisms. Routine management of ARDS is centered on lung-protective ventilation strategies such as low tidal volume ventilation and targeting low airway pressures to avoid exacerbation of lung injury, as well as a conservative fluid management strategy.

HIRA: Heart Rate Interval based Rapid Alert score to characterize autonomic dysfunction among patients with sepsis-related acute respiratory failure (ARF)

Objective. To examine whether heart rate interval based rapid alert (HIRA) score derived from a combination model of heart rate variability (HRV) and modified early warning score (MEWS) is a surrogate for the detection of acute respiratory failure (ARF) in critically ill sepsis patients.Approach. Retrospective HRV analysis of sepsis patients admitted to Emory healthcare intensive care unit (ICU) was performed between sepsis-related ARF and sepsis controls without ARF. HRV measures such as time domain, frequency domain, and nonlinear measures were analyzed up to 24 h after patient admission, 1 h before the onset of ARF, and a random event time in the sepsis controls. Statistical significance was computed by the Wilcoxon Rank Sum test. Machine learning algorithms such as eXtreme Gradient Boosting and logistic regression were developed to validate the HIRA score model. The performance of HIRA and early warning score models were evaluated using the area under the receiver operating characteristic (AUROC).Main Results. A total of 89 (ICU) patients with sepsis were included in this retrospective cohort study, of whom 31 (34%) developed sepsis-related ARF and 58 (65%) were sepsis controls without ARF. Time-domain HRV for Electrocardiogram (ECG) Beat-to-Beat RR intervals strongly distinguished ARF patients from controls. HRV measures for nonlinear and frequency domains were significantly altered (p< 0.05) among ARF compared to controls. The HIRA score AUC: 0.93; 95% confidence interval (CI): 0.88-0.98) showed a higher predictive ability to detect ARF when compared to MEWS (AUC: 0.71; 95% CI: 0.50-0.90).Significance. HRV was significantly impaired across patients who developed ARF when compared to controls. The HIRA score uses non-invasively derived HRV and may be used to inform diagnostic and therapeutic decisions regarding the severity of sepsis and earlier identification of the need for mechanical ventilation.

Prediction of respiratory failure risk in patients with pneumonia in the ICU using ensemble learning models

The aim of this study was to develop early prediction models for respiratory failure risk in patients with severe pneumonia using four ensemble learning algorithms: LightGBM, XGBoost, CatBoost, and random forest, and to compare the predictive performance of each model. In this study, we used the eICU Collaborative Research Database (eICU-CRD) for sample extraction, built a respiratory failure risk prediction model for patients with severe pneumonia based on four ensemble learning algorithms, and developed compact models corresponding to the four complete models to improve clinical practicality. The average area under receiver operating curve (AUROC) of the models on the test sets after ten random divisions of the dataset and the average accuracy at the best threshold were used as the evaluation metrics of the model performance. Finally, feature importance and Shapley additive explanation values were introduced to improve the interpretability of the model. A total of 1676 patients with pneumonia were analyzed in this study, of whom 297 developed respiratory failure one hour after admission to the intensive care unit (ICU). Both complete and compact CatBoost models had the highest average AUROC (0.858 and 0.857, respectively). The average accuracies at the best threshold were 75.19% and 77.33%, respectively. According to the feature importance bars and summary plot of the predictor variables, activetx (indicates whether the patient received active treatment), standard deviation of prothrombin time-international normalized ratio, Glasgow Coma Scale verbal score, age, and minimum oxygen saturation and respiratory rate were important. Compared with other ensemble learning models, the complete and compact CatBoost models have significantly higher average area under the curve values on the 10 randomly divided test sets. Additionally, the standard deviation (SD) of the compact CatBoost model is relatively small (SD:0.050), indicating that the performance of the compact CatBoost model is stable among these four ensemble learning models. The machine learning predictive models built in this study will help in early prediction and intervention of respiratory failure risk in patients with pneumonia in the ICU.

Long short-term memory model identifies ARDS and in-hospital mortality in both non-COVID-19 and COVID-19 cohort

OBJECTIVE

To identify the risk of acute respiratory distress syndrome (ARDS) and in-hospital mortality using long short-term memory (LSTM) framework in a mechanically ventilated (MV) non-COVID-19 cohort and a COVID-19 cohort.

METHODS

We included MV ICU patients between 2017 and 2018 and reviewed patient records for ARDS and death. Using active learning, we enriched this cohort with MV patients from 2016 to 2019 (MV non-COVID-19, n=3905). We collected a second validation cohort of hospitalised patients with COVID-19 in 2020 (COVID+, n=5672). We trained an LSTM model using 132 structured features on the MV non-COVID-19 training cohort and validated on the MV non-COVID-19 validation and COVID-19 cohorts.

RESULTS

Applying LSTM (model score 0.9) on the MV non-COVID-19 validation cohort had a sensitivity of 86% and specificity of 57%. The model identified the risk of ARDS 10 hours before ARDS and 9.4 days before death. The sensitivity (70%) and specificity (84%) of the model on the COVID-19 cohort are lower than MV non-COVID-19 cohort. For the COVID-19 + cohort and MV COVID-19 + patients, the model identified the risk of in-hospital mortality 2.4 days and 1.54 days before death, respectively.

DISCUSSION

Our LSTM algorithm accurately and timely identified the risk of ARDS or death in MV non-COVID-19 and COVID+ patients. By alerting the risk of ARDS or death, we can improve the implementation of evidence-based ARDS management and facilitate goals-of-care discussions in high-risk patients.

CONCLUSION

Using the LSTM algorithm in hospitalised patients identifies the risk of ARDS or death.

Real-time prediction of organ failures in patients with acute pancreatitis using longitudinal irregular data

It is extremely important to identify patients with acute pancreatitis who are at high risk for developing persistent organ failures early in the course of the disease. Due to the irregularity of longitudinal data and the poor interpretability of complex models, many models used to identify acute pancreatitis patients with a high risk of organ failure tended to rely on simple statistical models and limited their application to the early stages of patient admission. With the success of recurrent neural networks in modeling longitudinal medical data and the development of interpretable algorithms, these problems can be well addressed. In this study, we developed a novel model named Multi-task and Time-aware Gated Recurrent Unit RNN (MT-GRU) to directly predict organ failure in patients with acute pancreatitis based on irregular medical EMR data. Our proposed end-to-end multi-task model achieved significantly better performance compared to two-stage models. In addition, our model not only provided an accurate early warning of organ failure for patients throughout their hospital stay, but also demonstrated individual and population-level important variables, allowing physicians to understand the scientific basis of the model for decision-making. By providing early warning of the risk of organ failure, our proposed model is expected to assist physicians in improving outcomes for patients with acute pancreatitis.

Computational Simulation of Virtual Patients Reduces Dataset Bias and Improves Machine Learning-Based Detection of ARDS from Noisy Heterogeneous ICU Datasets

Goal: Machine learning (ML) technologies that leverage large-scale patient data are promising tools predicting disease evolution in individual patients. However, the limited generalizability of ML models developed on single-center datasets, and their unproven performance in real-world settings, remain significant constraints to their widespread adoption in clinical practice. One approach to tackle this issue is to base learning on large multi-center datasets. However, such heterogeneous datasets can introduce further biases driven by data origin, as data structures and patient cohorts may differ between hospitals. Methods: In this paper, we demonstrate how mechanistic virtual patient (VP) modeling can be used to capture specific features of patients' states and dynamics, while reducing biases introduced by heterogeneous datasets. We show how VP modeling can be used for data augmentation through identification of individualized model parameters approximating disease states of patients with suspected acute respiratory distress syndrome (ARDS) from observational data of mixed origin. We compare the results of an unsupervised learning method (clustering) in two cases: where the learning is based on original patient data and on data derived in the matching procedure of the VP model to real patient data. Results: More robust cluster configurations were observed in clustering using the model-derived data. VP model-based clustering also reduced biases introduced by the inclusion of data from different hospitals and was able to discover an additional cluster with significant ARDS enrichment. Conclusions: Our results indicate that mechanistic VP modeling can be used to significantly reduce biases introduced by learning from heterogeneous datasets and to allow improved discovery of patient cohorts driven exclusively by medical conditions.

Using Gated Recurrent Unit Networks for the Prediction of Hemodynamic and Pulmonary Decompensation

This paper presents a new medical severity scoring system, used to assess the risk of hemodynamic and pulmonary decompensation for patients being treated in intensive care units. The score presented here includes drug circulatory support and ventilation mode data for the evaluation of the patient's biosignals and laboratory values. It is shown that Gated Recurrent Unit-based neural networks are able to predict the maximal severity class within a 24 hour prediction time-frame (hemodynamic: 0.85 AUROC / pulmonary: 0.9 AUROC), and can estimate the underlying decompensation score for prediction times of up to 24 hours with mean errors of 6.3% of the maximal possible pulmonary, and 9.6% of the hemodynamic score. These results are based on 60h observation period. Clinical Relevance- Hemodynamic and pulmonary decom-pensation are life threatening dynamic events that can lead to death of patients. Early detection of these incidents is essential in order to intervene therapeutically and to improve survival chances. In everyday intensive care physicians are confronted with a vast number of laboratory values and vital parameters. There is a risk that early stages of hemodynamic and pulmonary decompensation are misjudged. The implementation of robust warning systems could support physicians in detecting these critical events and initiate therapeutical intervention in time which would achieve significant reduction of patient mortality.

Multi-Task Learning with Recurrent Neural Networks for ARDS Prediction using only EHR Data: Model Development and Validation Study (Preprint)

Background

Acute respiratory distress syndrome (ARDS) is a condition that is often considered to have broad and subjective diagnostic criteria and is associated with significant mortality and morbidity. Early and accurate prediction of ARDS and related conditions such as hypoxemia and sepsis could allow timely administration of therapies, leading to improved patient outcomes.

Objective

The aim of this study is to perform an exploration of how multilabel classification in the clinical setting can take advantage of the underlying dependencies between ARDS and related conditions to improve early prediction of ARDS in patients.

Methods

The electronic health record data set included 40,703 patient encounters from 7 hospitals from April 20, 2018, to March 17, 2021. A recurrent neural network (RNN) was trained using data from 5 hospitals, and external validation was conducted on data from 2 hospitals. In addition to ARDS, 12 target labels for related conditions such as sepsis, hypoxemia, and COVID-19 were used to train the model to classify a total of 13 outputs. As a comparator, XGBoost models were developed for each of the 13 target labels. Model performance was assessed using the area under the receiver operating characteristic curve. Heat maps to visualize attention scores were generated to provide interpretability to the neural networks. Finally, cluster analysis was performed to identify potential phenotypic subgroups of patients with ARDS.

Results

The single RNN model trained to classify 13 outputs outperformed the individual XGBoost models for ARDS prediction, achieving an area under the receiver operating characteristic curve of 0.842 on the external test sets. Models trained on an increasing number of tasks resulted in improved performance. Earlier prediction of ARDS nearly doubled the rate of in-hospital survival. Cluster analysis revealed distinct ARDS subgroups, some of which had similar mortality rates but different clinical presentations.

Conclusions

The RNN model presented in this paper can be used as an early warning system to stratify patients who are at risk of developing one of the multiple risk outcomes, hence providing practitioners with the means to take early action.

Applying artificial neural network for early detection of sepsis with intentionally preserved highly missing real-world data for simulating clinical situation

Purpose

Some predictive systems using machine learning models have been developed to predict sepsis; however, they were mostly built with a low percent of missing values, which does not correspond with the actual clinical situation. In this study, we developed a machine learning model with a high rate of missing and erroneous data to enable prediction under missing, noisy, and erroneous inputs, as in the actual clinical situation.

Materials and methods

The proposed artificial neural network model was implemented using the MATLAB ANN toolbox, based on stochastic gradient descent. The dataset was collected over the past decade with approval from the appropriate institutional review boards, and the sepsis status was identified and labeled using Sepsis-3 clinical criteria. The imputation method was built by last observation carried forward and mean value, aimed to simulate clinical situation.

Results

The mean area under the receiver operating characteristic (ROC) curve (AUC) of classifying sepsis and nonsepsis patients was 0.82 and 0.786 at 0 h and 40 h prior to onset, respectively. The highest model performance was found for one-hourly data, demonstrating that our ANN model can perform adequately with limited hourly data provided.

Conclusions

Our model has the moderate ability to predict sepsis up to 40 h in advance under simulated clinical situation with real-world data.