Research Priorities for Coronavirus Disease 2019 in Critical Illness

The Surviving Sepsis Campaign

Craig M. Coopersmith, MD, MCCM; Massimo Antonelli, MD; Seth R. Bauer, PharmD, FCCM; Clifford S. Deutschman, MS, MD, MCCM; Laura E. Evans, MD, MSc, FCCM; Ricard Ferrer, MD, PhD; Judith Hellman, MD; Sameer Jog, MD; Jozef Kesecioglu, MD, PhD; Niranjan Kissoon, MB BS, MCCM; Ignacio Martin-Loeches, MD, PhD; Mark E. Nunnally, MD, FCCM; Hallie C. Prescott, MD, MSc; Andrew Rhodes, MB BS, MD(Res); Daniel Talmor, MD, MPH; Pierre Tissieres, MD, DSc; Daniel De Backer, MD, PhD


Crit Care Med. 2021;49(4):598-622. 

In This Article


Questions are presented in order of committee prioritization. When a tie occurred, the question that was rated as a priority by more committee members is listed first compared with a question that was rated as a higher priority by fewer committee members.

1) Should the Approach to Ventilator Management Differ From the Standard Approach in Patients With Acute Hypoxic Respiratory Failure?

What is Known. Most patients with COVID-19 admitted to the ICU present with acute respiratory insufficiency. COVID-19–related respiratory insufficiency cases generally fulfill the Berlin description of acute respiratory distress syndrome (ARDS).[13] The Surviving Sepsis Campaign guidelines for managing COVID-19 recommend managing COVID patients with ARDS similarly to classical ARDS patients.[6,7] Although patients with respiratory failure from COVID-19 fulfill most or all of the classical criteria of ARDS shortly after ICU admission, their lung mechanics have been observed to, at times, significantly differ from standard ARDS. In many patients, severe hypoxemia is associated with modestly altered lung compliance, a situation that is very uncommon in classical ARDS. The concept that there are different phenotypes in COVID-19–induced respiratory failure that require different ventilation strategies which may be distinct from that in classical ARDS has proven to be controversial. A study performed by Ferrando et al[14] on 742 patients concluded that COVID-19 ARDS patients have features similar to other causes of ARDS, with high compliance with lung-protective ventilation and risk of 28-day mortality increasing with degree of ARDS severity. However, they did not differentiate whether the patients had high or low lung compliance at inclusion. Further, Haudebourg et al[15] did not observe differences in respiratory mechanics between patients with COVID-19–induced ARDS and patients with classical ARDS from other causes of similar severity. Specifically, they described that lung compliance and driving pressure were identical although they also noted that airway closure phenomenon was more frequent in COVID-19 ARDS, potentially reflective of changes in airway inflammation. Additionally, recruitability was slightly higher in COVID-19 ARDS, and no relationship between time from symptom onset to lung compliance was identified. Ziehr et al[16] also reported that lung compliance was markedly altered in their patients with COVID-19 ARDS in a small study in which patients had a high response to prone positioning.

Gaps in Knowledge. There is not global agreement as to whether or not ventilator management should differ in COVID-19 patients from other patients with hypoxic respiratory failure. Gattinoni et al[17] proposed the "L and H" phenotypes in patients with respiratory failure from COVID-19 based on retrospective data, CT scans, case observation, and communication with colleagues. Type L was proposed to represent the beginning of COVID-19 pneumonia, characterized with low elastance (high compliance), low ventilation to perfusion, low lung weight, and low lung recruitability (attributed to mainly open lung). Furthermore, they proposed that hypoxemia occurs in compliant lungs due to a loss of lung perfusion regulation and hypoxic vasoconstriction. In contrast, Type H was proposed as a phenotype seen at later stages, characterized by high elastance, high right-to-left shunt, high lung weight, and high recruitability, similar to severe classical ARDS criteria. Based upon these proposed phenotypes, they suggested rapid intubation of type L patients with large respiratory drive to avoid excessive intrathoracic negative pressures and self-inflicted lung injury, transforming to Type H.[17] They also advised avoiding high positive end-expiratory pressure (PEEP) and recruitment attempts and maintain gentle ventilation to avoid further damage. However, some authors have stated that additional information is needed to understand respiratory failure from COVID-19 more comprehensively and that premature phenotyping might expose these patients to considerable and preventable risk.[18] Whether or not the H and L types exist (and if so how common they are), whether there are fundamental differences in respiratory failure in patients with COVID-19, and whether this should direct clinical management remain gaps in current knowledge.

Future Directions. A fundamental question is whether there are different phenotypes in patients with respiratory failure due to COVID-19 that require different management strategies. If so, research is needed to identify the determinants of which phenotype a patient has, if these occur in different patients or consecutively in the same patient and if the evolution from one phenotype to another can be prevented. A complementary question of profound importance is whether lung-protective ventilation with low tidal volume and high PEEP should be applied in all COVID-19 ARDS patients. Further, it has been demonstrated that large respiratory efforts are associated with relapse of respiratory failure in patients with COVID-19–induced ARDS during weaning from mechanical ventilation,[19] but it is unknown whether large respiratory efforts at the early stages also contribute to lung deterioration. Research is also needed to determine if standard measurements used in non-COVID patients with respiratory insufficiency (arterial blood gas, respiratory rate, clinical assessment) are the appropriate metrics to determine when to intubate COVID-19 patients and whether this is altered in patients with excessive inspiratory drive. Further, it remains to be determined how best to identify patients who should be intubated early even though gas exchange may be still be acceptable and conversely how to identify patients in whom intubation may potentially be avoided despite hypoxemia. It is also unclear if prone position should be used more liberally in ARDS associated with COVID-19.

2) Can the Host Response Be Modulated for Therapeutic Benefit?

What is Known. Patients with COVID-19 can have a dysregulated inflammatory and hemostatic response following infection with SARS-CoV-2. Although patients were initially described as having a "cytokine storm," subsequent larger studies demonstrated that cytokine levels are either equivalent to or lower than seen in other forms of critical illness.[20–23] Corticosteroids have been shown to improve 28-day survival in critically ill patients with COVID-19,[24–28] but it is unclear whether their benefit is due to manipulation of the cytokine response or other effects. The largest of multiple randomized controlled trials to support a benefit of corticosteroids in COVID-19 was the Randomized Evaluation of COVID-19 Therapy (RECOVERY) trial which found lower mortality in 2104 patients randomized to dexamethasone compared with usual care (age-adjusted rate ratio, 0.83; 95% CI, 0.75–0.93).[24] Notably, the mortality benefit was restricted to patients on mechanical ventilation (rate ratio, 0.64; 95% CI, 0.51–0.81) or receiving oxygen without invasive mechanical ventilation (rate ratio, 0.82; 95% CI, 0.72–0.94). A meta-analysis of seven randomized controlled trials of various corticosteroids including dexamethasone, hydrocortisone, and methylprednisolone in 1,703 critically ill patients (over 90% on mechanical ventilation) demonstrated improved short-term survival in patients randomized to receive corticosteroids (odds ratio, 0.66; 95% CI, 0.53–0.82).[28]

Gaps in Knowledge. The host response to SARS-CoV-2 is heterogeneous. Immune profiling suggests distinct "immunotypes" in patients with COVID-19, each with differing T cell and B cell activation characteristics and clinical trajectories.[29] Furthermore, a maladapted immunologic response may occur over time in severely ill patients.[30] The best way to target the heterogeneous host response is unknown, and where a patient is on their longitudinal inflammatory response may have profound implications for clinical approaches used at the bedside. A large number of host-modulating agents have been examined in randomized controlled trials, propensity matched studies, smaller observational trials, and case series. An incomplete list of agents designed to alter the host response include SARS-CoV-2 immunoglobulins (specific and nonspecific), mesenchymal stem cells, interferon alpha, interferon beta, interleukin-1 inhibitors, interleukin-6 inhibitors, Bruton's tyrosine kinase inhibitors, and janus kinase inhibitors. Data on each of these (as well as additional experimental agents) are inconclusive to determine whether they are beneficial toward improving survival—or other patient-centric outcome—in patients with COVID-19. There are many reasons behind these gaps in knowledge, but one example is the difficulty in performing well-done randomized trials in the time of a pandemic (see question 10 below). Several uncertainties remain regarding the use of corticosteroids. In particular, only short-term outcomes (21–30 d survival) were evaluated in RECOVERY and the meta-analysis.[28] Previous trials of corticosteroids in ARDS demonstrated a negative impact on long-term outcome (evaluated at 60 and 180 d) despite favorable short-term outcomes.[31] Also, identification of the patients who mostly benefit from steroids remains uncertain as the RECOVERY trial suggested that the subgroup of patients on mechanical ventilation benefitted most from steroids, whereas the meta-analysis reported the reverse although the number of nonvented patients in the meta-analysis was small and did not include patients from RECOVERY since this trial did not distinguish between critically ill and noncritically ill patients. Additionally, adverse events should be better reported, in particular, occurrence of secondary infections and neuromuscular weakness. Finally, it is unclear if there is efficacy in extracorporeal therapies such as plasmapheresis and cytokine removal with various membranes as has been suggested in nonrandomized cohorts.[3,23]

Future Directions. Fundamentally, modulating the host response requires an appropriate target and the appropriate patient population. A number of targets have been identified including specific cytokines, mediators that augment or suppress inflammation, endothelial activation, etc. Specific targets are outlined above and in other questions in this article; however, the committee explicitly chose not to prioritize which targets should be approached first, as a comprehensive assessment of which elements of the host response are best candidates for modification is beyond the scope of this question. Although conceptually targeting both viral replication and the host response has appeal, limited studies on combination therapy have been performed, and unexpected side effects need to be understood since modulation of the host response may potentially suppress the immune system, which could promote SARS-CoV-2 growth/replication or secondary bacterial infection. In choosing which patients to study, clinical studies to date have predominantly characterized ICU patients by presence/absence of mechanical ventilation and/or high-flow nasal cannula. More broadly, studies have used a seven-point ordinal scale in which ICU patients make up ordinal 7 (mechanical ventilation or ECMO), ordinal 6 (high-flow nasal cannula), and in rare circumstances ordinal 5 (supplemental oxygen alone). Although these studies have shown clear differences between patient populations responding to the same intervention based upon severity of illness, lumping all ICU patients into a single category almost certainly misses critical differences that could be exploited for therapeutic gain. Although the presence/absence of mechanical ventilation has inherent benefit in clinical trials since it is easily measurable, it is also does not account for individual genetic and epigenetic variability, or how patients with different subphenotypes are likely to respond differently to the identical agent. As such, biomarker or physiomarker enriched trials may lead to a more detailed understanding of which patients would benefit, be harmed, or have no effect from manipulating the host response. Further, specifically distinguishing time of disease from symptom onset, from hospital admission, and from ICU admission may help stratify which patients would benefit greatest from manipulating the host response.

3) What Specific Cells Are Directly Targeted by SARS-CoV-2 and How Do These Cells Respond?

What is Known. Although the defining characteristic of COVID-19 is infection-induced hypoxemic respiratory failure, dysfunction has also been reported in multiple other organs. Designing appropriate therapeutic responses requires a knowledge of the mechanisms that the virus uses to enter cells, which may subsequently lead to cellular and then organ dysfunction. SARS-CoV-2 has been isolated in a variety of cells and locations in COVID-19 patients.[34] Nucleocapsid antigen or viral RNA has been directly identified in respiratory epithelial cells and in bronchoalveolar lavage fluid, bronchoscopic brush biopsies, sputum samples, nasal and pharyngeal swabs, feces, blood and urine,[35] and in the cytoplasm of gastric, duodenal, and rectal glandular epithelium of a COVID-19 patient[36] Coronavirus particles were detected on an endomyocardial biopsy in a patient who later died of Gram-negative pneumonia.[37] Autopsy results revealed substantial deposition of SARS-CoV-2 in pneumocytes and renal tubular epithelial cells, with lesser amounts in heart, liver, brain, and blood.[3,83] Microscopic examination of the kidney specimens revealed clusters of spiked particles in tubular epithelium and podocytes.[40]

Primarily based on studies of other coronaviruses, it appears that SARS-CoV-2 enters cells after the viral spike protein attaches to a cell surface receptor (most commonly type 2 angiotensin-converting enzyme [ACE2][41]), and the spike is cleaved by a protease, typically transmembrane serine protease 2 (TMPRSS2).[42] Use of single-cell RNA sequencing has demonstrated ACE2 and TMPRSS2 coexpression in nasal and bronchial epithelium, especially type II cells.[43–46] Low level expression has also been noted in airways, cornea, esophagus, small intestine, colon, liver, gallbladder, heart, kidney, bladder, and testis.[4,74] Within the lung and airway, ACE2 is expressed in multiple airway cell types and is most highly expressed in the nasal epithelium. One study using samples obtained via bronchoscopy noted that ACE2 is more highly expressed in the trachea and large airway epithelium than in small airway epithelia.[49] TMPRSS2 is more broadly distributed—thus ACE2 is likely a limiting factor in infection.[50] Coexpression has also been identified in skin and other "barrier" sites.[51] Of note, it has been postulated that the proteolytic function of TMPRSS2 may also be served by cathepsin B or L, which are coexpressed with ACE2 in many different cell types, or by furin, because these proteases are involved in cellular entry of other coronaviruses.[49,52,53] In contrast, despite the fact that both SARS-CoV and Middle East respiratory syndrome coronavirus (MERS)-CoV have been shown to infect them, ACE2 is rarely expressed in immune cells.[5,45] Cleavage of a protein by a protease, such as TMPRSS2, uncovers a region that then fuses with the membrane of the target cell[42,56] which can result in either direct entry of viral RNA into host cells or the formation of vesicles containing viral replicative material. Although ACE2 also binds other coronaviruses, the affinity of this receptor for SARS-CoV-2 is much higher, explaining its high rate of infectivity[57,58] SARS-CoV-2 can also enter cells via other receptors, which is especially important given the absence of ACE2 on immune cells. CD147 is expressed on T cell catheters and is used for cell entry by the SARS-CoV, HIV-1 and measles viruses.[5,96] CD147 interacts with numerous intracellular molecules expressed in immune cells and epithelium and suppresses T cell receptor-dependent activation.

Viral entry into and replication within cells can coopt cellular machinery, preventing cells from performing their normal functions. Alternatively, SARS-CoV-2 might stimulate intracellular pathways to excess, resulting in exhaustion of available cellular energy supplies. Additionally, COVID-19 is associated with damage to and loss of endothelial cells (see question 7 below), accompanied by intra- and extravascular thrombin deposition, occasional microthrombi, and trapped neutrophils (question 9 below). Cardiac abnormalities also include thrombosis, but the most prominently reported abnormality is acute multifocal cardiomyocyte injury. These findings provide potential explanations for COVID-induced myocardial dysfunction.[61] Reports of arrhythmias suggest the presence of electrophysiology abnormalities[62] although these changes may reflect effects of SARS-CoV-2 infection on the extracellular milieu.

Acute kidney injury is the most common extrapulmonary manifestation of the disease and is associated with an increased risk of mortality.[6,36] Proteinuria and/or hematuria are both common. The presence of SARS-CoV-2 viral particles in tubular cells and in the urine of patients with COVID-19 as well as leak of protein into the urine suggests that the virus causes direct damage to renal cells, but functional data have not been reported.[65] Autopsy studies have demonstrated renal abnormalities consistent with ischemia-reperfusion injury. Despite speculation that it is involved in COVID-19 pathology,[66] evidence implicating dysfunction of the renin-angiotensin-aldosterone system (RAAS) has not been noted.

Xiao et al[36] reported that 39 of 73 patients with COVID-19 had SARS-CoV-2 RNA in stool. A wide spectrum of GI abnormalities, particularly diarrhea, suggests that SARS-CoV-2 interferes with absorption,[67–71] but definitive studies are lacking. Mild-to-moderate elevation of liver enzymes suggests some degree of hepatocellular injury, whereas the presence of congestion, cholestasis, and steatosis suggests either obstruction or defective fatty acid breakdown.

Gaps in Knowledge. It remains unknown whether other cell types are susceptible to SARS-CoV-2 infection and what other tissues can harbor the virus. It is unclear whether entry of SARS-CoV-2 into cells not expressing ACE2 is dependent on CD147, CD25, or other cell surface proteins. It is unknown whether proteases other than TMPRSS2 contribute to virus uptake. Further, it is unclear if blocking the receptors or impairing the activity of proteases limit viral entry into cells. Although there are multiple drugs that directly affect the mechanisms that facilitate entry of SARS-CoV-2 into cells,[72–74] it remains to be determined if these compounds limit SARS-CoV-2 uptake in vivo or are effective in treating COVID-19. Although CD26 has been reported to be a site of entry for MERS-CoV, it remains to be determined if it serves a similar function for SARS-CoV-2 entry into immune cells.[75] Finally, little is known about how SARS-CoV-2–induced cellular dysfunction contributes to organ dysfunction and poor outcomes. Proposed mechanisms to explain SARS-CoV-2–induced organ dysfunction include (but are not limited to): 1) direct viral toxicity, 2) abnormalities secondary to ischemia caused by vasculitis/thrombosis/thrombo-inflammation, 3) immune dysregulation, and 4) RAAS dysregulation.

Future Directions. Given the high expression of viral-uptake genes in the nasal epithelium, intranasal drug administration should be explored. Another line of investigation is to determine how interfering with ACE2 and/or TMPRSS2 (or other receptors or proteases) affect/s normal cellular function. Further exploration of SARS-CoV-2 effects on the pathways used to activate/deactivate immune cell function is worth exploring. The hypothesis that vascular abnormalities underlie pathology in other organ systems should also be examined.

4) Can Early Data Be Used to Predict Outcomes of COVID-19 and, by Extension, to Guide Therapies?

What is Known. Respiratory symptoms of COVID-19 are extremely heterogeneous, ranging from none (asymptomatic) to minimal symptoms to significant hypoxia with ARDS.[76] Although the SARS-CoV-2 virus primarily affects the respiratory system, multiple organ failure can also occur.[77] Under the best of circumstances, it is valuable to be able to rapidly identify patients who are likely to require and benefit from hospitalization, ICU admission, and specific therapies. In times of pandemic when resources may be limited, this becomes absolutely vital. This may not only help to determine prognosis and guide the selection of best management strategies for a given individual, but it will also ultimately benefit the largest number of individuals in a resource-limited setting.[78]

Clinical data that are useful to the bedside practitioner for prognostication should be simple, objective, easy to collect, and correlated with physiologic variables in acute illness. Multiple risk factors have been identified that are associated with poor prognosis in COVID-19. Although viral load is not a good marker for early categorization,[79] patient characteristics such as older age, male gender, higher body mass index, and comorbidities[34,80] have been correlated with worse prognosis. Additionally, racial and ethnic disparities have also been identified[81,82] as being associated with higher risk of poor outcomes in certain countries. Preliminary studies suggest that lymphocyte count may be useful in predicting moderate, severe, and critical illness 1 day after disease onset with an area under the curve (AUC) at 0.81,[79] although this needs to be validated in larger trials. Imaging techniques such as high resolution CT scan[83] or lung ultrasound[84] have also been proposed as tools to evaluate the severity of illness at hospital admission and to predict clinical course and outcome. However, the utility of these modalities may be limited by availability, infection control concerns, operator skillset, and interobserver variability. Genetic factors[85] and ex vivo responses of cells may be informative in evaluating potential disease evolution,[86] but these are generally more applicable for research and are less easily measurable at the bedside for broad clinical use.

Gaps in Knowledge. Beyond obvious respiratory distress and shock, the clinical factors indicating the need for hospitalization and ICU admission vary between hospitals and geographic locations. When a patient is admitted, many variables can be incorporated in a minimal dataset to help identify severe disease earlier and subsequently improve prognosis.[87] However, the specific elements to be collected—and their prognostic power—remain to be determined. Multiple biomarkers have been proposed to carry prognostic significance and information about patient trajectory in COVID-19, including (but not limited to) C-reactive protein, serum amyloid A, interleukin-6, lactate dehydrogenase, lymphocyte count, neutrophil-to-lymphocyte ratio, D-dimer, CARDIACTROPONIN, RENALBIOMARKERS, ANDPLATELETCOUNT.[88] However, the relative importance of each of these has yet to be fully delineated. Further, the majority of these indices have been described in isolation without a concerted effort to compare the respective value of these variables/markers or to construct scores combining the different variables (see question 13 below). Notably, although a COVID-19 Scoring System based on age, presence of chronic heart disease, lymphocyte count, procalcitonin, and D-dimer HASBEENPROPOSED,[89] it has not been validated in different settings. Finally, there is significant variability in case fatality across hospitals between geographic regions, according to the prevalence of COVID-19 and need for surge capacity,[90–92] whereas the overall mortality for COVID-19 in the ICU has decreased significantly worldwide over time.[2] Variability between locations at a point in time and then further longitudinal variability may limit the generalizability of the capacity to predict outcome.

Future Directions. There is a need for a defined dataset that could potentially include biomarkers, imaging techniques and physiologic variables that phenotype patients with COVID-19. Beyond basic demographic information, identifying clinical factors that can be used to identify patients at risk of disease progression is an important research direction as are identification of biological factors that carry prognostic significance. A "big data" approach using datasets from multiple healthcare systems and multiple countries represents an attractive approach to identify these factors. Conceptually, large datasets could be examined for optimal sensitivity/specificity in a training set using advanced machine learning approaches. This should then be validated using different datasets. Balancing ease of reporting (likely seen with less data elements) with accuracy of prediction (likely seen with more data elements) would be critical in optimizing resulting algorithms. The relative importance—if any—of various imaging techniques requires further study as does prioritization of which imaging should be obtained. Finally, it is important to identify which outcome should be considered. Although the majority of attention to data has focused on survival in hospitalized patients, other patient-centric outcomes should be considered including longer term survival (beyond 28 d), need for sustained organ support (supplemental oxygen, renal replacement therapy), cognitive function, emotional well-being, and long-term physical outcome.

5) What Is the Role of Prone Positioning and Noninvasive Ventilation in Nonventilated Patients With COVID?

What is Known. Treatment of ARDS in general requires endotracheal intubation and mechanical ventilation. Intubated patients with ARDS can benefit from prone positioning, which improves oxygenation and reduces mortality in non-COVID-19–related moderate-to-severe ARDS.[9,39] Ventilated patients with COVID-19 have also shown an excellent response to prone positioning.[16]

The COVID-19 pandemic has significantly impacted the capacity of hospitals worldwide, such that patients with moderate-to-severe ARDS are often being treated outside the ICU. The desire to prevent the need for intubation and mechanical ventilation[95,96] has produced significant interest in evaluating the efficacy of prone positioning in nonintubated patients, a strategy that was only rarely tried prior to COVID-19. Several small trials and case series have assessed the feasibility and effectiveness of prone positioning in awake, nonintubated patients with COVID-19 and found the strategy to be promising.[97–104] In a prospective, cohort study, Coppo et al[97] proned 56 patients for at least 3 hours with a confirmed diagnosis of COVID-19–related pneumonia and receiving supplemental oxygen or noninvasive continuous positive airway pressure. PaO2/FIO2 ratio increased from 180.5 ± 76 mm Hg in supine position to 285.5 ± 113 mm Hg in prone position (p < 0·0001). These authors concluded that prone positioning was feasible and effective in rapidly ameliorating blood oxygenation in awake patients with COVID-19–related pneumonia requiring oxygen supplementation. In contrast, a multicenter, adjusted cohort study of 199 patients found that patients on high-flow nasal cannula had a 40% chance of being intubated regardless of whether they had awake prone positioning.[105]

Gaps in Knowledge. The efficacy of prone positioning in awake, nonintubated patients with COVID-19–related pneumonia remains unclear. Conceptually, prone positioning in spontaneously breathing patients treated with high-flow oxygen therapy or noninvasive ventilation may prevent the need for intubation, avoiding the risks connected with the ICU stay and ventilator-induced lung injury.[106] A number of questions remain regarding both the safety and efficacy of prone ventilation in nonintubated patients. First, does prone positioning actually prevent intubation and reduce the risks associated with invasive mechanical ventilation? There is no way to know based upon the existing literature whether patients who are not intubated using this strategy represent a self-selected group who might otherwise not have progressed to intubation. Next, are there risks associated with delaying a necessary intubation in patients who are selected to prone first? Further, how will treatment failure be recognized on proned patients such that they are rapidly identified and intubated? Also, is there a length of time that spontaneously breathing patients should prone for (either maximal duration or time per session)? Next, it remains to be determined who the best (or at a minimum appropriate) candidates are to receive awake prone positioning.

Future Directions. Although awake proning for nonintubated patients has rapidly been adopted at multiple centers worldwide, well-conducted trials using a stringent methodologic approach are required to determine whether this should be considered standard of care in a subset of patients with severe COVID-19. Trials should directly examine whether prone ventilation prevents intubation in patients with hypoxemic respiratory insufficiency from COVID-19. In addition to efficacy, trials need to determine whether proning awake patients lead to increased complications from delayed extubation. Finally, if self-proning is effective, trials are required to determine if there is an optimal length of time to prone for, balancing the physiologic benefits of proning with patient comfort.

6) Which Interventions Are Best to Use for Viral Load Modulation and When Should They Be Given?

What is Known. Many drugs have been proposed as candidates for potential viral modulation although only a small minority have gone through rigorous clinical trials. Remdesivir is an adenosine nucleotide analog that interferes with the function of the viral RNA-dependent RNA polymerase.[107] A multinational National Institutes of Health–sponsored trial of 1,063 patients compared remdesivir with placebo in a double-blind randomized trial[108] with a median time of 9 days from symptom onset to randomization. Remdesivir significantly reduced the time to recovery compared with placebo from 15 days to 11 days in the primary outcome of the trial (recovery rate ratio, 1.32; 95% CI, 1.12–1.55). The benefit of remdesivir was most pronounced in the subgroup of hospitalized patients who required supplemental oxygenation (but not high-flow nasal cannula), and this patient population had a decreased mortality at day 14 in subgroup analysis. In contrast, there was no benefit to remdesivir in either time to recovery or 14-day mortality in patients who required high-flow oxygen, noninvasive ventilation, mechanical ventilation, or ECMO. A concurrent industry-sponsored trial randomized 402 nonintubated patients to receive either 5 days or 10 days of remdesivir.[109] No difference in outcomes were noted between treatment lengths. Finally, a multicenter double-blind trial compared remdesivir with placebo in China. A total of 237 patients were enrolled (158 received remdesivir, 79 received placebo). No difference in time to clinical improvement or mortality was identified. However, the study was terminated early because it did not reach its target enrollment of 453 patients, making it difficult to interpret as it was underpowered to reach its primary endpoint.[110]

Chloroquine and hydroxychloroquine with or without azithromycin have been studied extensively in COVID-19. Despite early enthusiasm for the drugs (alone or in combination), these drugs have not been shown to be beneficial in multiple randomized trials and large retrospective observational trials,[111–113] and they are not recommended for use. Further, safety concerns have been raised with adverse effects of these agents being higher than seen in placebo arms.

Plasma from patients who have recovered from COVID-19 may contain antibodies that could both suppress the virus and modify the host inflammatory response.[114] After early usage of convalescent plasma in China to treat hospitalized patients with COVID-19,[115] randomized trials began in multiple countries. In the United States, an expanded access program was developed in parallel with these trials in order to provide broader access to this agent, but without the intent to generate definitive data on safety or efficacy. Ultimately, over 70,000 patients in the United States received this therapy in an unblinded fashion, without a control arm to determine whether or not it was effective. Based upon a retrospective subset analysis from the expanded access program suggesting that high-titer convalescent plasma might be more effective compared with low-titer convalescent plasma in subsets of hospitalized patients, the Food and Drug Administration in the United States issued an Emergency Use Authorization—which is not drug approval but rather facilitates availability and unapproved use during a public health emergency.[116]

Lopinavir is used as a retroviral therapy to exert an anti-HIV effect through HIV-1 viral protease inhibition, and ritonavir boosts the effects of lopinavir.[117] In the context of coronaviruses, the 3CL protease is the target. The combination of lopinavir and ritonavir has been used for the treatment of MERS.[118] A randomized trial of these two agents in 199 patients did not find a benefit in time to improvement or mortality compared with usual care, nor was a difference noted in detectable viral RNA.[119] Notably, the treatment was stopped in 13.8% of patients because of adverse events. Additionally, a pharmacokinetic study showed that plasma concentration using typical doses of lopinavir and ritonavir is lower than levels that may be needed to inhibit SARS-Cov-2 replication.[120]

Ivermectin is an antiparasitic drug that inhibits importin alpha/beta-1 nuclear transport proteins, which are part of an intracellular transport process that viruses interfere with.[121] There is a theoretic rationale for using this agent (which has in vitro activity against multiple viruses) in COVID-19 as it inhibits a pathway that suppresses host-antiviral response. However, although ivermectin inhibits SARS-Co-V-2 in vitro, pharmacokinetic and pharmacodynamic studies suggest that achieving the plasma concentration necessary in patients would require doses 100 times higher than those approved for human use and that predicted plasma and lung tissue concentrations would be lower than the half-maximal inhibitory concentration against the virus.[122–125]

Gaps in Knowledge. The efficacy of remdesivir in patients with COVID-19 on noninvasive ventilation or mechanical ventilation is unclear. If there are benefits to remdesivir, the ideal timing of when to start the drug and how long after symptom onset it might be effective remain to be determined in addition to the optimal treatment length. It is also unclear if the combination of corticosteroids and remdesivir is superior to corticosteroids alone in critically ill patients. It is unknown whether any other antivirals (either those listed above or other experimental agents) improve any patient-centric outcome in COVID-19.

Of note, although this article was under peer review, the results of the World Health Organization-sponsored SOLIDARITY trial on repurposed antiviral drugs for COVID-19 was released in preprint form.[12] This study included 11,266 adults in 30 countries. Patients were randomized to remdesivir, hydroxychloroquine, lopinavir, interferon plus lopinavir, interferon alone, or no study drug. None of the interventions impacted overall mortality, initiation of ventilation, or hospital stay. This trial had not yet been subjected to peer review when this article was resubmitted.

Future Directions. Randomized controlled trials need to be performed on remdesivir specifically examining critically ill patients. Preliminary data need to be generated on other antiviral candidates which, if positive, should lead to randomized controlled trials.

7) Do Endothelial Cells Play a Central Role in Driving and/or Potentiating COVID-19 Disease?

What is Known. Observations implicate the endothelium as critical in COVID-19 illness. The ACE2 enzyme, important for SARS-CoV-2 binding and internalization,[56] is present on endothelial cells and plays a role in angiotensin signaling balance, with effects on apoptosis, vasoconstriction, and inflammatory signalling.[126,127] SARS-CoV-2 has been reported to invade and replicate in capillary-like organoids in vitro.[128] Data from patients with COVID-19 demonstrate viral infection of endothelial cells,[129] a pauci-immune vasculitis,[130] thrombin microvascular clot,[129–132] megakaryocytes,[132,133] and a characteristic intussusceptive angiogenesis that is unique from other viral pneumonias.[131] Von Willebrand factor, factor VIII, and thrombomodulin are products released from endothelium that have been demonstrated to be elevated in patients with severe COVID-19.[134] To determine the impact of SARS-CoV (the virus that caused the SARS outbreak) in complement, a preclinical model examined the role of the virus in mice deficient in C3, the central component of complement.[135] Compared with infected wild type mice, C3 knockout mice had less weight loss and less respiratory dysfunction despite equivalent viral loads in the lung, with reduced cytokine and chemokine levels in both the lungs and the blood, suggesting the complement system plays an important role as a mediator of coronavirus disease by regulating a systemic preinflammatory response to SARS-CoV infection. The relevance of these preclinical findings is enhanced by observations of significant complement deposition in the interalveolar septa and cutaneous microvasculature in the lung and skin of two patients with COVID-19.[130]

Gaps in Knowledge. These observations strongly suggest both direct and indirect viral involvement of the endothelium in COVID-19–related disorders such as ARDS, acute kidney injury, thrombosis, and cardiovascular events. At the same time, these data are all associative (except for preclinical mice data using a related virus) and thus cannot support a firm conclusion that the endothelium plays a crucial role in the pathogenesis of COVID-19. An important gap is in understanding the endothelial role in systemic viral dissemination. The role of complement-mediated vasculitis, described in other coronavirus infections, needs to be better characterized for SARS-CoV-2 beyond case reports. Whether the angiogenesis and thrombosis observed in autopsy findings represents a mechanistic step in the pathobiology of COVID-19 or a response to disease drivers is largely unexplored.

Future Directions. Future investigations should specifically address whether endothelial abnormalities reflect primary steps in the pathogenesis of COVID-19 or secondary effects driven by other cell types/organ systems. Research should focus on imaging techniques (radiographic imaging, staining, markers), in vivo models (animal models, tests of endothelial function), and in vitro studies using primary human endothelial cells and should address different vascular types in studies using SARS-CoV-2. Ideally, such models would provide mechanistic approaches to therapies and offer a means to assess for risk of severe infection or a worsening trajectory. These approaches should address existing endothelial-targeted therapies, ACE2 viral binding and inactivation, manipulation of the coagulation and immune systems, apoptosis, and endothelial cell infection. Further work should also investigate whether endotheliopathies in COVID-19 predict future cardiovascular disease.

8) What Is the Best Approach to Anticoagulation in Patients With COVID-19?

What Is Known. A high prevalence of venous and thromboembolic events has been reported in COVID-19 patients. Pulmonary embolism (PE) is the most important but increases in deep vein thrombosis (DVT), myocardial infarction, ischemic stroke, arterial thrombosis, and repeated clotting of continuous renal replacement therapy circuits have been identified.[61,136–142] Notably, the high prevalence of thromboembolic events has occurred despite patients being on prophylactic (often higher doses than used for other ICU patients) or even therapeutic anticoagulation. Some of these thrombotic events represent the initial presenting symptoms for hospital admission and may be central to the clinical presentation of COVID-19.[143,144] These diagnosed and sometimes undiagnosed events contribute to the high mortality of COVID-19 patients admitted to the ICU. Increased levels of procoagulant factors including fibrinogen, and D-dimer are associated with higher mortality in patients with COVID-19.[145] Clinical observations and autopsy findings seem to distinguish COVID-19–associated coagulopathy from thrombotic microangiopathy and disseminated intravascular coagulation (DIC).[145–149]

A retrospective study of 4,389 COVID-19 patients demonstrated that compared with no anticoagulation, therapeutic, and prophylactic anticoagulation were associated with lower in-hospital mortality. There was a trend toward improvement in patients started on therapeutic coagulation within 48 hours of admission (risk ratio, 0.86; 95% CI, 0.73–1.02; p = 0.08). In the entire cohort, 89 patients had major bleeding including 3.0% of patients on therapeutic anticoagulation. Of 26 autopsies, 11 (42%) had thromboembolic disease not clinically suspected.[150]

Gaps in Knowledge. Despite the widely seen increase in thromboembolic disease, the optimal method for anticoagulation is not known. Multiple study protocols have been published for existing, on-going studies that should help to clarify this question.[151–154] However, it is currently unknown whether there is a benefit to empiric therapeutic anticoagulation without clearly documented thromboembolic disease, and if so, in which patient population. Further, it is unclear whether there is a role for higher dose prophylaxis and, if so, in which patients. It is also unclear if specific anticoagulants are preferable in the setting of COVID-19. This lack of clarity has led to recommendations from professional societies regarding optimal anticoagulation that are not entirely harmonized and emphasize patient specific factors in decision-making until additional data are available.[155–159]

Future Directions. Clinical trials are needed to determine how to identify patients who may benefit from therapeutic anticoagulation. Further, trials are needed to determine whether to use regular or augmented doses of prophylactic anticoagulation and if a specific patient population benefits from higher doses of prophylactic anticoagulation. The optimal agent(s) for anticoagulation need to be determined. Studies also need to be performed to determine who should be screened for DVT (all patients vs a subset based upon clinical or laboratory findings) and how often screening should be carried out.

9) What Are Mechanisms of Vascular Dysfunction and Thrombosis in COVID-19 and Why Are Thrombotic Manifestations More Common in SARS-CoV-2 Infection Compared With Other Infections or Causes of Critical Illness?

What Is Known. Severe COVID-19 causes coagulopathy and vascular complications, including DVT, PE, ischemic strokes, myocardial infarction, peripheral arterial thrombosis, and limb ischemia.[138,160–164] Thrombosis of extracorporeal circuits has been reported during renal replacement therapy and ECMO.[160,161] These thromboembolic complications occur even in patients receiving prophylactic or therapeutic anticoagulation. Autopsy studies have revealed microvascular fibrin deposition and thrombi in organs, intravascular trapping of neutrophils, pulmonary infarcts, severe endothelial injury with intracellular virus, and microangiopathy, as well as higher levels of angiogenesis and new vessel growth than observed in patients infected with influenza virus.[131,165]

Elevated D-dimers have been reported in patients with severe COVID-19 disease, and higher levels of D-dimers early in the course have been shown to predict worse outcomes in preliminary studies although this association has been disputed.[137,160,161,166,167] Other early findings include normal to elevated fibrinogen, normal to minimally elevated prothrombin time (PT) and partial thromboplastin time (PTT), and normal platelet counts, although studies suggest that platelets are activated.[137,148,161,166] One single-center study of 73 patients on mechanical ventilation due to COVID-19 demonstrated positive lupus circulating anticoagulant in 85% of patients, although this finding was not associated with thrombotic complications.[168]

Thrombotic complications of COVID-19 occur in large vessels on the venous and arterial sides as well as in the microcirculation, which is distinct from other infectious causes of coagulopathy. Available data suggest that in the early stages, COVID-19–associated coagulopathy is not DIC.[137,161,169] However, studies suggest that over the course of unremitting SARS-CoV-2 infection, the coagulopathy evolves to be consistent with DIC (elevated, PT, PTT, D-dimers, and fibrin degradation products).[137,157] COVID-19 also induces hyperviscosity, with plasma viscosity above 3.5 centipoise associated with thrombotic complications.[170] Further, COVID-19 has been demonstrated to induce fibrinolysis shutdown as measured by rotational thromboelastometry in a study of 21 patients.[171] Notably, 89% of patients with venous thromboembolism met criteria for fibrinolysis shutdown in this study.

Gaps in Knowledge. The mechanisms of coagulopathy and vasculopathy in COVID-19 are poorly defined. There are gaps in the understanding of 1) host and microbial triggers of coagulopathy and vasculopathy, 2) the roles of inflammation, and of different cell populations (e.g., endothelial cells, platelets, leukocytes) in driving coagulopathy and vasculopathy, 3) the role of angiogenesis in ARDS and other organ injury and failure, and 4) why the thrombotic complications of COVID-19 are different from other infectious diseases. Although mechanisms remain to be determined, hypoxia-induced vasoconstriction with reduced blood flow, endothelial cell inflammation and dysfunction, and hypercoagulability with high concentrations of von Willebrand Factor and other prothrombotic constituents (Virchow's Triad) could lead to higher risk of micro- and macrovascular thrombosis in severe COVID-19 patients.[40,130,172–175] Additionally, identification of which plasma proteins cause COVID-19–induced hyperviscosity remains to be determined.

Future Directions. Further preclinical, translational, and clinical studies are needed to better understand the mechanisms of COVID-19–induced coagulopathy and vasculopathy. Specific areas of investigation include understanding the role of different cell populations (leukocytes, endothelial cells, epithelial cells, platelets) and of inflammation in driving the vascular complications of COVID-19 and defining similarities and differences in the mechanisms and course between COVID-19–associated vasculopathy and coagulopathy versus other infectious diseases. Clinical trials to determine whether targeting hyperviscosity via plasma exchange improves outcomes are warranted. Further, additional studies are required to determine the physiologic importance of fibrinolysis shutdown and if this can be targeted clinically.

10) Do the Long-Term Sequelae of Severe and Critical COVID-19 Disease Differ From Sequelae of Sepsis/ARDS?

What Is Known. Over the past 15 years, multiple studies have demonstrated that both acute respiratory failure and general critical illness are often followed by multifaceted and long-lasting sequelae.[176,177] The term "postintensive care syndrome" was established in 2012 to raise awareness of new or worsened physical, cognitive, and emotional symptoms following critical illness.[177] Critical illness survivors also have increased risk for recurrent infection, rehospitalization, and death.[176] Many ICU survivors cannot return to work or regain functional independence, even 6–12 months after acute illness.[178,179] A meta-analysis of 28 studies including 2,820 patients infected with SARS or MERS found high rates of posttraumatic stress disorder (38%), depression (33%), and anxiety (30%) at 6 months post illness, as well as pulmonary dysfunction, reduced exercise tolerance, and reduced health-related quality of life.[180]

Due to the recency of the pandemic, the long-term outcomes from severe and critical COVID-19 are unknown. However, survivors are anticipated to experience similar issues,[181] creating a large demand for rehabilitative services.[182] Early reports from Italy indicate high rates of persistent symptoms at 60 days, particularly fatigue.[183] Similarly, a study from the United States found that 40% of patients surviving hospitalization for COVID were unable to resume normal activities within 60 days of discharge.[184] There are increasing articles in lay media about COVID-19 survivors with persistent symptoms—coined "long COVID"—and patients have spontaneously formed online support groups.[185,186]

Gap in Knowledge. Because of the newness of the pandemic, relatively limited information exists about long-term outcomes of patients who recover from COVID-19, the expended duration of sequelae, and how these vary by severity of initial COVID-19 disease. It is unclear if long-term sequelae of severe and critical COVID-19 are similar to postintensive care syndrome or if they are distinct (albeit potentially overlapping) entities with unique causes and manifestations. The International Sepsis Forum recently identified key gaps in sepsis survivorship research.[187] Gaps include limited information for specific patient populations (e.g., in lower income countries and children), few studies of in-hospital or posthospital interventions to enhance recovery, and limited data on how to identify patients who are likely to benefit from recovery-focused interventions.[187]

Future Directions. Longitudinal cohort studies are needed to define the prevalence of individual sequelae after COVID-19, the average duration of sequelae, and the variation in prevalence across patient subgroups (e.g., subgroups defined by disease severity). Ideally, such studies would be embedded within existing longitudinal cohorts to capture patients' baseline pre-COVID health status or use proxy-respondent questionnaires or interrogation of available health records to capture this. Second, interventional or comparative effectiveness studies are needed to determine the best practices during and after hospitalization to limit physical and cognitive sequelae. For example, the optimal timing, duration, intensity, and patient selection for mobility interventions remain unclear. Third, studies are needed to improve the implementation of practices already known to improve long-term outcomes (e.g., ABCDEF bundle) but which are more difficult to implement during the pandemic due to high census and infection control precautions. Finally, given the number of patients surviving severe COVID-19 disease, there is an urgent need to test scalable follow-up and recovery programs such as centralized telehealth clinics to screen for and address new cognitive impairments, functional disability, new and worsening symptoms, and chronic comorbidities.

11) How Does SARS-CoV-2 Impair Immune Function?

What is Known. Immunologic profiling reveals a significant heterogeneity associated with severity of COVID-19 and temporal changes during the course of the disease.[29,188–191] Invariably, SARS-CoV-2 infection results in broad changes in circulating immune cell populations, humoral responses, and cellular functionality. Patients with COVID-19 often have severe persistent lymphopenia despite having an elevated WBC count. This finding is not associated with lymphocyte recruitment to the respiratory tract but may result from a peripheral depletion secondary to systemic inflammation or apoptosis.[192] It is also associated with an oligoclonal plasmablast expansion and heterogeneous T cell activation affecting both memory CD4+ and CD8 + T cells.[188,193] In a study of 44 patients, Qin et al[20] noted decreases in the total number of CD4+ and CD8+ T cells, B cells and Natural Killer cells at admission. Numbers of both helper and cytotoxic T cell subsets were also decreased, but the T helper/T suppressor ratio was normal as was theinterferon γ response of these cells to phorbol myristate acetate/ionomycin. In contrast, other researchers have found impaired cytokine production in CD8+ T cells.[194] Studies have also found that surviving T cells isolated from COVID-19 patients appear functionally exhausted.[195,196]

Neutrophilia is also an important hallmark of severe COVID-19. The loss of cell surface Fcγ receptor III CD16 on natural killer and neutrophils phenotypically signals an exaggerated immature neutrophilia (CD10LowCD101CXCR4±) and consequent emergency myelopoiesis and altered neutrophil response.[189,197,198] Neutrophil recruitment, entrapment in organs, and neutrophil extracellular traps-osis development may explain vascular thrombosis, vasculitis, and postinfective complications.[199–201] Monopenia is also systematically identified in the most severe cases. A decrease of nonclassical CD14LowCD16High, accumulation of human leukocyte antigen (HLA)–DRLow classical monocytes, and consecutive decrease in monocytes HLA – DR isotype expression are associated with acquired immunodepression and associated decrease in monocyte functionality. It has also been reported in a study of 987 patients with life-threatening COVID-19 pneumonia and 1,227 healthy individuals that 10% of patients have neutralizing autoantibodies against type I interferons.[202] These neutralizing autoantibodies, like inborn errors of type I interferon production, may tip the balance in favor of the virus in the severe forms of the disease, with insufficient innate and adaptive immune responses. The authors show multiple lines of evidence to suggest that these autoantibodies preceded infection with SARS-CoV-2 and concluded that they accounted for the severity of the disease.

Gaps in Knowledge. Although the early days of the pandemic have been accompanied by an explosion of knowledge on how the immune system is impacted by SARS-CoV-2 infection, significantly more remains to be discovered. A more comprehensive analysis of cells in both the innate and adaptive immune system in multiple immunologic components remains to be performed. How the immune system is impacted by (and/or helps propagate) the severity of the disease is mostly unknown. Although preliminary studies have longitudinally evaluated the immune system over time, the kinetics of how viral infection impacts the immune system are less well understood, and very little is known about long-term effects after patients are discharged from the hospital. Although the impact of medical comorbidities, age, and gender on the immune system have all been extensively analyzed outside the scope of COVID-19, how each of these interacts with the host immune response during SARS-CoV-2 infection remains to be determined. Whether immunophenotypes can be used for either prognostication or for targeted treatment via precision medicine is unknown. The importance of neutralizing autoantibodies remains to be more definitively determined.

Future Directions. Variability of infection severity as well as occurrence of postinfective complications, as seen in children with acute myocarditis,[203,204] suggests complex immune-pathophysiology involving at a minimum 1) age-specific immunity (e.g., early-age, senescent immunity); 2) immunization status (e.g., against coronavirus species, long-/short-lasting immunization, immunoglobulin specificity, and subclasses); 3) serologic response (e.g., immune complex, complement system); and 4) impact of acquired- or induced-immunosuppression (e.g., pregnancy, systemic diseases, cancer, immunotherapy). To unravel this complex response requires investigations in critically ill patients with appropriate clinical matching to characterize the immune system and leads to potential immunomodulatory therapy. This can be coupled with preclinical studies, autopsy studies, and larger scale biorepository studies to lead to a more comprehensive understanding of the immune response in COVID-19.

12) What Are the Predictors of ICU Admission in COVID-19?

What is Known. COVID-19–related disease has placed an enormous strain on ICU resources globally, and in some areas, the demand for ICU services has outstripped the supply. Being able to identify patients requiring ICU care is therefore critical to proper utilization of a scarce resource. There are now numerous reports on the epidemiology and characteristics of COVID-19 patients. Some patient characteristics associated with ICU admission, include male sex, obesity, and chronic kidney disease.[205] Among the initial biological variables (as discussed in question 4), lymphocytes and neutrophil-to-lymphocyte ratio may be beneficial,[79,206,207] whereas imaging modalities such as chest CT scan and lung ultrasound may also be helpful.[84,208] In addition, detection of viral RNA may have prognostic implications since SARS-CoV-2 RNA was detected more frequently, and levels were higher in a recent study of 123 patients admitted to the ICU.[209]

Initial attempts have also been made to derive and validate risk prediction scores. An example is the COVID-19–specific score (COVID-GRAM) which used characteristics of COVD-19 patients at time of admission to the hospital in a development cohort of 1,590 patients to predict critical illness (defined as admission to the ICU, mechanical ventilation or death).[210] Factors included in the score were chest radiographic abnormality, age, hemoptysis, dyspnea, unconsciousness, number of comorbidities, cancer history, neutrophil-to-lymphocyte ratio, lactate dehydrogenase, and direct bilirubin. The score performed well in a validation cohort of 710 patients with an AUC of 0.88. Liu et al[211] compared the performance of seven existing severity scores in predicting mortality in 673 COVID-19 patients admitted to the hospital floor. They found the National Early Warning Score had the highest performance with an AUC of 0.88.[211] In addition, a risk score generated from 641 patients with COVID-19 for predicting ICU admission or death (70% training, 30% testing) identified five significant variables predicting ICU admission and seven significant variables predicting death with an AUC of 0.74 for predicting ICU admission and 0.83 for predicting death.[212]

Gaps in Knowledge. It is unclear whether any single variable carries significant prognostic information related to disease progression and ICU admission. Further, the generalizability of all prediction models is uncertain. Not only do they need to be validated in different healthcare systems in different countries, scores may be influenced by rapidly changing protocols for the care of COVID-19 patients.

Future Directions. As multicenter data bases become available, new severity scores will be developed that are more likely to reflect a broad population. Studies need to determine if a single score can be used universally to predict ICU admission. Further, studies need to determine if prediction scores perform better than clinical assessment, biomarkers or imaging techniques, and if they can be made simple enough for wide usage and accurate enough to have clinical applicability at the bedside. Since there are no a priori reasons to suggest that one approach will be superior to another, we advocate approaching the development of predictive tools using a wide range of approaches including (but not limited to) algorithmic development using multiple machine learning, regularized regression analyses, and logistic regression.