Passive immune therapy and other immunomodulatory agents for the treatment of severe influenza: Systematic review and meta‐analysis

Abstract Background A range of immunomodulatory therapies have been proposed as adjuncts to conventional antivirals to suppress harmful inflammation during severe influenza infection. We conducted a systematic review to assess available data of the effect of adjunctive non‐corticosteroid immunomodulatory therapy and potential adverse effects. Method We searched MEDLINE, Embase, Web of Science and clinical trial databases for published and unpublished studies, and screened the references of included articles. We included RCTs, quasi‐RCTs and observational studies of virologically confirmed influenza infections in hospitalised patients. We did not restrict studies by language of publication, influenza type/subtype or age of participants. Where possible, we pooled estimates of effect using random‐effects meta‐analysis models. Results We identified 11 eligible studies for inclusion: five studies (4 RCTs and 1 observational; 693 individuals) of passive immune therapy; four studies (3 RCTs and 1 observational; 1120 individuals) of macrolides and/or non‐steroidal anti‐inflammatory drugs (NSAIDs), one RCT of mTOR inhibitors (38 individuals), and one RCT of statin therapy (116 individuals). Meta‐analysis of RCTs of passive immune therapy indicated no significant reduction in mortality (OR 0.84, 0.37‐1.90), but better clinical outcomes at Day 7 (OR 1.42, 1.05‐1.92). There was a significant reduction in mortality associated with macrolides and/or NSAIDs (OR 0.28; 0.10‐0.77). Conclusions Passive immune therapy is unlikely to offer substantial mortality benefit in treatment of severe seasonal influenza, but may improve clinical outcomes. The effect of other immunomodulatory agents is uncertain, but promising. There is a need for high‐quality RCTs with sufficient statistical power to address this evidence gap.


| INTRODUC TI ON
Seasonal influenza is a common viral infection of the respiratory tract. It is estimated to cause more than a billion infections annually, with three to five million severe illnesses and 250 000-650 000 deaths. 1,2 The highest mortality rates have been reported in adults aged over 75 years, children younger than 5 years and residents of sub-Saharan Africa or South-East Asia.
Recommended antiviral treatments of severe seasonal influenza are currently limited to the neuraminidase inhibitors (NAIs). 3,4 While effective at shortening the duration of influenza symptoms when administered early in the course of infection, debate continues as to the extent NAIs are able to prevent progression to severe infection, the development of complications in hospitalised individuals, or reduce mortality. 5,6 An effective immune response to the influenza virus following infection is necessary for viral clearance and recovery from infection. Viral shedding is prolonged in immunocompromised patients with influenza, associated with an increased risk of emergent NAI resistance, and secondary bacterial infections. [7][8][9] But in a delicate balance, this immune response to an infection can also be harmful to the host. For example, an excessively pro-inflammatory cytokine and chemokine environment has been cited as the key explanation for the severity of human infections with highly pathogenic avian influenza, and the 1918 H1N1 "Spanish flu" pandemic. 10 This "cytokine storm" can rapidly result in multi-organ dysfunction and acute respiratory distress syndrome (ARDS). Similarly, in seasonal influenza damage to the airways and alveolae is mediated both by viral replication and by the innate immune response. 11 A range of immunomodulators for severe influenza have been proposed, 12,13 but certainty as to their relative benefits and harms is lacking. Corticosteroid therapy, for example, is widely prescribed as part of the standard of care for treatment of influenza complications such as the treatment of exacerbations of chronic obstructive pulmonary disease (COPD) and asthma. 14,15 A Cochrane review in 2017 found moderate-quality evidence that corticosteroids also reduce mortality when used in severe community-acquired pneumonia (relative risk [RR] 0.58; 95% CI: 0.40-0.84). 16 Conversely, however, in the context of severe influenza, an updated Cochrane meta-analysis published in 2019 concluded that corticosteroid therapy was associated with increased mortality (odds ratio [OR] = 3.90; 95% CI: 2.31-6.60; I 2 = 68%; 15 studies). 17 This result must be interpreted with caution as it was mainly derived from observational studies and residual bias is likely to persist as patients with more severe influenza are more likely to receive corticosteroids.
The recent 2018 Infectious Diseases Society of America (IDSA) seasonal influenza guidelines do not recommend any immunomodulatory therapies as adjunctive treatments. 3 This systematic review focuses on immunomodulatory agents other than corticosteroids for the treatment of severe influenza. Three systematic reviews of passive immune therapy (convalescent plasma/serum or intravenous immunoglobulin) for the adjunctive care of severe influenza were previously published. [18][19][20] These reviews, however, included only data from non-randomised studies and historical reports from the 1918 Spanish influenza pandemic which are of uncertain relevance today. A number of randomised controlled studies of passive immune therapy have been published since.
This systematic review was commissioned by the World Health Organization (WHO) to inform the development of clinical practice guidelines for severe influenza. It aims to provide a comprehensive and up-to-date assessment of the available data investigating the clinical effectiveness and safety of non-corticosteroid immunomodulatory therapy adjunctive to conventional antiviral medication for the treatment of severe influenza.

| ME THODS
This systematic review and meta-analysis was conducted in accordance with the PRISMA Statement 2009. 21  Only studies conducted in humans with a virologically confirmed influenza infection (seasonal or zoonotic) were included, but without restriction on age or sex. There was no restriction on date or language of publication. We included randomised trials, quasi-experimental and observational studies published in academic, peer-reviewed literature. Population studies and studies with fewer than 10 participants were excluded. For studies with an observational design, only studies that attempted to adjust for differences between groups in disease severity and/or propensity to receive the immunomodulatory therapy were included.

Clinicaltrials.gov and the WHOs International Clinical Trials
Registry Platform (ICTRP) were also searched for ongoing clinical trials, and data from these studies were included if the study was completed and results were available from online sources such as the clinical study report. Web of Science was used for citation searching by collating the bibliographies and citations of included studies to identify additional studies which may be eligible.
We included studies of any of passive immune therapy, macrolides, mTOR inhibitors, non-steroid anti-inflammatory drugs (NSAIDs) or statins. Comparator groups were those who received antiviral therapy or supportive care alone.
Our primary outcome of interest was mortality, and secondary outcomes were severity of illness (eg requiring admission to intensive care and/or mechanical ventilation), duration of hospital admission, serious adverse events, duration of viral shedding and emergence of resistance.
Studies were selected in two stages: first review of study title and abstract, followed by analysis of the full text of the article. Each study was independently reviewed by two authors, and disagreements were resolved by discussion (VL, BY). Data from studies to be included in the review were then independently extracted by these two review authors using a standardised data collection form developed and piloted for this systematic review (Appendix ).
Two review authors independently assessed the methodological quality of included studies (VL, BY). RCTs were assessed using the Cochrane Risk of Bias (RoB 2.0) tool, while non-randomised studies were assessed using a Newcastle-Ottawa Scale (NOS) modified for the purposes of this review (Appendix ). 22,23 Mortality data from individual studies were tabulated and odds ratios (OR) with 95% confidence intervals (CI) calculated using Review Manager 5.3. 24 For RCTs, data were analysed on an intention-to-treat (ITT) basis. No form of data imputation was used for participants with missing outcome data. Outcome measures that have been adjusted for confounding, such as ORs or hazard ratios (HRs) with 95% CIs, were also extracted. Where multiple adjusted analyses were presented, the results from the most complete model were collected. Ordinal logistic regression and additional statistical tests were performed using R version 3.6.1 as required.
The I 2 statistic was used to assess heterogeneity across experimental and observational studies. An I 2 value >75% was inferred to reflect substantial heterogeneity between the findings from the studies.
Outcome data from observational studies were aggregated using a random-effects meta-analysis model to pool data to reflect expected differences in the measured effectiveness of adjunctive therapies-due to differing patient characteristics, interventions and outcome definitions. Data from different therapies and study designs (experimental vs observational) were aggregated separately and combined when considered appropriate.
The five Grading of Recommendations, Assessment, Development and Evaluation (GRADE) considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) were used to assess the quality of evidence from the studies that contribute data to the meta-analyses for the pre-specified outcomes. Reasons for the decision to downgrade or upgrade the quality of studies were provided.

| RE SULTS
The search strategy was implemented on 25 January 2019, and identified 5928 articles after removal of duplicates ( Figure 1). An additional 56 articles were identified through reference tracking (though none of these met eligibility criteria to be included in the full-text review). The full text of 266 articles was scrutinised, and seven were initially identified for inclusion in the systematic review. Four additional completed studies were identified from clinicaltrials.gov, which at the time of review were unpublished, but with data available from various online sources including the clinical study record.
Two of these studies have since been published. 26,27 All studies had virologically confirmed influenza through a combination of either laboratory polymerase chain reaction or rapid antigen test.
The main reason for article exclusion was no reported assessment of the effect of adjunctive immunomodulatory treatment on patient outcomes. Seventeen articles were identified as potentially eligible, but we were unable to locate full-text copies of the article.
Fourteen of these were not published in English, and all were published prior to 2011.
Five studies on passive immune therapy, two of macrolides, one of NSAIDs, one of NSAIDs in combination with macrolides, one of mTOR inhibitors, and one of statins were identified. The study design, intervention, participants and outcome characteristics are summarised in Table 1.

| Passive immune therapy
Four randomised controlled trials-three double-blind placebo-controlled 25 and one open-label 28 -and one prospective cohort study 29 assessed the effect of passive immune therapy on mortality. A total of 693 participants were enrolled in these studies, of which 655 were analysed for the primary endpoint. A total of 323 received passive immune therapy ("experimental") and 332 did not ("control").
Sixteen participants (Beigel et al 28 ) were also included in the study by Davey et al. 26 All five studies administered a single infusion of a polyclonal pas-

TA B L E 1 (Continued)
between intervention groups at Day 3 in the three studies which reported this outcome (Beigel et al, 28 Davey et al, 26 Beigel et al 27 ).
No transfusion-associated side effects were described. The incidence of SAEs was similar in the studies from Beigel et al 27 and Davey et al. 26 Hung et al 25 did not systematically describe adverse events, but length of ICU and hospital stay were not significantly different between treatment groups.

| Macrolide and NSAID therapy
Two studies investigated the effect of macrolide therapy alone: one prospective cohort study 30 and one open-label RCT 31 (Table 1) In total, 1120 participants were enrolled for these four studies, 322 received a macrolide ("experimental") and 678 did not ("control"), while 167 received an NSAID ("experimental") and 170 did not ("control").
The cohort study enrolled adults admitted to the ICU with primary viral pneumonia and pandemic A/H1N1-2009 influenza.
Macrolides administered were clarithromycin (n = 99, 52.1%), azithromycin (n = 90, 47.4%) and erythromycin (n = 1, 0.5%). Experimental and control groups were similar in age and gender, but were more likely to be pregnant, immunosuppressed or have COPD. They were also more likely to receive adjunctive corticosteroid therapy, but had lower APACHE II scores and were less likely to have a haematological disease.
Lee et al 31  There were no deaths recorded by Lee et al. 31 The OR estimate derived from the two RCTs (Hung et al, 32 Hung et al 33

| mTOR inhibitor
One open-label RCT studied the effect of the mTOR inhibitor sirolimus in the treatment of patients critically ill with influenza in ICU (Table 1). 34 Thirty-eight participants were enrolled in this study, 19 received 2mg sirolimus for 10 days ("experimental"), and 19 did not ("control"). All subjects also received corticosteroids. Three died

| Statin
One unpublished blinded, placebo-controlled RCT of adjunctive statin therapy was identified. 35 Study investigators administered atorvastatin 40mg daily for 5-7 days in patients hospitalised with influenza who were not receiving regular statins prior to hospital admission. Results were available from the study record at clinicaltrials.
No deaths were reported in this study in either treatment group.
For the study primary endpoint, no significant difference between experimental and control groups in the change in IL-6 levels from baseline to 72 hours was reported (P = .611). However, a significant improvement in symptom score over this time period was described in the atorvastatin treatment group (P = .029). Incidence of AEs was similar between treatment groups (6.8% vs. 8.8%), and no SAEs occurred.

| Risk of bias assessment
The two observational studies were assessed to be of high quality. 29,30 The cohort study 29  The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). b Certainty downgraded by two for imprecision and indirectness due to differences between study populations. c Certainty downgraded by two for indirectness due to differences in interventions and populations. therapy. The cohort study of macrolides therapy 30 received the maximum nine points.
Overall, the five published RCTs were assessed at low risk of bias. 25,28,31,32,34 While in four 28,31,32,34 of these RCTs participants and investigators were not blinded to group allocation and intervention received, it was judged unlikely this would have a significant effect on assessment of mortality. Complete assessment of the risk of bias was not possible for the four unpublished studies 33,35 ; however, all were reported as investigator and participant blinded placebo-controlled RCTs and no bias concerns were evident from the available data.

| GRADE assessment
GRADE assessments of the certainty of evidence were conducted based on summary effect data from RCTs only. Certainty of evidence was graded as low for passive immune therapy, due to imprecision and uncertainty surrounding the applicability of evidence in individuals with severe influenza at highest risk of mortality and imprecision. The two largest clinical trials (Davey et al, 26 Beigel et al 27 ) were not powered to detect mortality difference. Certainty of evidence for other adjunctive therapies was also graded as low reflecting variability in interventions and/or imprecision (Table 2).

| Ongoing studies
One RCT of passive immune therapy, which has been reported as completed, was identified from the screen of clinical trial databases, but results are not expected till 2020 (Appendix ). One additional RCT of the macrolide clarithromycin was also identified but at the time of review had not yet been initiated.

| D ISCUSS I ON
The systematic review identified eleven completed studies of noncorticosteroid immunomodulatory therapies, including some highquality RCTs. The results of these studies indicate there is unlikely to be a substantial mortality benefit from passive immune therapy as an adjunct to conventional antiviral therapy for the treatment of severe seasonal influenza. Currently, there is insufficient evidence to recommend routine administration of any of the reviewed adjunctive therapies; however, the results from study of macrolides and NSAIDs warrant further study in well-designed RCTs.
The highest quality evidence uncovered through this systematic review was for the effect of passive immune therapy on severe influenza mortality. Our pooled estimate after including two recently completed and relatively large RCTs indicated it is unlikely to be of benefit in reducing mortality in most situations. This conclusion is different from previous reviews of passive immune therapy, but exclusion of non-randomised and uncontrolled historical studies conducted during the 1918 pandemic suggests our summary effect estimate is more likely relevant to treatment of influenza today.
One further study of passive immune therapy is expected to report results within the next year. This is a phase 2 dose-ranging study with pharmacologic and safety primary endpoints, which recruited 65 individuals, and its results are unlikely to substantially alter the conclusions of this review.
There was statistically significant evidence of clinical benefit from passive immune therapy in the post hoc meta-analysis of The immunomodulatory effects of statins have been well described, and this may be the mechanism of action for some of their observed cardiovascular benefits. 40 The immune modulation in longterm use has been associated with reduced influenza vaccine effectiveness 41 though not consistently. 42 Conversely, influenza infection in people who are receiving statins has been reported to be at lower risk of hospitalisation or death in a large observational study. 43 The RCT by Chase 2019 35 suggested some benefit on clinical symptoms with initiation of statin therapy but was not powered to detect any mortality benefit. Of note, an RCT of rosuvastatin for sepsis-associated ARDS was halted for futility and may have contributed to hepatic and renal dysfunction. 44  with potentially eligible studies for inclusion, but where we were unable to locate full-text copies of articles. We also included data from a number of unpublished studies, and it is possible published results may change-though this is unlikely for mortality.
The decision to exclude observational studies, which did not perform multivariate adjustment for confounding, is not likely to alter the overall conclusions of our review, as the majority of excluded studies that were included in other systematic reviews were of small size. Incorporating these studies into the review would ideally require an individual patient data meta-analysis.
The influenza treatment landscape may also change significantly over the coming years, if any of the antivirals currently in clinical development demonstrate superiority to current standard of care with NAI. If so, a re-assessment of the role of immunomodulatory therapy will be required.