SCCM Resource Library-Heterogeneity of Effect of PEEP Strategy by Subphenotypes Derived From Clinical Data in ARDS

Video Transcription

Hello, my name is Rachel Kast and I'm a data scientist at Endpoint Health. Today I will be presenting research on behalf of my collaborators entitled  Heterogeneity of Effective Heap Strategy by Subphenotype Derived from Clinical Data in ARDS.  I'd like to thank the Society of Critical Care Medicine and the program committee for allowing us to share this work today.  My only disclosure is that I'm an employee of Endpoint Health. Background.  Acute Respiratory Distress Syndrome or ARDS is defined according to the Berlin criteria as a PF ratio less than 300 with bilateral chest opacities within a week of clinical  insult or worsening respiratory symptoms. This definition is purposely broad and it's meant to incorporate a diverse range of underlying etiologies and pathologies.  Despite the diversity in the definition, the PF ratio by itself is most frequently used to stratify ARDS patients, whether for prognosis, severity, or clinical intervention.  There is a need for improved stratification methodologies that can group patients on similarities that extend beyond just the PF ratio. Recently, our team and others have  explored the feasibility of ARDS subphenotyping for clinical applications.  The early models that were developed were able to identify two distinct groups of patients with differences in severity, outcomes, and response to treatment. However, they relied  on the use of novel biomarkers, which limits their immediate clinical application.  Recently, our team published a new methodology for ARDS subphenotyping that only uses data  that's already routinely collected as standard of care. Specifically, we look at three vital  signs, mean arterial pressure, heart rate, and respiratory rate, two lab values, creatinine and  bilirubin, and four arterial blood gas measures, arterial pH, PaO2, bicarbonate, and FiO2.  We input those nine clinical elements into an unsupervised k-means clustering model trained on the ARDSnet FACT and EDEN trials to identify two groups of patients.  We then validated that same model on the ARDSnet SALES, ALVEOLI, and ARMA trials, as well as the ART trial out of Brazil. What we found was two consistent and distinct groups  of patients in each of the studies that we looked at. In every study, patients who were assigned  to subphenotype A were characterized by overall lower levels of inflammation and better outcomes,  including lower mortality. Patients assigned to subphenotype B were characterized by higher levels of inflammation and worse outcomes, including higher mortality. While these  subphenotypes are interesting, they have yet to show clinical utility for actual application.  What we hypothesize is that the ARDS subphenotypes not only can identify two distinct groups of patients, but also that those groups of patients will have differential  responses to therapies. The objective of this work was to evaluate that hypothesis by assessing the heterogeneity of treatment effect of high versus low PEEP strategies  according to ARDS subphenotype in adult patients with ARDS methodology. This study was a  retrospective reanalysis of the ALVEOLI and the ART trials, which both were randomized clinical  trials exploring high versus low PEEP strategies. Patients were eligible to enroll in the ALVEOLI  trial if they met the American-European consensus criteria for ARDS and had a PF ratio of 300 or  less. They were enrolled in the trial within 48 hours of symptom onset. High versus low PEEP  strategies were assessed according to the high versus low PEEP tables defined in the study  protocol. The ART trial only looked at patients with moderate to severe ARDS with a PF ratio of  200 or less, and patients were eligible to enroll in the trial within 72 hours of symptom onset.  The ART trial had a similar low PEEP strategy using the ARDS-Net PEEP tables. However, the high PEEP intervention for the ART trial was titrated PEEP with recruitment maneuver.  For both studies, we looked at the patients who had the nine clinical elements required for our  unsupervised clustering. Each of those patients was assigned to either subphenotype A or subphenotype  B using the algorithm that I just described. All of our subsequent analysis was published online  prior to initiation of our work. We assessed the heterogeneity of treatment effect of high versus  low PEEP using a Bayesian hierarchical logistic model with pre-planned and weekly informative  priors, which aimed to encompass all effect settings for the primary outcome of mortality  at 28 days. We also assessed secondary outcomes of total ventilator days and total ventilator  free days at 28 days using the median and the interquartile range. Results. A total of 1,559  patients were enrolled in the two trials. Of those, 557 patients were missing at least one  of the nine elements required for our k-means clustering methodology, so they were therefore  excluded from analysis. The remaining 1,002 patients were included here. At study enrollment,  you can see that patients who are in subphenotype B were sicker than patients in subphenotype A.  This was characterized by the higher rate of vasopressor use as well as the worst severity  scores. The alveoli trial looked at the APACHE-3 score and the ARK trial looked at the SAPS-3 score.  We also see similarly that patients in subphenotype B had worse outcomes than subphenotype A. They had higher mortality at 28 days and fewer ventilator free days.  These Kaplan-Meier curves show mortality over time based on subphenotype and PEEP strategy used.  What we see here is that patients in subphenotype A receiving high PEEP had higher mortality than  their low PEEP counterparts over the entire study period. However, in subphenotype B, we saw similar rates of mortality over time regardless of the PEEP intervention used.  Bayesian analysis allowed us to further quantify those findings. What we see here is that in the  pooled data set comprised of both alveoli and ARK, patients in subphenotype A had a 99.4% risk  of harm if they received high PEEP. Conversely, patients in subphenotype B had a 63.9% probability  of benefit if they received high PEEP. These results were also consistent when we looked at  each study individually. In the alveoli trial, patients in subphenotype A who received high PEEP  had a 94.3% probability of harm. In the ARK trial, patients in subphenotype A who received high PEEP  had a 97.7% probability of harm. Conclusions. In conclusion, this study used a recently developed unsupervised clustering methodology to identify two subphenotypes of ARDS patients.  Not only were the subphenotypes clinically different, but they also displayed a heterogeneity of treatment effect based on PEEP strategy. Patients in subphenotype A had an increased  risk of harm from high PEEP strategies, while patients in subphenotype B had a modest probability  of benefit from high PEEP strategies. What we think is important about this clustering methodology  is that it uses routinely collected clinical data collected at the time of study enrollment to identify the patient subphenotype. We think that if this was to be implemented  into clinical practice, it would not introduce a significant measurement burden to either the patient,  the clinician, or the healthcare system overall. While we do need to perform a prospective randomized clinical trial to validate these findings, we do think that these results are  promising. We plan to expand on what we've learned here using unsupervised clustering  with clinical data to identify other targeted therapies and interventions for precision treatment  of ARDS, as well as expand out what we're doing into other acute indications such as sepsis.  If you'd like to learn more about some of our other heterogeneity of treatment effect work  based on ARDS subphenotype, we will also be presenting these abstracts as part of SCCM 2022.  Thank you for taking the time to watch this talk. I look forward to answering any questions in the chat.

Video Summary

In this video, Rachel Kast, a data scientist at Endpoint Health, presents research on heterogeneity of effective heap strategy in patients with Acute Respiratory Distress Syndrome (ARDS). The study explores the use of subphenotyping to identify distinct groups of patients with different responses to therapies. The research utilizes clinical data routinely collected from patients, such as vital signs and lab values, to identify two subphenotypes. The results show that patients in subphenotype A had an increased risk of harm from high PEEP strategies, while patients in subphenotype B had a modest probability of benefit. These findings suggest the potential for targeted therapies in ARDS.

Asset Subtitle

Pulmonary, Research, 2022

Asset Caption

INTRODUCTION/HYPOTHESIS: Acute respiratory distress syndrome (ARDS) is a heterogeneous condition. Recently, two subphenotypes using widely available clinical data were identified. Subphenotype B exhibits clinical and laboratory signals consistent with higher inflammation and has higher mortality compared to subphenotype A. The presence of heterogeneity of treatment effect (HTE) between these subphenotypes has not been explored. The objective of this study was to assess the HTE of different positive end-expiratory pressure (PEEP) strategies on 28-day mortality according to these subphenotypes in adult patients with ARDS. METHODS: This is a retrospective evaluation of two prior ARDS trials (ALVEOLI and ART) that compared high vs. low PEEP strategies. Patients were classified in two subphenotypes using a K-means clustering algorithm on nine clinical elements collected at randomization (pH, PaO2, bicarbonate, bilirubin, creatinine, mean arterial pressure, heart rate, respiratory rate and FiO2). This model was constructed using data from the two large trials (EDEN and FACTT), and validated using data from four trials (ALVEOLI, ARMA, SAILS, and ART). Biological characteristics of the subphenotypes were validated by evaluating the levels of pro-inflammatory plasma biomarkers. The primary outcome was 28-day mortality. HTE of PEEP was assessed following a pre-planned Bayesian hierarchical logistic model and reported as odds ratio (OR), 95% credible interval (CrI) and probability of benefit (OR < 1.00) or harm (OR >1.00). Priors were weakly informative, and sensitivity analyses considering different priors were performed. RESULTS: Data from 1559 ARDS patients were analyzed. High PEEP resulted in higher risk for 28-day mortality compared to low PEEP in patients in subphenotype A (OR, 1.66 [95% CrI, 1.13 to 2.47]; probability of harm of 99.4%) but not in subphenotype B (OR, 0.94 [95% CrI, 0.65 to 1.34]; probability of benefit of 63.9%). These effects were not modified when patients were stratified by either PaO2/FiO2 ratio or driving pressure, when the trials were assessed separately or using different priors. CONCLUSIONS: Our study demonstrates HTE of PEEP strategies across two subphenotypes derived from common clinical variables. This work may allow predictive enrichment in future clinical trials.

Meta Tag

Content Type Presentation

Knowledge Area Pulmonary

Knowledge Area Research

Knowledge Area Procedures

Knowledge Level Advanced

Membership Level Select

Tag Acute Respiratory Distress Syndrome ARDS

Tag Positive End Expiratory Pressure PEEP

Tag Mechanical Ventilation

Tag Outcomes Research

Year 2022

Keywords

ARDS

subphenotyping

clinical data

targeted therapies

heterogeneity

	Society of Critical Care Medicine 500 Midway Drive Mount Prospect, IL 60056 USA Phone: +1 847 827-6888 Fax: +1 847 439-7226 Email: support@sccm.org	Contact Us About SCCM Newsroom Advertising & Sponsorship DONATE	MySCCM LearnICU Patients & Families Surviving Sepsis Campaign Critical Care Societies Collaborative	GET OUR NEWSLETTER
© Society of Critical Care Medicine. All rights reserved. \| Privacy Statement \| Terms & Conditions The Society of Critical Care Medicine, SCCM, and Critical Care Congress are registered trademarks of the Society of Critical Care Medicine.