Evaluation of Oseltamivir Used to Prevent Hospitalization in Outpatients With Influenza A Systematic Review and Meta-analysis

IMPORTANCE Despite widespread use, summary evidence from prior meta-analyses has contradictory conclusions regarding whether oseltamivir decreases the risk of hospitalization when given to outpatients. Several large investigator-initiated randomized clinical trials have not yet been meta-analyzed. OBJECTIVE To assess the efficacy and safety of oseltamivir in preventing hospitalization among influenza-infected adult and adolescent outpatients.


B
efore the COVID-19 pandemic, influenza was one of the most clinically burdensome respiratory viruses. 1 The US Centers for Disease Control and Prevention estimated 29 million cases, 380 000 hospitalizations, and 28 000 deaths from influenza in the US during the 2018 to 2019 season. 2 While COVID-19 led to a temporary reduction in infections, influenza is now expected to have a resurgence. 3Novel strains or a rise in a relatively less immune population could trigger an influenza pandemic resembling the crises experienced in 1968 or 2009. 4As such, the availability of safe and effective treatments is critical to avoid overwhelming health care systems and to reduce morbidity and mortality.Indeed, a breakthrough in the COVID-19 pandemic occurred when outpatient randomized clinical trials demonstrated reductions in hospitalization and death. 5In contrast, despite the substantial threat that influenza poses, there are no evidence-based outpatient treatments proven to prevent the progression to hospitalization.
Oseltamivir (Tamiflu) is an antiviral that is commonly prescribed to outpatients with influenza to accelerate recovery and prevent complications.Detailing by key opinion leaders, guideline panels, and the manufacturer has even led to stockpiling of the medication as part of national pandemic responses. 6Yet, despite guideline recommendations, 7,8 and millions of doses prescribed, it is unclear whether oseltamivir reduces severe complications requiring hospitalization.3][14][15] We, therefore, sought to clarify whether oseltamivir is a highvalue medical treatment (achieving optimal results for patients balanced with an efficient use of resources). 16To do so, we conducted a systematic review and meta-analysis of RCTs of oseltamivir for the prevention of first hospitalization in adolescent and adult outpatients (efficacy) and treatmentassociated adverse events (safety).

Methods
The protocol for this systematic review and meta-analysis was prospectively registered on PROSPERO (CRD42022299030). 17indings are reported following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline. 18

Search Methods
We searched PubMed, Ovid MEDLINE, Embase, Europe PubMed Central, Web of Science, and Cochrane Central, as well as ClinicalTrials.govand WHO International Clinical Trials Registry (eTable in Supplement 1) from inception to January 4, 2022.The central search strategy consisted of MeSH (Medical Subject Headings) terms and keywords corresponding to the subjects of "influenza," "randomized clinical trials," and "oseltamivir."This was then adapted to meet each databasespecific or registry-specific terminology.Bibliographies of included articles and relevant systematic reviews were hand searched.Unpublished Roche-sponsored clinical study reports (CSRs) were obtained from the British Medical Journal's open database for oseltamivir. 19No language restrictions, filters, or limits were applied.

Selection Criteria
Published and unpublished RCTs were included in this review.Observational studies were excluded.Each included trial compared oseltamivir at the recommended oral dosage of 75 mg twice daily for 5 days vs a nonactive control equivalent (placebo or standard of care) and reported the outcome of hospitalization.Only the first hospitalization was considered; readmissions were not counted.Study populations included outpatients aged 12 years and older diagnosed with natural influenza infections based on clinical history and laboratory evidence.Most often infection was confirmed by viral culture or polymerase chain reaction (PCR); however, in certain Rochesponsored studies, an infection could also be established by a 4-fold rise in antibody titers at day 30.

Study Selection and Extraction
Search results were imported to EndNote, version 9.3.3(Clarivate), and duplicates were removed.Unique studies were uploaded to Rayyan, and 2 reviewers (R.H. & É.B.C.) independently screened all titles and abstracts, removed clearly irrelevant results, selected eligible studies from full-text review, and recorded reasons for exclusion.The same reviewers then independently extracted the data from included studies using a pre-established data extraction table in Microsoft Excel, version 16.54 (Microsoft).Disagreements were resolved by consensus with a third reviewer (T.C.L.).

Outcomes and Data Items
Extracted study characteristics included the year, number of participants, method of confirming influenza, follow-up duration, and study sponsor.Relevant participant demographics were extracted (eg, race and ethnicity, sex, influenza A or B).Missing study demographics were assumed to be unavailable.
The primary efficacy outcome was the number of first allcause hospitalizations per treatment group in the intentionto-treat infected (ITTi) population-individuals confirmed to have influenza according to the study definition.Hospitalization was defined as the first admission to a hospital or health care center during the treatment or follow-up period, for any cause and any duration.Emergency department visits with direct discharge home were excluded.When this was not specifically reported in the ITTi population, we made up to 8 email data requests to the senior and/or corresponding author.The British Medical Journal database contains unpublished Roche CSRs. 19Within these, we identified hospitalized patients from the serious adverse event narratives and cross-referenced their participant identification numbers to the study's diagnostic results to confirm their case positivity.
The primary safety outcome was the rate of any adverse event, regardless of grade, and included nausea, vomiting, diarrhea, cardiac, psychiatric, neurologic, and a composite of any gastrointestinal symptoms (eg, nausea, diarrhea, gastritis, and others).Nonindustry studies either had thresholds for reporting neurological adverse effects or did not report these at all; for example, Roche CSRs recorded neurological adverse effects (severe and nonsevere) but excluded headache and fatigue if these occurred during the 5 days of treatment and were accompanied by 1 or more additional typical symptoms of influenza (eg, myalgias).In addition, total serious adverse events (as defined by the studies) were analyzed separately whenever possible based on reporting.Adverse events were measured within the safety population (all randomized patients who received at least 1 treatment dose).

Risk of Bias
The risk of bias for each study was assessed by 2 independent reviewers (R.H. & É.B.C.) using the Cochrane Risk of Bias tool, version 2.0. 20Disagreements were settled by consensus, and assessments were rendered by the Risk-of-Bias Visualization tool. 21

Statistical Analyses
Under the Cochrane Handbook's assumption that some heterogeneity is inevitable, a restricted maximum-likelihood random effects model was used for the meta-analyses using the meta-command in Stata, version 17.0 (StataCorp). 22Hospitalization was summarized as a risk ratio (RR) with 95% CIs.A continuity correction of 0.5 was used for cells with zero events.Using the metaprop_one module, we estimated the pooled control event rate using a generalized linear mixed model.We multiplied the pooled control event rate by (1 − risk ratio) and its 95% CIs to estimate the absolute risk difference (RD) with 95% CIs.Common adverse event types were meta-analyzed using a restricted maximum-likelihood random effects model and were reported on the RR and RD scales.If statistically significant, the number needed to harm was reported. 22Statistical heterogeneity was examined with the I 2 test whereby a value greater than 50% was considered statistically significant heterogeneity. 23everal secondary analyses were conducted for the outcome of hospitalization on the RR scale.First, to explore potential causes of heterogeneity, prespecified subgroup analyses were performed based on each study's: mean population age (above and below 65 years); method of confirming influenza (PCR, viral culture, or rapid antigen); population risk level (high-risk [mean population age older than 65 years or high-risk comorbidities] vs not); and trial sponsor (Roche vs other).A subgroup analysis was also conducted based on study quality with studies grouped as either low or at greater than low risk of bias.Finally, for hospitalization, we performed a remove-one metaanalysis to ensure no singular study significantly influenced the pooled estimate and a cumulative meta-analysis to investigate for a change in efficacy over time.
Finally, since Roche studies confirmed influenza infections via viral culture as well as a 4-fold or greater increase in antibody titer, and given prior studies have found oseltamivir reduces the odds of a 4-fold antibody rise by almost 20%, we conducted a post hoc analysis using the ITT populations from Roche-sponsored studies. 24o assess for publication bias, we visually inspected a funnel plot and performed an Egger test for asymmetry. 23A threshold of P < .10 was selected as an indicator of statistically significant publication bias. 25

Certainty of Evidence
Two independent reviewers (R.H. and É.B.C.) evaluated the certainty of the evidence for the outcome of hospitalization using the GRADE (Grading of Recommendations, Assessment, Development and Evaluations) framework. 26

Study and Population Characteristics
The ITTi population comprised 6295 individuals, 3443 of whom were assigned to oseltamivir (54.7%).Overall trial demographics are included in Table 1 based on the total study populations.Participants had a mean (SD) age of 45.3 (14.5) years, and 53.6% (5610 of 10 471) were female individuals.Where reported, 70.2% (4225 of 6017) of participants identified as White individuals, and 20.8% (1253 of 6017) identified as Asian individuals.A total of 60.3% (3668 of 6079) of participants were infected with influenza A. At the study level, 9 of 15 (60%) trials were sponsored by Roche and were conducted between 1998 and 2006.Across studies, the control rate of hospitalization was low (0.6%).

Efficacy Outcome
Overall, oseltamivir was not associated with reduced risk of first hospitalization in the ITTi population (RR, 0.77; 95% CI,

Sensitivity Analyses
A remove-one analysis found Butler, 13 2020, and Roche WV15819, WV15876, and WV15978, 35 2000, had a greater association with the overall effect size (eFigure 3 in Supplement 1).Cumulatively, estimated efficacy decreased over time, particularly when nonindustry studies began to dominate the literature.(eFigure 4 in Supplement 1).A post hoc analysis restricted to placebo-controlled trials also found no difference in the efficacy of oseltamivir (RR, 0.66; 95% CI, 0.38-1.13).The sensitivity analysis using the ITT populations for Roche-sponsored studies shifted the overall effect size toward the null (RR, 0.85; 95% CI, 0.55-1.30;eFigure 5 in Supplement 1).Similarly, when analyzed as a subgroup, Rochesponsored studies also shifted closer to the null (RR, 0.68; 95% CI, 0.41-1.15).Additional subgroup analyses revealed no appreciable changes in the point estimates (eFigures 6A-6E in Supplement 1).

Publication Bias
Visual inspection of the funnel plot revealed asymmetry; however, the Egger test was not statistically significant (P = .66;eFigure 7 in Supplement 1).

GRADE Certainty of Evidence
It was concluded with moderate-certainty evidence that oseltamivir had little to no effect on hospitalization.Although all included studies were RCTs directly evaluating oseltamivir, there was imprecision in the estimates due to wide variability between study results, not all studies were placebocontrolled, and some studies were at risk of bias.Although the present analysis stratified by risk of bias produced similar estimates, these aforementioned factors decreased the present study's certainty from strong to moderate.

Discussion
This systematic review and meta-analysis of oseltamivir for the outpatient treatment of laboratory-confirmed influenza included approximately 3400 more patients than prior analyses. 9,11  fied high-risk subgroups.Interestingly, the subgroup analysis limited to industry-sponsored studies did suggest a reduced risk of hospitalization in the ITTi population.One possible explanation is that the industry studies used viral culture, and it is possible that the PCR used in modern trials detects milder cases with lower viral loads and/or residual nonviable virus when compared with viral culture.Another possibility includes Roche's allowance of a 4-fold rise in antibody responses to confirm infection.There is evidence that oseltamivir may reduce seroconversion and therefore patients hospitalized with negative serology in the oseltamivir group may have been misclassified as noninfected. 24A final explanation could be the lower prevalence of oseltamivir resistance when industry studies were conducted.Since that time, a greater than 10-fold rise in resistance has been observed (0.32% in the early 2000s to 3.56% between 2008 and 2013). 38,39Nonetheless, many of these industry trials only came to light after a legal challenge, and it is reasonable to look at the evidence in total.It was reassuring that there was no increase in severe adverse events observed despite oseltamivir being strongly associated with an increased risk of gastrointestinal adverse effects (nausea and vomiting).
Based on the present analyses, it appeared unlikely that administration of oseltamivir to a general outpatient population had a meaningful effect on serious influenza-related outcomes culminating in hospitalization.That said, it should be noted that the rate of hospitalization was exceedingly low, with a control event rate of 0.6% (95% CI, 0.14%-1.07%).For oseltamivir to continue to be part of a viable influenza response with respect to preventing severe complications, future studies should focus on identifying the groups of higher-risk participants, with laboratoryconfirmed influenza, who may derive benefit.Conducting an adequately powered trial would require a large sample size; however, given millions have received oseltamivir, such a trial doesn't seem unreasonable.As examples, we modeled 2 possible scenarios.First, if the risk of hospitalization is very low (eg, approximately a 1% rate as observed in the general population), 40 a study of 30 716 participants would be required to demonstrate a 30% relative risk reduction with 80% power and 2-sided α = .05.By comparison, to conduct a trial focused on patients at greater risk of hospitalization, (eg, the 2% event rate among this population in the present analysis), 15 232 participants would be required.To succeed at recruitment, such trials would need to either take place during an epidemic or pandemic year or over several years of seasonal influenza.Though the required sample size is large, it is potentially achievable; PANORAMIC (Platform Adaptive Trial of Novel Antivirals for Early Treatment of COVID-19 in the Community) recruited 25 783 participants for early outpatient COVID-19 treatments between December 8, 2021, and April 27, 2022. 41

Limitations
The present meta-analysis had several limitations.First, we analyzed CSRs together with published and nonindustry trials; these differed in the time frame over which they took place, the mechanism for diagnosing infection, and the granularity of the data included.Second, the mean age of the patients was young (mid-40s) and the rate of hospitalization was low.This might have limited the power to detect an effect but also implies that any missed effect would have a very high number needed to treat.We also chose a priori to analyze first hospitalization, whereas others have included readmissions. 24This difference, and the inclusion of the newer trials, should be factored in when comparing these results with prior analyses.Similarly, we excluded patients assigned to high-dose oseltamivir as this is not the approved dosing and therefore is less clinically relevant.Third, although the present search methods were robust, there is always the possibility that some studies, particularly unpublished ones, were missed.Fourth, we did not study symptomatic improvement, which, along with the associated outcome of return to work, could be important during a pandemic.Prior studies have reported small improvements in symptom duration (16.8-25.2hours). 9,11hether this decrease is meaningful when compared with medication costs, an increase in nonsevere adverse events, and the opportunity cost of missing out on the discovery of more effective therapies is a topic of study for health care economists and could be discussed on an individual patient basis.Finally, we excluded observational data from the present analysis.While observational data can contribute larger numbers of patients, and the data can be more affordable and faster to access, it is also subject to substantial biases (eg, immortal time bias, confounding by indication, residual confounding), which make it unsuitable for evaluating medication efficacy even with the most robust statistical methods. 42o our knowledge, this is the first systematic review and meta-analysis focusing on oseltamivir specifically for the reduction of all-cause hospitalization, an important outcome to prevent overwhelming health care systems.Given many of the adverse effects of oseltamivir can overlap with symptoms of influenza, we thought it important to study all-cause hospitalization, rather than influenza-related hospitalizations, which would overlook complications related to the medication's adverse effects.This is particularly relevant for older high-risk adults where seemingly mild gastrointestinal adverse effects might still increase the risk of hospitalization through anorexia and dehydration.

Conclusions
Based on the available RCT data in this systematic review and meta-analysis, there is a lack of convincing evidence that oseltamivir reduces serious complications in outpatients with influenza, though its use is associated with an increase in nonsevere gastrointestinal adverse events.This meta-analysis provides important data for clinicians, patients, and policy makers to contextualize the evidence and inform guidelines.Future research should focus on the conduct of an adequately powered placebo-controlled trial in a suitably highrisk population.

Table 2
summarizes the risk of key adverse events.Patients prescribed oseltamivir experienced significantly more nausea (RR, 1.43; 95% CI, 1.13-1.82),vomiting (RR, 1.83; 95% CI, 1.28- This PRISMA flow diagram was for a systematic review that included searches of databases, registers, and other sources.All records identified by other methods included CSRs provided by Roche to the British Medical Journal.Abbreviation: CSR, clinical study report.

Table 1 .
Study Design and Population Characteristics (continued) used serological testing alongside viral culture to confirm influenza diagnoses.Overall trial demographics were based on the total study populations.Demographics for the desired 12 years and older population extracted from Fry 27 Abbreviations: BMI, body mass index; ITTi, intention-to-treat infected; PCR, polymerase chain reaction; RT-PCR, reverse transcription polymerase chain reaction; NR, not reported.aIn this review, "other" race and ethnicity is defined as any other race or ethnicity other than White, Black, or Asian (ie, Hispanic, American Indian, or Alaska Native) as well as those participants whose identity was not provided or collected (ie, not available).bDenotes studies that Figure 2. Random Effects Meta-analysis on the Outcome of Hospitalization Within the ITTi Population Aged 12 Years and Older Despite this, oseltamivir was not associated with significantly reduced hospitalization in general or in prespeci-"Yes" indicates the number of individuals hospitalized, and "No" indicates the number of individuals who were not.Statistical heterogeneity was examined with the I 2 test whereby a value greater than 50% was considered statistically significant heterogeneity.