Clinical characteristics and severity of influenza infections by virus type, subtype, and lineage: A systematic literature review

Aim Studies carried out in the early 2000s found that the number of influenza‐associated hospitalizations and deaths was highest in seasons dominated by A(H3N2), suggesting that the clinical presentation and severity of influenza may differ across virus types, subtypes, and lineages. We aimed to review the studies that examined this hypothesis. Method We conducted a literature review of studies published until January 2017 that compared the clinical presentation, disease severity, and case‐fatality ratio of influenza patients infected with different virus types (A, B), subtypes (pre‐pandemic A(H1N1), A(H1N1)p, A(H3N2)), and lineages (Victoria, Yamagata). Results The literature search resulted in over 1700 entries: After applying in‐ and exclusion criteria, 47 studies were included in the literature review. Studies showed a wide diversity in setting and populations. Only a minority of studies provided results adjusted by patient's age and other potential confounders. There were very few differences in the clinical presentation of patients infected with different influenza viruses. We found weak evidence that the A(H1N1)p subtype in the post‐pandemic period was more often associated with secondary bacterial pneumonia, ICU admission, and death, than the other influenza virus (sub)types. Conclusion Contrary to what is commonly assumed, the causal virus subtype does not seem to be a major determinant of clinical presentation and severity of influenza illness. However, drawing conclusions was made difficult by the low comparability and methodological shortcomings of included studies, and more well‐designed studies are warranted.


| INTRODUC TI ON
Influenza illness is clinically characterized by non-specific signs and symptoms that are common to other respiratory infections, such as sudden onset, fever, malaise, headache, and cough. 1 Influenza illness is usually short-lived (3-5 days), and severe outcomes are rare unless the person is elderly or has an underlying disease (such as chronic heart disease, diabetes, and cancer), a weakened immune system, or other medical condition. Influenza was described as "an unvarying disease caused by a varying virus" in 1975, 2 suggesting that the illness caused by the different virus types and subtypes is clinically indistinguishable, but this has been challenged in recent years. Two ground-breaking studies published by Thompson et al in 2003 and found that the number of hospitalizations and influenza-associated deaths in the United States was highest during seasons in which A(H3N2) was the dominant subtype among the circulating viruses, followed by seasons in which influenza B or influenza A(H1N1) was dominant, and this was confirmed in later studies. [3][4][5] Although these studies were not based on individual-level clinical data but modeled data with aggregated national mortality, hospital discharge, and viral surveillance data, they have led to the hypothesis that the clinical presentation, severity, and risk of unfavorable outcomes of influenza illness may indeed differ across virus types and subtypes.
In recent years, the hypothesis that influenza severity is dependent on the causal virus type and subtype has been examined in several studies, [6][7][8][9][10][11][12] which differed considerably between one another in terms of study setting and design, populations being examined, sample size, influenza viruses being compared, and ability to control for potential confounders (eg, patient's age, underlying comorbidities, and other predictors of disease severity and outcome). To our knowledge, no systematic review has been carried out to date that has attempted to summarize the available evidence, yet this question is of considerable importance from both a clinical and public health perspective, as it may have implications for the management of influenza patients, for communication and preparedness during seasonal epidemics (eg, regarding the number of influenza-related hospitalizations to be expected during the influenza season), and for producing accurate cost-benefit estimates of influenza vaccination campaigns and other prevention and control strategies. To help clarify this issue, we conducted a systematic review of published studies that compared the clinical presentation, course severity, and casefatality ratio of influenza patients infected with different virus types, subtypes and lineages.

| Literature search and inclusion criteria
We searched articles in MEDLINE using the following search string: influenza AND (sign(s) OR symptom(s) OR clinical OR comorbidity OR severity OR complication(s) OR death) AND (comparison OR compare/s/d). We considered all papers published until January 31, 2017, that were written in English or in another language mastered by at least one study researcher (ie, French, Spanish, Italian, or Dutch). Two study researchers independently carried out an initial screening of all entries based on their title and abstract: Papers that were considered eligible for the review were obtained and read in full copy text format. In the next step, the eligibility of each paper was independently assessed by two study researchers; any disagreements were resolved via consensus. Papers were considered to be eligible for inclusion if they compared the clinical presentation (signs and symptoms), the presence of underlying conditions, or the disease severity (eg, complications, hospitalization, admission to an intensive care unit [ICU], need for ventilation support, or casefatality ratio) between laboratory-confirmed influenza patients infected with different influenza virus types (A, B), subtypes (prepandemic A(H1N1), A(H1N1)p, A(H3N2)), and lineages (Victoria, Yamagata). We excluded studies in which all included influenza cases were infected with only one influenza virus (sub)type, those focusing on avian influenza viruses, and those that were carried out during the pandemic period (ie, all patients were enrolled between April 2009 and July 2010). The references of all retrieved papers were tracked to find additional publications.

| Data extraction
Data were extracted from each article by one study researcher, entered into a database expressly developed for the project, and independently cross-checked by a second study researcher. In addition to main outcomes, we extracted information on factors that were considered to be relevant for the correct interpretation of the results, namely: 1. Country, region, and years in which the study was conducted; 2. Study setting and criteria for inclusion of laboratory-confirmed influenza patients (eg, patients reported to community-based surveillance system, individuals visiting the emergency room of hospitals and clinics, inpatients), and whether the study was conducted among specific population subgroups (eg, asthma patients, healthcare personnel, pregnant women);

| Assessment of the quality of studies
For observational studies, such as the studies included in our systematic review, several quality assessment tools or grids exist, 13 many of which are, however, specifically developed for studies with a case-control or cohort design. Considering most of the studies included in our review have a cross-sectional design, we opted to score all included studies using a slightly modified version of the Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies developed by the National Heart, Lung and Blood Institute, 14 which is an adequate tool to assess the quality of the studies and the risk of bias.

| Statistical analysis
The main characteristics of all selected studies are reported in Tables 1 and 2. The studies were divided into two groups: studies in which all included influenza cases were treated as inpatients (ie, hospital-based studies) and studies in which only a subset of patients were eventually hospitalized (these included community-based studies, studies in which patients were enrolled among those visiting the emergency room of a hospital, and others). This was done based on the expectation that results may differ when all patients are hospitalized, because these patients may be more severely ill compared to patients from settings that include outpatients or are community-based patients.
The studies differed in the statistical methods that were used to compare the clinical presentation and severity of influenza illness between patients infected with different virus (sub)types.
Some studies presented a measure of relative risk (RR) (ie, odds ratio or risk ratio) calculated through regression models: These were reported in Table 3 (for signs and symptoms) and Table 4 (for underlying conditions, complications, and outcomes), along with the variables that were used for adjusting the RR estimates.
We had initially planned to pool study-specific RRs into a summary estimate using random-effects meta-analysis models; however, this was not possible because of the large diversity of studies in terms of settings, populations, and definitions (see The majority of studies performed no adjustment for the patient's age (although some of them focused on specific age groups such as children, 7,15,16 adults, 6 or the elderly 17 ) or other potential confounders. In these studies, proportions (for binary variables such as the presence/absence of signs and symptoms, underlying conditions, or complications) and mean/median values (for continuous variables such as the length of hospital stay) were reported and frequently compared using appropriate statistical tests. When no test was performed by the authors, we applied a large-sample test to compare proportions, provided that the group-specific sample size and proportions were reported by the study authors. The results of these studies were summarized in Tables S1 and S2 (for signs and symptoms) and Table S3 (for complications, outcomes, and underlying conditions).  Figure 1).

| RE SULTS
An overview of the studies (Table 1 9 The assessment of the quality of included studies is provided in the Data S1. Limitations common to most of the included studies were the following: lack of a sample size justification (or a precise calculation of the statistical power), poor clarity about how the outcome in the study was defined and assessed, and lack of adjustment for potential confounding (see below). Also, the participation rate and proportion of patients lost to follow-up were not reported in many studies.
Only six papers reported odds ratios or risk ratios for differences in the frequency of symptoms and signs (Table 3 10 All study participants were nursing home residents.     p were admitted more often to the ICU and died more often).
The assessment of the unadjusted differences in the frequency of symptoms and signs (Table S1 15,21-28,30,31,36 and Table S2 6-8, [10][11][12]16,31,[37][38][39][40][41]44,[46][47][48][49][50][51]53,54 ) also showed few differences between the influenza viruses. Compared to influenza B, there was some evidence that patients with influenza A (not further specified) less often presented with myalgia (four studies-all focusing on children-of fifteen) were less often sent to the hospital for medical advice and/or further investigation (two studies of fourteen) and more often presented with cough (two studies of nine). With the exception of the finding for myalgia, there were no further age-specific differences in the frequency of symptoms and signs between influenza viruses.
Concerning the frequency of complications and underlying conditions (Table S3 9

| D ISCUSS I ON
We aimed to assess the difference in clinical characteristics and ill- Likewise, the virus subtype did not seem to be a major determinant of severity, especially once the patient's age and pre-existing health conditions were taken into account, with the possible exception for the A(H1N1)p virus subtype.
Knowing the virus type and subtype may help with the clinical management of a patient, and some researchers have stressed the importance of rapid testing tools to identify the type of virus, 40 while, others have suggested that clinical relevance is low. 40 40,43 or bacterial co-infections, 41,44 to promote a more prudent use of antiviral and antibacterial drugs. From a public health perspective, Yap and coll. 43  A number of studies have found that influenza-associated hospitalizations and deaths are highest in seasons dominated by A(H3N2), [3][4][5] suggesting that the clinical presentation and severity of influenza may be worse for this subtype. However, we did not con-  9 Another limitation was that signs and symptoms may vary between mild and severe; therefore, their clinical presentation may not provide a precise measure of the severity of influenza (only a small number of studies made a distinction in the severity of signs and symptoms, for instance, by focusing on "high fever" instead of on fever in general). We did not focus on the age signature of the different influenza viruses in our review: However, some studies suggested that there is a difference between age groups affected by different influenza viruses, 8,10,30,31,33,40,41,44 and reviewing these data could provide additional knowledge. Another limitation of our review may lie in our search strategy. Studies were only searched in MEDLINE, and, although its coverage has been demonstrated to be generally high, 59 some eligible papers were missed in the initial search. Concerning the search string, we used the Boolean operator OR several times to be as sensitive as possible in the earliest steps of the literature search; however, we were also forced to include "influenza" and "compare/d/s/comparison" in order to keep the number of screened entries to within reasonable limits, and some eligible papers may have also been missed because of this approach. The snowballing method revealed a significant number of additional papers and, while this increased the coverage of our search, we cannot rule out the possibility of having missed some studies.
In conclusion, we found very limited evidence that the different influenza virus types, subtypes, and lineages differ between one another in terms of clinical presentations, prevalence of underlying medical conditions, illness severity, or case-fatality ratio. However, an important gap in knowledge still exists in this area, as drawing firm conclusions was made difficult by the low comparability and methodological limitations of many of the studies that were included. A minimum set of quality requirements for future studies on this topic should include a clear description of the study populations, settings, and in-/exclusion criteria; a follow-up of each patient during the entire illness course, that is, from onset until recovery or death (and including details of in-hospital stay for patients that were hospitalized); and the use of multivariate regression techniques providing relative risk estimates adjusted by (at least) patient's age, underlying conditions, vaccine status, and antiviral treatment.

ACK N OWLED G EM ENTS
We would like to thank François Schellevis Joke Korevaar and at NIVEL for supervising and advising this research project.

CO N FLI C T O F I NTE R E S T
Clotilde El Guerche-Séblain is an employee of Sanofi Pasteur.
Clotilde El Guerche-Séblain is the scientific coordinator at Sanofi Pasteur of the research project, helped define the study objectives, and critically revised the manuscript. When reviewing the manuscript, the revisions did not concern the public health findings or conclusions. All the other authors declare they have no conflict of interest to disclose.