A simple classification model for hospital mortality in patients with acute lung injury managed with lung protective ventilation

A systematic review of machine learning models for management, prediction and classification of ARDS

AIM

Acute respiratory distress syndrome or ARDS is an acute, severe form of respiratory failure characterised by poor oxygenation and bilateral pulmonary infiltrates. Advancements in signal processing and machine learning have led to promising solutions for classification, event detection and predictive models in the management of ARDS.

METHOD

In this review, we provide systematic description of different studies in the application of Machine Learning (ML) and artificial intelligence for management, prediction, and classification of ARDS. We searched the following databases: Google Scholar, PubMed, and EBSCO from 2009 to 2023. A total of 243 studies was screened, in which, 52 studies were included for review and analysis. We integrated knowledge of previous work providing the state of art and overview of explainable decision models in machine learning and have identified areas for future research.

RESULTS

Gradient boosting is the most common and successful method utilised in 12 (23.1%) of the studies. Due to limitation of data size available, neural network and its variation is used by only 8 (15.4%) studies. Whilst all studies used cross validating technique or separated database for validation, only 1 study validated the model with clinician input. Explainability methods were presented in 15 (28.8%) of studies with the most common method is feature importance which used 14 times.

CONCLUSION

For databases of 5000 or fewer samples, extreme gradient boosting has the highest probability of success. A large, multi-region, multi centre database is required to reduce bias and take advantage of neural network method. A framework for validating with and explaining ML model to clinicians involved in the management of ARDS would be very helpful for development and deployment of the ML model.

Specific upregulation of extracellular miR-6238 in particulate matter-induced acute lung injury and its immunomodulation

Acute lung injury (ALI) and acute respiratory distress syndrome (ARDS) are life-threatening diseases characterized by a severe inflammatory response and the destruction of alveolar epithelium and endothelium. ALI/ARDS is caused by pathogens and toxic environmental stimuli, such as particulate matter (PM). However, the general symptoms of ALI/ARDS are similar, and determining the cause of lung injury is often challenging. In this study, we investigated whether there is a critical miRNA that characterizes PM-induced ALI. We found that the expression of miR-6238 is specifically upregulated in lung tissue and lung-derived extracellular vesicles (EVs) in response to PM exposure. Notably, bacterial endotoxin (Lipopolysaccharide; LPS or peptidoglycan; PTG) does not induce the expression of miR-6238 in the lung. Instead, the expression of miR-155 is dramatically increased in LPS-induced ALI. We further demonstrated that human lung epithelial cells and macrophages predominantly produce miR-6238 and miR-155, respectively. Mechanistically, EV-miR-6238 is effectively internalized into alveolar macrophages (AMs) and regulates inflammatory responses in vivo. CXCL3 is a main target of miR-6238 in AMs and modulates neutrophil infiltration into the lung alveoli. Collectively, our findings suggest that miR-6238 is a novel regulator of pulmonary inflammation and a putative biomarker that distinguishes PM-induced ALI from endotoxin (LPS/PTG)-mediated ALI.

A narrative review of progress in the application of artificial intelligence in acute respiratory distress syndrome: subtypes and predictive models

Background and Objective

Acute respiratory distress syndrome (ARDS) occurs in different populations, and it is very challenging to manage heterogeneous patient groups. Artificial intelligence (AI) aids in interpreting complex data of patients with ARDS and can be used to detect adverse events as it can automatically capture complex relationships. This review aimed to explore the application and progress of AI in ARDS (e.g., subgroup classification of patients with ARDS via unsupervised clustering and supervised predictive models for early detection) and identify the current ARDS-related problems that can be solved using AI.

Methods

This comprehensive and narrative review was performed to obtain information about the application of AI in ARDS and summarize its subtypes and predictive models.

Key Content and Findings

The current applications of AI and machine learning in ARDS include ARDS subgroup classification, diagnosis, and survival prediction. In this review, the current problems that should be addressed by AI in ARDS were identified, and our findings may serve as a useful reference for its translational use in the ARDS field.

Conclusions

Owing to the discovery of hyper- and hypoinflammatory subtypes, individualized treatment of ARDS is possible, and diagnosis and survival prediction are essential in disease management and planning. However, prospective studies should clarify the reliability and generalizability of the results using AI and machine learning and performing bedside testing in larger populations to establish a more stable and time-resilient model. Therefore, a consensus on conducting and reporting machine learning studies in medicine should be urgently established.

The Impact of Chronic Medical Conditions on Mortality in Acute Respiratory Distress Syndrome

PURPOSE

To examine the impact of chronic comorbidities on mortality in Acute Respiratory Distress Syndrome (ARDS).

MATERIALS AND METHODS

Retrospective cohort study of adults with ARDS (ICD-10-CM code J80) from the National Inpatient Sample between January, 2016 and December, 2018. For the primary outcome of mortality, we conducted weighted logistic regression adjusting for factors identified on univariate analysis as potentially significant or differing between the two groups at baseline. We used negative binomial regression adjusting for the same comorbidities to identify risk factors for longer length of stay (LOS) among ARDS survivors.

RESULTS

After exclusions, 1046 records were analyzed (3355 ARDS survivors and 1875 non-survivors.) The comorbidities examined included hypertension, diabetes mellitus, obesity, hypothyroidism, alcohol and drug use, chronic kidney disease (CKD), cardiovascular disease, chronic liver disease, chronic pulmonary disease and malignancy. In multivariate analysis, we found that malignancy (OR 2.26, 95% CI 1.84-2.78, p < 0.001), cardiovascular disease (OR 1.54, 95% CI 1.23-1.92, p < 0.001), and CKD (OR 1.75, 95% CI 1.22-2.50, p  =  0.002) increased the risk of death. In interaction analyses, cardiovascular disease combined with either malignancy or CKD conferred higher odds of death compared to either risk factor alone.

CONCLUSIONS

The comorbidity of malignancy confers the most reliable risk of poor outcomes in ARDS with higher odds of hospital death and a simultaneous association with longer hospital LOS among survivors.

A trauma-related survival predictive model of acute respiratory distress syndrome

Purpose

The purpose of this study was to construct and validate a simple model for the prediction of survival in patients with trauma‐related ARDS.

Methods

This is a single‐center, retrospective cohort study using MIMIC‐III Clinical Database.

Results

842 patients were included in this study. 175 (20.8%) died in‐hospital, whereas 215 (25.5%) died within 90 days. The deceased group had higher Acute Physiology Score (APS III), Sequential Organ Failure Assessment (SOFA), and Simplified Acute Physiology Score II (SAPS II). In multivariate logistic regression model, independent risk factors for mortality in ARDS patients included age ([odds ratio] OR, 1.035; 95% confidence interval [CI], 1.020–1.049), body mass index (OR, 0.957; 95% CI, 0.926–0.989), red blood cell distribution width (OR, 1.283; 95% CI, 1.141–1.443), hematocrit (OR, 1.055; 95% CI, 1.017–1.095), lactate (OR, 1.226; 95% CI, 1.127–1.334), blood urea nitrogen (OR, 1.025; 95% CI, 1.007–1.044), acute kidney failure (OR, 1.875; 95% CI, 1.188–2.959), sepsis (OR, 1.917; 95% CI, 1.165–3.153), type of admission (emergency vs. elective [OR, 2.822; 95% CI, 1.647–4.837], and urgent vs. elective [OR, 5.156; 95% CI, 1.896–14.027]). The area under the curve (AUC) of the model was 0.826, which was superior than the SAPS II (0.776), APS III (0.718), and SOFA (0.692). In the cross‐validation model, the accuracy of the test set was 0.823, the precision was 0.643, and the AUC was 0.813.

Conclusions

We established a prediction model using data commonly used in the clinic, which has high accuracy and precision and is worthy of use in clinical practice.

Can Big Data and Machine Learning Improve Our Understanding of Acute Respiratory Distress Syndrome?

Acute respiratory distress syndrome (ARDS) accounts for 10% of all diagnoses in the Intensive Care Unit, and about 40% of the patients succumb to the disease. Clinical methods alone can result in the under-recognition of this heterogeneous syndrome. The purpose of this study is to evaluate the role that big data and machine learning (ML) have played in understanding the heterogeneity of the disease and the development of various prediction algorithms. Most of the work in the field of ML in ARDS has been in the development of prediction models that have comparable efficacies to that of traditional models. Prediction algorithms have been useful in identifying new variables that may be important to consider in the future, supplementing the unknown information with the help of available noninvasive parameters, as well as predicting mortality. Phenotype identification using an unsupervised ML algorithm has been pivotal in classifying the heterogeneous population into more homogenous classes. Big data generated from ventilators in the form of ventilator waveform analysis and images in the form of radiomics have also been leveraged for the identification of the syndrome and can be incorporated into a clinical decision support system. Although the results are promising, lack of generalizability, "black box" nature of algorithms and concerns about "alarm fatigue" should be addressed for more mainstream adoption of these models.

Evaluating the association between unmet healthcare needs and subsequent clinical outcomes: protocol for the Addressing Post-Intensive Care Syndrome-01 (APICS-01) multicentre cohort study

INTRODUCTION

As short-term mortality declines for critically ill patients, a growing number of survivors face long-term physical, cognitive and/or mental health impairments. After hospital discharge, many critical illness survivors require an in-depth plan to address their healthcare needs. Early after hospital discharge, numerous survivors experience inadequate care or a mismatch between their healthcare needs and what is provided. Many patients are readmitted to the hospital, have substantial healthcare resource use and experience long-lasting morbidity. The objective of this study is to investigate the gap in healthcare needs occurring immediately after hospital discharge and its association with hospital readmissions or death for survivors of acute respiratory failure (ARF).

METHODS AND ANALYSIS

In this multicentre prospective cohort study, we will enrol 200 survivors of ARF in the intensive care unit (ICU) who are discharged directly home from their acute care hospital stay. Unmet healthcare needs, the primary exposure of interest, will be evaluated as soon as possible within 1 to 4 weeks after hospital discharge, via a standardised telephone assessment. The primary outcome, death or hospital readmission, will be measured at 3 months after discharge. Secondary outcomes (eg, quality of life, cognitive impairment, depression, anxiety and post-traumatic stress disorder) will be measured as part of 3-month and 6-month telephone-based follow-up assessments. Descriptive statistics will be reported for the exposure and outcome variables along with a propensity score analysis, using inverse probability weighting for the primary exposure, to evaluate the relationship between the primary exposure and outcome.

ETHICS AND DISSEMINATION

The study received ethics approval from Vanderbilt University Medical Center Institutional Review Board (IRB) and the University of Utah IRB (for the Veterans Affairs site). These results will inform both clinical practice and future interventional trials in the field. We plan to disseminate the results in peer-reviewed journals, and via national and international conferences.

TRIAL REGISTRATION DETAILS

ClinicalTrials.gov (NCT03738774). Registered before enrollment of the first patient.

Acute Respiratory Distress Syndrome: Bench-to-Bedside Approaches to Improve Drug Development

Despite 50 years of extensive research, no definite drug is currently available to treat acute respiratory distress syndrome (ARDS), and the supportive therapies remain the mainstay of treatment. To improve drug development for ARDS, researchers need to deeply analyze the “omics” approaches, reevaluate the suitable therapeutic targets, resolve the problems of inadequate animal modeling, develop the strategies to reduce the heterogeneity, and reconsider new therapeutic and analytical approaches for better designs of clinical trials.

External validation of a biomarker and clinical prediction model for hospital mortality in acute respiratory distress syndrome

PURPOSE

Mortality prediction in ARDS is important for prognostication and risk stratification. However, no prediction models have been independently validated. A combination of two biomarkers with age and APACHE III was superior in predicting mortality in the NHLBI ARDSNet ALVEOLI trial. We validated this prediction tool in two clinical trials and an observational cohort.

METHODS

The validation cohorts included 849 patients from the NHLBI ARDSNet Fluid and Catheter Treatment Trial (FACTT), 144 patients from a clinical trial of sivelestat for ARDS (STRIVE), and 545 ARDS patients from the VALID observational cohort study. To evaluate the performance of the prediction model, the area under the receiver operating characteristic curve (AUC), model discrimination, and calibration were assessed, and recalibration methods were applied.

RESULTS

The biomarker/clinical prediction model performed well in all cohorts. Performance was better in the clinical trials with an AUC of 0.74 (95% CI 0.70-0.79) in FACTT, compared to 0.72 (95% CI 0.67-0.77) in VALID, a more heterogeneous observational cohort. The AUC was 0.73 (95% CI 0.70-0.76) when FACTT and VALID were combined.

CONCLUSION

We validated a mortality prediction model for ARDS that includes age, APACHE III, surfactant protein D, and interleukin-8 in a variety of clinical settings. Although the model performance as measured by AUC was lower than in the original model derivation cohort, the biomarker/clinical model still performed well and may be useful for risk assessment for clinical trial enrollment, an issue of increasing importance as ARDS mortality declines, and better methods are needed for selection of the most severely ill patients for inclusion.

Novel translational approaches to the search for precision therapies for acute respiratory distress syndrome

In the 50 years since acute respiratory distress syndrome (ARDS) was first described, substantial progress has been made in identifying the risk factors for and the pathogenic contributors to the syndrome and in characterising the protein expression patterns in plasma and bronchoalveolar lavage fluid from patients with ARDS. Despite this effort, however, pharmacological options for ARDS remain scarce. Frequently cited reasons for this absence of specific drug therapies include the heterogeneity of patients with ARDS, the potential for a differential response to drugs, and the possibility that the wrong targets have been studied. Advances in applied biomolecular technology and bioinformatics have enabled breakthroughs for other complex traits, such as cardiovascular disease or asthma, particularly when a precision medicine paradigm, wherein a biomarker or gene expression pattern indicates a patient's likelihood of responding to a treatment, has been pursued. In this Review, we consider the biological and analytical techniques that could facilitate a precision medicine approach for ARDS.

Correlation between transition percentage of minute volume (TMV%) and outcome of patients with acute respiratory failure

PURPOSE

We have previously shown in patients receiving adaptive support ventilation (ASV) that there existed a Transition %MinVol (TMV%) where the patient's work of breathing began to reduce. In this study, we tested the hypothesis that higher TMV% would be associated with poorer outcome in patients with acute respiratory failure.

MATERIALS AND METHODS

In this prospective observational study, we recruited patients with acute respiratory failure on ASV between December 2012 and September 2013 in a mixed ICU. The TMV% was determined by adjusting % MinVol until mandatory respiratory frequency was between 0 and 1breath/min. TMV% was measured on the first two days of mechanical ventilation.

RESULTS

A total of 337 patients (age: 70±16years) were recruited. In patients whose TMV% increased between Day 1 and Day 2, aOR for mortality was 7.0 (95%CI=2.7-18.3, p<0.001) compared to patients whose TMV% decreased. In patients whose TMV% was unchanged between Day 1 and Day2, aOR for mortality was 3.91 (95%CI=1.80-8.22, p<0.01).

CONCLUSIONS

An increase, or lack of decrease, of TMV% from Day 1 to Day 2 was associated with higher risk of in-hospital death.

Severe hypercapnia and outcome of mechanically ventilated patients with moderate or severe acute respiratory distress syndrome

PURPOSE

To analyze the relationship between hypercapnia developing within the first 48 h after the start of mechanical ventilation and outcome in patients with acute respiratory distress syndrome (ARDS).

PATIENTS AND METHODS

We performed a secondary analysis of three prospective non-interventional cohort studies focusing on ARDS patients from 927 intensive care units (ICUs) in 40 countries. These patients received mechanical ventilation for more than 12 h during 1-month periods in 1998, 2004, and 2010. We used multivariable logistic regression and a propensity score analysis to examine the association between hypercapnia and ICU mortality.

MAIN OUTCOMES

We included 1899 patients with ARDS in this study. The relationship between maximum PaCO₂ in the first 48 h and mortality suggests higher mortality at or above PaCO₂ of ≥50 mmHg. Patients with severe hypercapnia (PaCO₂ ≥50 mmHg) had higher complication rates, more organ failures, and worse outcomes. After adjusting for age, SAPS II score, respiratory rate, positive end-expiratory pressure, PaO₂/FiO₂ ratio, driving pressure, pressure/volume limitation strategy (PLS), corrected minute ventilation, and presence of acidosis, severe hypercapnia was associated with increased risk of ICU mortality [odds ratio (OR) 1.93, 95% confidence interval (CI) 1.32 to 2.81; p = 0.001]. In patients with severe hypercapnia matched for all other variables, ventilation with PLS was associated with higher ICU mortality (OR 1.58, CI 95% 1.04-2.41; p = 0.032).

CONCLUSIONS

Severe hypercapnia appears to be independently associated with higher ICU mortality in patients with ARDS.

TRIAL REGISTRATION

Clinicaltrials.gov identifier, NCT01093482.

Vasoactive intestinal peptide suppresses macrophage-mediated inflammation by downregulating interleukin-17A expression via PKA- and PKC-dependent pathways

Interleukin (IL)-17A is a pro-inflammatory cytokine that markedly enhances inflammatory responses in the lungs by recruiting neutrophils and interacting with other pro-inflammatory mediators. Reducing the expression of IL-17A could attenuate inflammation in the lungs. However, whether VIP exerts its anti-inflammatory effects by regulating the expression of IL-17A has remained unclear. Here, we show that there is a remarkable increase of IL-17A in bronchoalveolar lavage fluid (BALF) and lung tissue of mice with acute lung injury (ALI). Moreover, lipopolysaccharides (LPS) stimulated elevated expression of IL-17A, which was evident by the enhanced levels of mRNA and protein observed. Furthermore, we also found that VIP inhibited LPS-mediated IL-17A expression in a time- and dose-dependent manner in an in vitro model of ALI and that this process might be mediated via the phosphokinase A (PKA) and phosphokinase C (PKC) pathways. Taken together, our results demonstrated that VIP might be an effective protector during ALI by suppressing IL-17A expression.

Is there still a role for the lung injury score in the era of the Berlin definition ARDS?

Background

The Lung Injury Score (LIS) remains a commonly utilized measure of lung injury severity though the additive value of LIS to predict ARDS outcomes over the recent Berlin definition of ARDS, which incorporates severity, is not known.

Methods

We tested the association of LIS (in which scores range from 0 to 4, with higher scores indicating more severe lung injury) and its four components calculated on the day of ARDS diagnosis with ARDS morbidity and mortality in a large, multi-ICU cohort of patients with Berlin-defined ARDS. Receiver Operator Characteristic (ROC) curves were generated to compare the predictive validity of LIS for mortality to Berlin stages of severity (mild, moderate and severe).

Results

In 550 ARDS patients, a one-point increase in LIS was associated with 58% increased odds of in-hospital death (95% CI 14 to 219%, P = 0.006), a 7% reduction in ventilator-free days (95% CI 2 to 13%, P = 0.01), and, among patients surviving hospitalization, a 25% increase in days of mechanical ventilation (95% CI 9 to 43%, P = 0.001) and a 16% increase (95% CI 2 to 31%, P = 0.02) in the number of ICU days. However, the mean LIS was only 0.2 points higher (95% CI 0.1 to 0.3) among those who died compared to those who lived. Berlin stages of severity were highly correlated with LIS (Spearman’s rho 0.72, P < 0.0001) and were also significantly associated with ARDS mortality and similar morbidity measures. The predictive validity of LIS for mortality was similar to Berlin stages of severity with an area under the curve of 0.58 compared to 0.60, respectively (P-value 0.49).

Conclusions

In a large, multi-ICU cohort of patients with ARDS, both LIS and the Berlin definition severity stages were associated with increased in-hospital morbidity and mortality. However, predictive validity of both scores was marginal, and there was no additive value of LIS over Berlin. Although neither LIS nor the Berlin definition were designed to prognosticate outcomes, these findings suggest that the role of LIS in characterizing lung injury severity in the era of the Berlin definition ARDS may be limited.

One-year mortality and predictors of death among hospital survivors of acute respiratory distress syndrome

PURPOSE

Advances in supportive care and ventilator management for acute respiratory distress syndrome (ARDS) have resulted in declines in short-term mortality, but risks of death after survival to hospital discharge have not been well described. Our objective was to quantify the difference between short-term and long-term mortality in ARDS and to identify risk factors for death and causes of death at 1 year among hospital survivors.

METHODS

This multi-intensive care unit, prospective cohort included patients with ARDS enrolled between January 2006 and February 2010. We determined the clinical characteristics associated with in-hospital and 1-year mortality among hospital survivors and utilized death certificate data to identify causes of death.

RESULTS

Of 646 patients hospitalized with ARDS, mortality at 1 year was substantially higher (41 %, 95% CI 37-45%) than in-hospital mortality (24%, 95% CI 21-27%), P < 0.0001. Among 493 patients who survived to hospital discharge, the 110 (22%) who died in the subsequent year were older (P < 0.001) and more likely to have been discharged to a nursing home, other hospital, or hospice compared to patients alive at 1 year (P < 0.001). Important predictors of death among hospital survivors were comorbidities present at the time of ARDS, and not living at home prior to admission. ARDS-related measures of severity of illness did not emerge as independent predictors of mortality in hospital survivors.

CONCLUSIONS

Despite improvements in short-term ARDS outcomes, 1-year mortality is high, mostly because of the large burden of comorbidities, which are prevalent in patients with ARDS.

The acute respiratory distress syndrome

The acute respiratory distress syndrome (ARDS) is an important cause of acute respiratory failure that is often associated with multiple organ failure. Several clinical disorders can precipitate ARDS, including pneumonia, sepsis, aspiration of gastric contents, and major trauma. Physiologically, ARDS is characterized by increased permeability pulmonary edema, severe arterial hypoxemia, and impaired carbon dioxide excretion. Based on both experimental and clinical studies, progress has been made in understanding the mechanisms responsible for the pathogenesis and the resolution of lung injury, including the contribution of environmental and genetic factors. Improved survival has been achieved with the use of lung-protective ventilation. Future progress will depend on developing novel therapeutics that can facilitate and enhance lung repair.