Identifying Clinical Research Priorities in Adult Pulmonary and Critical Care. NHLBI Working Group Report

Abstract

Preventing, treating, and promoting recovery from critical illness due to pulmonary disease are foundational goals of the critical care community and the NHLBI. Decades of clinical research in acute respiratory distress syndrome, acute respiratory failure, pneumonia, and sepsis have yielded improvements in supportive care, which have translated into improved patient outcomes. Novel therapeutics have largely failed to translate from promising preclinical findings into improved patient outcomes in late-phase clinical trials. Recent advances in personalized medicine, “big data,” causal inference using observational data, novel clinical trial designs, preclinical disease modeling, and understanding of recovery from acute illness promise to transform the methods of pulmonary and critical care clinical research. To assess the current state of, research priorities for, and future directions in adult pulmonary and critical care research, the NHLBI assembled a multidisciplinary working group of investigators. This working group identified recommendations for future research, including 1) focusing on understanding the clinical, physiological, and biological underpinnings of heterogeneity in syndromes, diseases, and treatment response with the goal of developing targeted, personalized interventions; 2) optimizing preclinical models by incorporating comorbidities, cointerventions, and organ support; 3) developing and applying novel clinical trial designs; and 4) advancing mechanistic understanding of injury and recovery to develop and test interventions targeted at achieving long-term improvements in the lives of patients and families. Specific areas of research are highlighted as especially promising for making advances in pneumonia, acute hypoxemic respiratory failure, and acute respiratory distress syndrome.

For 50 years, caring for patients with acute pulmonary disease has been a defining element of critical care (1, 2). Identifying interventions to prevent, treat, and facilitate recovery from pneumonia, acute hypoxemic respiratory failure (AHRF), and acute respiratory distress syndrome (ARDS) has been a priority for the critical care community and the NHLBI (3). Preclinical research (4, 5) and clinical trials (6) have yielded major improvements in supportive care, which have translated into improved patient outcomes (7, 8). However, novel therapeutics have largely failed to improve outcomes when tested in late-phase clinical trials (9–13).

To assess the current state of, research priorities for, and future directions in adult pulmonary and critical care research, in September 2018, the NHLBI assembled a working group for clinical research in adult pulmonary and critical care. This multidisciplinary group included leaders in clinical, translational, and basic science research in pulmonary, critical care, and emergency medicine. Specific areas of content expertise (pneumonia, AHRF, ARDS, and sepsis) and methodological expertise (molecular biology, preclinical disease models, early-phase clinical trials, late-phase clinical trials, long-term outcomes, health services research, and biostatistics) were represented. The meeting was divided into two types of sessions: 1) sessions addressing principles applicable to the design of clinical research in adult pulmonary and critical care and 2) sessions addressing specific disease processes. Each session was followed by group discussion of the common themes and priorities.

The first section of this report describes the predominant themes that emerged as having the greatest potential for contributing to progress in clinical research in adult pulmonary and critical care (Table 1). The second section outlines research priorities for three specific conditions highlighted for discussion by the working group: pneumonia, AHRF, and ARDS.

Table 1.

Summary of Overarching Recommendations for Future Critical Care Research

Heterogeneity of patients, conditions, and treatment effect	For critical illness syndromes such as sepsis or ARDS, research should focus on understanding the clinical, physiological, and biological underpinnings of heterogeneity in syndromes, diseases, and treatment response to improve the development of targeted interventions and the design of clinical trials.
Preclinical research and early-phase clinical trials	Optimizing preclinical models by incorporating comorbidities, cointerventions, and organ failure and support and improving the use of biomarkers and surrogate outcomes in early-phase clinical trials can close the gap between mechanistic discovery and successful late-phase clinical trials.
Clinical study designs	Adaptive and platform designs may increase the efficiency of evaluating novel therapies for a disease or subphenotypes of a disease. Cluster-level trials and trials leveraging electronic health records may increase the efficiency of comparing the effectiveness of common therapies for broad groups of patients.
Long-term patient-important outcomes	Future research should incorporate measurement of long-term patient-important outcomes and potential surrogate outcomes into observational and interventional studies to advance mechanistic understanding of injury, recovery, and adaptation. Trials should test interventions targeting these mechanisms to improve the lives of patients and their caregivers.

Definition of abbreviation: ARDS = acute respiratory distress syndrome.

Overarching Recommendations for Future Critical Care Research

Heterogeneity of Patients, Conditions, and Treatment Effect

A fundamental challenge for critical care research today is to identify which critically ill patients are likely to experience improved clinical outcomes with a given therapy. Meeting this challenge will require understanding three types of risk: 1) a patient’s overall risk of an outcome, 2) the proportion of a patient’s overall risk of an outcome that is attributable to the disease of interest, and 3) the proportion of the patient’s overall risk of an outcome that is amenable to treatment with the therapy being evaluated (Figures 1 and E1 in the online supplement).

Figure 1.

Relationships between overall risk, disease-attributable risk, and treatment-responsive risk. The x-axis displays, for a theoretical population of critically ill patients with a disease of interest, each patient’s predicted risk of death, ranging from 0% (will almost certainly survive) to 100% (will almost certainly die). The y-axis displays the observed percentage of patients who die, ranging from 0% (no patients died) to 100% (all patients died). The light gray shading denotes the deaths that were caused by the disease of interest (disease-attributable risk). The medium gray shading denotes deaths that were caused by comorbidities or concurrent organ failures rather than the disease of interest. The dark gray denotes deaths prevented by treatment. (A) A disease-specific treatment will generally be able to prevent more deaths among patients who have a greater overall risk of death (prognostic enrichment) and whose risk of death is primarily due to the disease being treated rather than comorbidities or concurrent organ failures. (B) A specific therapy may have a greater effect on outcomes in a highly treatable disease compared with a poorly treatable disease. Which patients respond to a specific treatment depends on the mechanism of action and the biology of the disease. (C) The probability of responding to a specific treatment might be greatest among patients at high overall risk or low overall risk, or it might be completely independent of overall risk. Disease-specific treatments would generally be expected to prevent deaths resulting from the disease of interest (light gray), whereas some supportive therapies might be expected to prevent deaths resulting from the disease of interest (light gray) and deaths resulting from comorbidities or concurrent organ failures (medium gray).

Overall risk

Within a population of critically ill patients, individual patients may differ greatly in their risk of an outcome (14). Patients at higher risk of a specific outcome have the potential to experience greater absolute reductions in that outcome with treatment (when disease-attributable and treatment-responsive risk are constant) (Figure 1) (14, 15). Patients at low risk of a specific outcome may have little potential to benefit from the treatment but may still experience its adverse effects (14). Enrolling patients at high risk of an outcome (prognostic enrichment) in critical care trials can ensure a high absolute event rate in the control group, potentially decreasing sample size, increasing power, and improving benefit–risk balance.

Attributable risk

Critical illness is complex, and frequently only a portion of a patient’s overall risk of an outcome is due to the disease of interest (attributable risk), whereas much of the patient’s risk of an outcome is due to comorbidities or concurrent organ failures (16–19) (Figure 1). For example, the overall in-hospital mortality among patients with ARDS is 30–40% (7). As much as 80% of the mortality risk among patients with ARDS, however, may be attributable to the underlying disease or concurrent organ failures rather than to ARDS itself (20). Thus, preventing or treating ARDS may produce only a small decrease in overall mortality. Hence, moribund patients, who have a high overall risk of death, of which only a small portion is attributable to the disease of interest, are frequently considered unlikely to benefit from study treatments and thus are excluded from clinical trials of ARDS therapies. Historically, critical care trials have selected sample sizes using overall rates of an outcome (21). Using more conservative estimates focused on the risk attributable to the disease, rather than the patient’s overall risk, may allow more realistic sample size estimations (Figure 1A).

Treatment-responsive risk

The difference between an individual’s outcome with a treatment versus without the treatment is the “treatment-responsive risk” (Figure 1C). Patients more likely to respond to a particular treatment may be identified using demographic characteristics (22), pathophysiology (23), prior response to the treatment (24), or a disease characteristic related to the treatment’s mechanism of action (e.g., genomic or proteomic factor) (25). Selectively enrolling patients in critical care trials who are more likely to respond to the treatment (predictive enrichment) can produce greater absolute and relative risk reductions, decrease sample size, and improve the benefit–risk balance for patients (23, 26–28).

Critical Illness Syndromes, Phenotypes, and Endotypes

Identifying patients likely to respond to a disease-specific treatment requires understanding the biology of the disease, the likely mechanistic effect of the treatment, and the interaction between the two. Historically, many critical illnesses (e.g., sepsis and ARDS) have been considered syndromes, defined by observable signs and symptoms, but not necessarily specific to underlying biological processes. Ideally, such syndromic definitions would be replaced with increasingly precise disease phenotypes (29), subphenotypes, and endotypes (Table 2). Critical care may learn from the fields of oncology and asthma, which have used biological pathways to define disease, identify therapeutic targets, and design efficient clinical trials (25, 30–32). This will require critical care research to develop valid proxy measures of mechanism in the causal pathway between intervention and outcome.

Table 2.

Approaches to Grouping Patients for Research

Term	Definition	Example
Syndrome	A set of medical signs and symptoms that are correlated with each other but may or may not be directly related to a specific disease	Asthma
Phenotype	Observable characteristics (clinical, physiological, morphological, biological, or response to treatment) without known relation to a pathophysiological mechanism	Early-onset allergic asthma
Subphenotype	A subset of a phenotype that is characteristic of a subset of a population	Early-onset allergic asthma with sputum eosinophilia
Endotype	A subset of a disease or syndrome defined by a distinct functional or pathophysiological mechanism	Th2 cytokines and allergen-specific IgE mediate atopy and eosinophilic inflammation responsive to corticosteroids and antibodies to IL-13 and IL-5

Definition of abbreviation: Th2 = T-helper cell type 2.

Large observational and interventional datasets with standardized clinical phenotyping, physiological measures, and biological samples from both acute illness and postacute recovery should be used to propel understanding of phenotypes, subphenotypes, and their mechanisms. Partnerships between academia, industry, and the U.S. Food and Drug Administration will be required to develop and validate approaches to rapid phenotyping and theragnostics (strategies that combine a diagnostic test which identifies patients likely to respond and targeted therapy based on that test’s results) for use in clinical trials. Markers of subphenotypes may be “static,” such as baseline plasma cytokine concentration (33), or “dynamic,” such as a patient’s physiological response to a change in positive end-expiratory pressure at the beginning of an ARDS trial (34). Ultimately, new interventions in critical care may be developed and tested using biological criteria rather than clinical syndromes, including potentially testing treatments in subphenotypes of patients that cut across historically separate clinical syndromes.

Preclinical Research

Major advances that have improved clinical outcomes for patients with ARDS began with preclinical animal studies. Research in small- and large-animal models established that higher Vts and high airway pressures were injurious to the normal and injured lung, paving the way for clinical trials to identify a marked reduction in mortality with use of lower Vts (4–6). Similarly, the potential benefit of a conservative approach to fluid management in ARDS was based on animal models of acute lung injury that demonstrated less pulmonary edema with reduced pulmonary vascular hydrostatic pressure (35).

Studies involving preclinical models must be designed with a specific aim and must be understood within that specific framework. For some preclinical models, the aim will be purely to advance mechanistic understanding (e.g., examining lung epithelial tissue factor–knockout mice to understand determinants of alveolar–capillary barrier permeability). With many preclinical models, the aim is to examine a new therapeutic and establish a mechanism of benefit and a proof of principle that the new therapy might have potential value in the clinical setting. To accomplish this objective, there is substantial need for preclinical models that have more direct clinical relevance. For example, a recent international expert consensus initiative provided recommendations for more representative preclinical models of sepsis (many of which apply to other critical illnesses) (36). Overall, preclinical models need to

• 1.Examine model systems beyond rodents (e.g., large animals [37], ex vivo human lungs [38], and LPS inhalation in human volunteers [39]);
• 2.Include covariates and comorbidities in animal models (e.g., age, diabetes, and immunosuppression);
• 3.Induce sepsis in a manner similar to the onset of sepsis in humans (e.g., microorganisms common in human sepsis [40], sites other than the peritoneal cavity [41], and prior microbial exposure to establish immunological memory and trained immunity [42]);
• 4.Include organ dysfunction and organ support in animal models (e.g., mechanical ventilation for respiratory failure);
• 5.Administer cointerventions similarly to current clinical care for human patients with sepsis (e.g., administration of antimicrobials and crystalloid solutions after, rather than before, sepsis onset);
• 6.Use clinically relevant outcomes over an appropriate time scale.

Ideally, research using animal models should evaluate the full spectrum of critical illness from prevention and early treatment through established disease and the resolution and recovery phases. In addition to more clinically relevant preclinical models, preclinical research should leverage highly controlled and replicable circumstances to understand not just whether a therapy works but also the circumstances under which it does not work, as well as the effects of specific aspects of the disease on the model and the therapy. An American Thoracic Society consensus document provides detailed recommendations for how animal studies of acute lung injury can be designed to encompass key elements of ARDS (43).

Another important issue with animal studies in sepsis and ARDS is the widespread lack of transparent, quality reporting of study design and implementation, including reporting randomization, blinding, and statistical power (44, 45). Ideally, preclinical research in pulmonary and critical care should move to adopt the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines (46). Optimizing preclinical models by incorporating randomization and blinding may improve the eventual translation of promising novel therapeutics from preclinical research into human clinical trials.

Early-Phase Clinical Trials and Surrogate Outcomes

Early-phase clinical trials are highly controlled studies enrolling small numbers of carefully selected patients. The aim of early-phase clinical trials is to advance understanding of pathophysiology (47, 48); define the route, dose, and side effects of a novel therapy (49); or assess the effect of a novel therapy on surrogate outcomes (50). Identifying surrogate outcomes for use in early-phase clinical trials is an urgent unmet need for critical care research (51). A surrogate outcome is a laboratory-, survey-, or performance-based measurement or a physical sign used as a substitute for a clinical outcome. For a surrogate outcome to be valid, the effect of the therapy on the surrogate outcome must predict the effect of the therapy on the clinical outcome. Development of validated, short-term surrogate outcomes could smooth the transition from preclinical models to early-phase clinical trials, increase the efficiency of adaptive trials, and facilitate the incorporation of long-term, patient-important outcomes into a broader range of trials. For long-term outcomes specifically, biological (e.g., S100B), structural (e.g., cortical thickness on magnetic resonance imaging), or functional (e.g., neuronal activity on functional magnetic resonance imaging) biomarkers would ideally eventually predict cognitive outcomes (52–54), assess the short-term efficacy of interventions (55), and elucidate the mechanisms of injury and recovery in critical illness (54, 56).

Clinical Study Design

In addition to the challenges summarized above regarding heterogeneity of patients, conditions, and treatment effects, the therapeutic window for many critical care interventions is narrow and early. Patients are often vulnerable or cognitively impaired, which complicates participation in research (57–59). To address these challenges, critical care research should 1) develop and apply novel trial designs (e.g., adaptive trials and platform trials), 2) leverage electronic health records (EHRs) to conduct pragmatic trials and observational research, 3) advance the understanding of and approaches to informed consent in acute care research, and 4) apply advanced observational research methods to inform both the design of trials and clinical care.

Adaptive and platform trials

Adaptive and platform trial designs hold the potential to minimize exposure of critically ill adults to ineffective therapies while shepherding effective therapies from development to clinical practice quickly and cost-effectively. An adaptive trial is defined as a trial designed to allow prospectively planned modifications to trial features (e.g., number of study groups, sample size per group, enrollment criteria, and dose or frequency of the intervention) based on accumulating data from subjects in the trial (Figure E2). Adaptive designs may gain efficiency from adaptive enrichment of the study population, early discontinuation of worse-performing treatments, and seamless transition from early- to late-phase trials.

A platform trial is defined as a trial using a single master protocol and research infrastructure to simultaneously evaluate multiple interventions and/or disease subpopulations in multiple substudies (Figure E2). Platform trials gain efficiencies from shared control groups, adaptive borrowing of information from similar groups of patients, and shared infrastructure and governance. However, there may be challenges, including considerable pretrial planning (e.g., statistical modeling, patient selection, organization of study arms, and within-trial adaptations), funding (e.g., traditional funding mechanisms may not easily handle a trial with no fixed sample size; a trial that tests multiple experimental agents owned by multiple companies cannot easily be sponsored by a single company), study monitoring, and consent (e.g., communicating risks and benefits to patients for each separate agent being tested).

Trials that combine features of adaptive and platform designs (adaptive platform trials) are currently addressing choice of antibiotic, use of steroids, and optimal ventilator strategy in community-acquired pneumonia (NCT02735707) and evaluating interventions to improve outcomes of elective surgery (NCT03861767). Additional efficiency may be gained by integrating adaptive platform trials with the EHR in the form of randomized, embedded, multifactorial, adaptive platform (“REMAP”) trials (60).

Adaptive and platform trials in critical care face specific design challenges and potential limitations. Methods developed for oncology trials with long enrollment windows, interventions implemented and monitored over long durations, and biologically established short-term surrogate outcomes may not extrapolate straightforwardly to critical care trials. The complexity of critical illness and the significant potential for interactions between multiple interventions and cointerventions require careful consideration during the design and sample size estimation of adaptive and platform trials.

Pragmatic trials

Pragmatic trials aim to evaluate the effectiveness of a treatment under “real-world” conditions by incorporating broad eligibility criteria, administration of the intervention by treating clinicians, and simplified data collection for patient-important outcomes (61). Pragmatic trial designs are well suited to critical care comparative effectiveness questions for which the required sample size is large, generalizability is a primary focus, or the intervention is more reliably delivered by clinical personnel than by study personnel (e.g., a trial of tracheal intubation vs. laryngeal mask airway during out-of-hospital cardiac arrest when research personnel are not present and the intervention is urgent). In the unpredictable environments of the prehospital setting, emergency department, and ICU, executing traditional study procedures (e.g., screening, consenting, enrolling, randomizing, and delivering the intervention) during a narrow therapeutic window may be challenging, disruptive to clinical care, or impossible. For questions in which early intervention is paramount, specific pragmatic designs may improve representative patient enrollment and reliable delivery of interventions. For example, cluster-randomized, cluster-crossover, or stepped-wedge trials, in which providers or units are randomized, may facilitate examination of urgent or emergent critical care interventions that would be challenging to rigorously evaluate using patient-level randomization (62, 63). Pragmatic trials may provide the sample size and breadth of patients necessary to examine not only the average effect of an intervention among all patients to whom the intervention might be applied in clinical practice but also the estimated effects of the intervention for individual patients (64).

Leveraging EHRs

Trials with traditional, adaptive, platform, or pragmatic designs (and observational research) may all potentially gain efficiency by leveraging the EHR. Applications that interface directly or indirectly with the EHR may facilitate multiple steps of a traditional randomized trial, including the following:

• 1.Screening for eligible patients (e.g., an ARDS “sniffer” [65])
• 2.Prompting and documenting informed consent (e.g., point-of-care consent for patients and clinicians [66])
• 3.Generating study group assignment (e.g., EHR-embedded randomization [67])
• 4.Delivering the intervention (e.g., via order sets or an advisor [68])
• 5.Monitoring treatment compliance and adverse events (69)
• 6.Providing data on cointerventions and short- and long-term outcomes (70)

Challenges to be addressed include improving quality and standardization of data entry in the EHR, developing registries and reproducible approaches for automatically extracting validated EHR data, building systems for real-time monitoring of adverse events, embedding machine learning models for phenotyping patients using EHR data, addressing data security and data-sharing concerns in multicenter trials, and collaborating with third-party and EHR vendors for support of clinical research. A potential benefit of integrating clinical research into the delivery of critical care through the EHR is to facilitate implementation of new evidence into practice as a part of a learning healthcare system (71).

Informed consent

Critical care research faces unique challenges regarding informed consent. Critical care research occurs in complex environments with patients who frequently lack decision-making capacity and need interventions on a time scale of minutes or hours rather than days. Current federal regulations specify two approaches to informed consent applicable to critical care research: 1) enrollment with prospective informed consent from patients who have full decision-making capacity or from an available legally authorized representative or 2) enrollment with waiver of or exception from informed consent when obtaining informed consent is impracticable (e.g., because of patients’ illness) and obtaining informed consent from a legally authorized representative is impracticable (e.g., because of the urgency of the intervention). Alternative approaches to prospective informed consent could be optimized for patients who are acutely ill but awake (e.g., abbreviated consent with informed refusal), sedated or in severe pain (e.g., abbreviated consent, assent, or dissent), or clinically unstable but possibly aware (e.g., simple disclosure) (72). Federal regulators and stakeholders in critical care research should evaluate alternative approaches to informed consent in the specific context of prehospital, emergency medicine, and critical care research. Simply denying current and future critically ill patients the opportunity to benefit from new research because traditional models of informed consent do not work well is unacceptable. Instead, models of informed participation in research are needed that reflect the realities of the context in which critical care is practiced and recognize the obligation of healthcare professionals and institutions to continuously improve the quality of clinical care through research (73). Achieving these aims will be facilitated by involving patients, families, and community members in each stage of critical care research from design to dissemination (74).

Advanced methods for observational research

For clinically important questions in critical care that cannot be readily answered in a clinical trial, analyses of observational data provide an important source of causal inference. Observational data can quantify the incidence of the disease (75), estimate the frequency of outcomes (76), identify subphenotypes (29), provide risk-adjusted associations between interventions and outcomes (77), and provide a sustainable context into which interventional studies are integrated (78). Increasingly, modern analytic techniques (79) applied to well-curated datasets may address many sources of confounding and either mirror the results of randomized trials (80) or provide generalizable information about patients underrepresented in traditional explanatory trials. The advent of “big data,” from “-omics” to million-patient EHR registries, provides the opportunity for complex, dynamic, multilevel datasets to be gathered, curated, and integrated across ICUs, hospitals, and regions. Priorities for future critical care research using big data are summarized in Table 3.

Table 3.

Recommendations for Critical Care Research Using “Big Data”

1. Promote collaboration across health centers, disciplines, and federal institutes

2. Revise funding models for information technology proposals to keep pace with rapidity of technological advancement

3. Incentivize standardized data entry in the electronic health record during clinical care to facilitate research

4. Implement a common data model and data curation to minimize missing data and optimize the quality of real-world data derived from electronic health records, health plan data, and patient-reported outcomes

5. Build distributed data networks and promote a standard framework for securely sharing and exchanging patient information and ensuring application interoperability

6. Link electronic health record registries with biospecimen repositories

7. Explore unsupervised approaches to phenotype derivation

8. Employ infrastructure developed for observational research to facilitate the conduct of pragmatic trials; phase IV postmarketing studies for drug safety and effectiveness; and data collection during pandemics, emergencies, and natural disasters

Long-Term, Patient-Important Outcomes

Core outcomes and measures

Over the past decade, substantial progress has been made in understanding which outcomes after critical illness are important to patients and how these outcomes should be measured (81). A program of research (www.improvelto.com)—including systematic reviews (82); international surveys of patients, caregivers, clinicians, and researchers (83); and an international modified Delphi consensus process (84, 85)—produced a recommended core outcome set and core outcome measurement set for all studies evaluating outcomes after hospital discharge for survivors of acute respiratory failure. Statistical methods have been developed for analyzing such functional outcomes while addressing truncation due to mortality (86). To facilitate interpretation of results across trials, future research evaluating patient-important outcomes after hospital discharge should include the established core outcome set in addition to study-specific outcomes. The core outcome set consists of survival, health-related quality of life, mental health, pain, cognition, physical function, muscle and/or nerve function, and pulmonary function (84). The established core outcome measurement set (85) should be used, and measurement instruments for core outcomes currently lacking consensus should be evaluated (87).

Long-term outcomes in randomized trials

Long-term, patient-important outcomes within a randomized trial may serve as the primary outcome (capturing a causal process affected by the intervention with a measurable effect size) (88, 89), secondary outcomes (important explanatory variables for which the study may have insufficient power) (90), safety outcomes (measures of harm through the primary causal mechanism or other pathways), or descriptive outcomes (characterizing the trajectory of the cohort over time, regardless of differences between groups). Specific, long-term, patient-important outcomes should be considered as the primary outcome for future critical care trials in which the mechanistic effect of the intervention is anticipated to affect such outcomes and as safety outcomes in many critical care trials in which the effect of the intervention on such outcomes is uncertain.

Knowledge gaps about long-term outcomes

Several knowledge gaps currently limit the broader use of long-term, patient-important outcomes in critical care research. First, limited progress has been made in understanding and measuring patients’ preillness baseline function and trajectory (which is a prerequisite to understanding how much disability is induced by critical illness and might be preventable with intervention) (91). A standardized approach to retrospectively ascertaining preillness conditions and trajectories could be developed, validated, and applied to observational and interventional critical care research.

Second, understanding the mechanism and timing of injury and recovery is a prerequisite to designing trials with the right intervention, dose, timing, frequency, intensity, and duration to improve long-term, patient-important outcomes. Large-cohort studies should advance mechanistic understanding of injury and recovery for patients and families by collecting clinical and biological samples before, during, and after acute illness. Contemporaneously, we should endeavor to understand adaptation to injuries that cannot be prevented or healed but whose impact on health and function can be mitigated.

Third, future research should evaluate the effects of critical illness not only on patients but also on families, caregivers, and healthcare professionals. Interventions should be evaluated that target a family-centered model of care, including enhanced social and community support, mental health interventions, or respite care. Engaging patients and families in critical care research may provide new insights into questions, methods, and outcomes and ensure accountability and a patient- and family-centered research focus (74). Improving long-term outcomes for critically ill patients may also require evaluating interventions to prevent and treat “burnout” among healthcare professionals, either organizationally (e.g., shift scheduling) or personally (e.g., mindfulness training) (92).

Research in Specific Diseases

Pneumonia

Pneumonia is the most common cause of both sepsis and ARDS and is a leading cause of death in the United States (6, 93, 94). Decreasing the incidence, severity, and sequelae of pneumonia is a priority for research in adult pulmonary and critical care. Improving understanding of and treatment for pneumonia faces several challenges. First, pneumonia is falsely considered to be merely an acute illness. Patients with pneumonia experience residual effects on cytokine concentrations past hospital discharge (95, 96) and clinical symptoms for months (97). Research should delineate host factors that determine susceptibility to pneumonia and factors that influence resolution of lung injury and long-term sequelae (42, 98–100). Second, pneumonia is incorrectly perceived to be merely a localized illness. Research should examine factors that influence the transition from local infection in the lung (pneumonia) to inflammatory responses elsewhere in the lung (ARDS) or systemically (sepsis). Research should also examine the mechanisms linking pneumonia to cardiac, cerebrovascular, renal, and other nonpulmonary sequelae, independent of sepsis, hypotension, or metastatic infection (101). Third, pneumonia is frequently treated as a syndrome. Cohort studies collecting clinical and biological samples before, during, and after severe pneumonia (98) should identify biological pathways (endotypes) underlying clinical presentations of pneumonia and mediators of injury resolution (102, 103) to target interventions at mechanisms rather than clinical syndromes (104). Specific research priorities for community-acquired pneumonia are presented in Table 4.

Table 4.

Research Priorities for Community-acquired Pneumonia

	Summary of Research Priorities	Illustrative Examples
Before pneumonia (prevention)	• Evaluate systemic and pulmonary markers of host susceptibility to pneumonia	• Use imaging and biospecimens to define and differentiate mechanistic signals in “protected” individuals (e.g., young healthy humans and animal models with relevant exposures establishing lung protection) versus more “susceptible” individuals (e.g., humans at high risk for pneumonia and animal models with relevant risk factors); test if those signals calibrate pneumonia risk when measured in human populations or shift pneumonia risk when manipulated in animal models
	• Examine mechanisms that defend the lung against pneumonia, including the ability to eliminate microbes (immune resistance) and the ability to withstand injury from infection and inflammation (tissue resilience)
	• Develop and test interventions that reverse or slow the progression of increasing pneumonia susceptibility
	• Examine the effects of antecedent respiratory infections on innate immunity in the lung (e.g., alveolar macrophage phenotypes) and local immunological memory in the lung (e.g., resident memory T cells)
During pneumonia (treatment)	• Advance cross-disciplinary evaluation of host–pathogen interactions	• Identify candidate pharmacologic agents that phenocopy signaling pathways tied to protection or reverse signaling pathways tied to injury and test lead candidates in experimental systems for abilities to bolster immune killing of microbes or resilience of lung and extrapulmonary tissues
	• Collect pulmonary and nonpulmonary biospecimens in observational and interventional studies to identify biological pathways (endotypes) underlying clinical presentations of pneumonia to target interventions at mechanisms rather than clinical syndromes
	• Examine factors that influence the transition from localized infection (pneumonia) to inflammatory responses elsewhere in the lung (ARDS) or systemically (sepsis)
	• Develop and test interventions that 1) help the host to more efficiently kill microbes, 2) diminish the anatomical and physiological damage from infection and inflammation, and 3) interrupt pneumonia’s nonpulmonary sequelae (e.g., muscle wasting)
	• Leverage new technology for rapid, comprehensive pathogen identification to ensure that evaluation of pneumonia treatments occurs in the context of early and appropriate antibiotic therapy
	• Elucidate the transition from infection and inflammation to resolution and repair (e.g., the role of specialized proresolving mediators) and the implications for immunomodulatory therapy
After pneumonia (facilitating resolution and recovery)	• Examine the mechanisms linking pneumonia to long-term pulmonary sequalae (e.g., accelerated chronic respiratory disease) and nonpulmonary sequelae (e.g., cardiac and cerebrovascular events and impaired cognition)	• Identify discrete molecular, physiological, or clinical signals that discriminate patients with pneumonia who developed adverse long-term outcomes from others who did not; test those signals for prognostic value in human patients and causation in animal models
	• Incorporate noninfectious outcomes (e.g., cardiovascular events) into long-term outcomes in pneumonia research
	• Define trajectories of recovery after pneumonia, as well as the biological, clinical, and social factors that influence these trajectories

Definition of abbreviation: ARDS = acute respiratory distress syndrome.

AHRF

AHRF, characterized by tachypnea and diminished arterial partial pressure of oxygen, occurs in 25–50% of ICU patients (105). A fundamental challenge in AHRF research is that numerous distinct diseases (e.g., cardiogenic pulmonary edema, community-acquired pneumonia, and acute eosinophilic pneumonia) or syndromes (e.g., sepsis and ARDS) may present as AHRF. Research in AHRF should balance the imperative to identify effective, common therapies for AHRF across a broad range of patients with recognition of the reality that identifying targeted, mechanistic therapies may require dividing patients with AHRF by the underlying biological processes, trajectories, and phases of illness. The degree to which the underlying biology and response to therapy are shared between the syndromes of AHRF and ARDS remains unclear, including whether the recently proposed subphenotypes of ARDS (33) have AHRF counterparts. Large, observational cohort studies with detailed phenotyping, endotyping, and genotyping are needed to understand the relationships between pneumonia, AHRF, and ARDS. Examples of important unanswered questions in the clinical care of mechanically ventilated patients with AHRF are provided in Table 5.

Table 5.

Research Questions in Acute Hypoxemic Respiratory Failure with Invasive Mechanical Ventilation

	Examples of Research Questions
Extracorporeal support	• Does venovenous extracorporeal membrane oxygenation improve outcomes for patients with mild-to-moderate ARDS or for patients with AHRF without ARDS?
	• Should extracorporeal carbon dioxide removal be used to facilitate ultralow-Vt ventilation?
Spontaneous breathing	• Can extracorporeal carbon dioxide removal facilitate safe spontaneous breathing?
	• Should we encourage spontaneous breathing by early transition to pressure support ventilation?
	• Should we encourage spontaneous breathing with airway pressure release ventilation?
	• Among spontaneously breathing patients, should we actively limit Vt?
Personalized ventilation	• Should we measure patient effort and titrate ventilatory support to achieve normal levels?
	• Does limiting driving pressure or mechanical power improve outcomes beyond pressure- or volume-limited ventilation?
	• Does minimizing asynchrony improve patient outcomes?
Comparative effectiveness	• How do outcomes compare between higher positive end-expiratory pressure and prone positioning overall and for specific subpopulations of patients?

Definition of abbreviations: AHRF = acute hypoxemic respiratory failure; ARDS = acute respiratory distress syndrome.

ARDS

ARDS has been an archetypal illness for adult critical care (6, 106, 107). Decades of basic, translational, and clinical research have improved scientific understanding of ARDS (108). Clinical trials have identified strategies for supportive care (many related to minimizing the harms of mechanical ventilation) that reduce systemic inflammation (109) and improve patient outcomes (6, 110). No pharmacological therapy for ARDS has been shown to reduce short-term or long-term mortality. In this way, ARDS exemplifies the challenges and opportunities facing adult pulmonary and critical care research outlined in this summary.

Improving outcomes for patients at risk for ARDS, for those with established ARDS, and for those recovering from ARDS remains a priority for adult pulmonary and critical care research. Each of the overarching recommendations for future critical care research described above applies to future clinical research in ARDS (111), as summarized in Table 6. Meeting these challenges in ARDS research may help clear the path for clinical research in adult pulmonary and critical care more broadly and advance the shared goal of enhancing health, lengthening life, and reducing illness and disability for critically ill adults and their families.

Table 6.

Methodologic Recommendations for Acute Respiratory Distress Syndrome Research

	Summary of Methodologic Recommendations	Illustrative Examples
Precision medicine	• Dissect the clinical syndrome of ARDS into the underlying subphenotypes or endotypes and test interventions targeted at biological pathways	• A trial comparing the efficacy of multiple interventions based on biomarkers and clinical outcomes among critically ill adults with a specific ARDS phenotype
Preclinical models	• Develop preclinical models of ARDS that better represent human disease	• Blinded, randomized trial of interventions in large-animal models that recapitulate ARDS etiology and biology in humans and subsequent treatment
Biomarker development	• Develop reliable biomarkers of prognosis and treatment response to bridge from preclinical research to late-phase clinical trials	• Research using preclinical models and existing or prospective human studies to understand and validate the role of specific molecules in the development, progression, and resolution of ARDS
Novel research designs	• Employ novel trial designs and leverage the EHR to improve the efficiency of ARDS clinical trials	• Cluster-randomized trial using electronic health records to identify potential participants, monitor delivery of the intervention, and continuously transmit curated, deidentified data to a coordinating center
Long-term outcomes	• Develop and incorporate a core set of long-term patient-important outcomes into ARDS clinical trials to improve understanding of patient trajectory, acute illness, and recovery	• Trial comparing a novel intervention with current clinical care with regard to long-term patient-important outcomes

Definition of abbreviations: ARDS = acute respiratory distress syndrome; EHR = electronic health record.

Conclusions

This NHLBI working group made several recommendations for future clinical research in adult pulmonary and critical care. Increased focus on understanding the clinical, physiological, and biological underpinnings of heterogeneity in syndromes, diseases, and treatment response will improve the development of targeted interventions to be tested with novel and efficient clinical trials. Optimizing preclinical research by using designs with randomization and blinding and incorporating comorbidities, cointerventions, and organ support into animal models may improve the translation of promising novel therapeutics from preclinical research to late-phase clinical trials. Developing novel biomarkers associated with patient prognosis and treatment response may help ensure that clinical trials evaluate novel therapeutics for the right patients at the right time. Adaptive, platform, and pragmatic trial designs may efficiently identify efficacious new therapies and compare the effectiveness of existing ones. Incorporating long-term, patient-important outcomes into large-scale observational studies and clinical trials will improve understanding of patient trajectories and mechanisms of injury and recovery and allow evaluation of targeted interventions to improve the lives of patients, caregivers, and healthcare professionals. Pneumonia, AHRF, and ARDS offer specific promising research opportunities to improve mechanistic understanding of disease and test available therapies with the potential to improve patient outcomes.