Performance of surveillance case definitions for respiratory syncytial virus infections through the sentinel influenza surveillance system, Portugal, 2010 to 2018

Background Well-established influenza surveillance systems (ISS) can be used for respiratory syncytial virus (RSV) surveillance. In Portugal, RSV cases are detected through the ISS using the European Union (EU) influenza-like illness (ILI) case definition. Aim To investigate clinical predictors for RSV infection and how three case definitions (EU ILI, a modified EU acute respiratory infection, and one respiratory symptom) performed in detecting RSV infections in Portugal. Methods This observational retrospective study used epidemiological and laboratory surveillance data (October 2010–May 2018). Associations between clinical characteristics and RSV detection were analysed using logistic regression. Accuracy of case definitions was assessed through sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). A 0.05 significance level was accepted. Results The study involved 6,523 persons, including 190 (2.9%) RSV cases. Among 183 cases with age information, RSV infection was significantly more frequent among individuals < 5 years (n = 23; 12.6%) and ≥ 65 years (n = 45; 24.6%) compared with other age groups (p < 0.0001). Cough (odds ratio (OR): 2.4; 95% confidence interval (CI): 1.2–6.5) was the best RSV-infection predictor considering all age groups, while shortness of breath was particularly associated with RSV-positivity among ≤ 14 year olds (OR: 6.7; 95% CI: 2.6–17.4 for 0–4 year olds and OR: 6.7; 95% CI: 1.5–28.8 for 5–14 year olds). Systemic symptoms were significantly associated with RSV-negative and influenza-positive cases. None of the case definitions were suitable to detect RSV infections (AUC = 0.51). Conclusion To avoid underestimating the RSV disease burden, RSV surveillance within the Portuguese sentinel ISS would require a more sensitive case definition than ILI and, even a different case definition according to age.


Introduction
The human respiratory syncytial virus (RSV) is a major cause of morbidity and mortality worldwide since it is the predominant viral agent affecting the respiratory tract, causing acute, sometimes fatal lower respiratory tract infections in infants, young children and the elderly [1]. In fact, RSV has been associated with a substantial disease burden in adults, especially in the elderly, with an estimation of 1.5 million episodes of acute respiratory infection (ARI) in industrialised countries in 2015 [2][3][4]. Moreover, according to a study in the same year, the global burden of RSV-associated acute lower respiratory infection has been estimated at 33.1 million annually resulting in over 3.2 million severe illness that required hospitalisation in children younger than 5 years [5].
Currently, the available options for clinical management of RSV disease are symptomatic supportive care [6] as well as Palivizumab. Palivizumab is a humanised antibody against the F glycoprotein of the virus. It prevents RSV infection and has been shown to reduce the number of hospitalised cases by half [7]. It can also be employed to treat RSV but it does not reduce RSV mortality and its use is limited to selected populations in high-resource settings [6,[8][9][10]. Several RSV vaccines are progressing in phase III clinical trials and RSV vaccines are expected to become available in the coming 5 to 10 years. With this perspective, evidence-based support for vaccination policies at the national, regional and global levels is necessary and, in 2015, the World Health Organization (WHO) made it a high priority to establish robust age-specific estimates of those affected by RSV and globally-compatible RSV-diseaseburden surveillance systems [11]. Many countries however detect RSV infections within existing surveillance systems for influenza [12][13]. Moreover, one of the challenges to implement a global RSV surveillance system is the lack of a uniform case definition. Influenza case definitions may be less sensitive for RSV and, consequently, have the potential to underestimate the RSV burden [14]. Indeed, a broader ARI case definition, which includes a sudden onset of symptoms and at least one respiratory symptom (cough, sore throat, shortness of breath or coryza), has been considered to be more suitable for capturing RSV infections [1,15].
In Portugal, RSV cases are detected using the standard European Union (EU) influenza-like illness (ILI) case definition through the influenza surveillance system (ISS). The EU ILI case definition includes sudden onset of symptoms, at least one respiratory symptom (cough, sore throat, shortness of breath), and at least one systemic symptom (fever or feverishness, malaise, headache or myalgia) [15]. Therefore, we aimed to evaluate signs and symptoms as clinical predictors of RSV, and to estimate the sensitivity and specificity of three case definitions, including EU ILI, a modified EU ARI [15], and one respiratory symptom, for detecting RSV infections through the country's sentinel ISS.

Portuguese Influenza Surveillance System
The Portuguese ISS comprises a sentinel and a nonsentinel component. The sentinel component, which is the sentinel ISS, exists since 1990 and is composed by the General Practitioners' (GP) Sentinel Network, GP from the I-MOVE's Euro EVA project (which is a project to monitor the effectiveness of influenza vaccine in the EU at a community-level) and the Emergency and Obstetric Departments Networks. The sentinel ISS includes primary healthcare centres, general hospitals with paediatric and adult emergency rooms and medical wards, and one reference paediatric hospital in Lisbon. Age and sex distribution of the population under observation in the GP Sentinel Network is similar to that of the Portuguese population [16]. Using a data reporting form, the sentinel ISS weekly reports demographic, clinical and laboratory data of individuals tested for influenza and other respiratory viruses to the National Reference Laboratory for Influenza and other Respiratory Viruses at the National Institute of Health Doutor Ricardo Jorge (INSA) during the influenza season. The influenza season comprises the period between week 40 (October) and week 20 (May) of the next year. In addition, the sentinel ISS sends nasopharyngeal swabs for influenza and other respiratory viruses diagnosis to INSA.

Study population and data
Individuals tested through the sentinel ISS from 2010 to 2018 in Portugal for any respiratory virus (adenovirus, enterovirus D68, human coronavirus, human metapneumovirus, parainfluenza (1, 2 and 3), rhinovirus and RSV A and B) with epidemiological, clinical and laboratory data were included in the study. Data were collected based on the detection of influenza cases through the sentinel ISS using the EU ILI case definition. Clinical characteristics were interpreted by the doctors or reported by the patients (or their parents if

Ethical statement
Data used within this study were anonymised and were collected: (i) in the scope of epidemiological surveillance for which submission to an ethical committee is not required and (ii) specific projects including GP Sentinel Network and I-MOVE's Euro EVA, which had already been approved by the Health Ethic Committee of INSA.

Case definitions tested
The EU ILI case definition is used in the sentinel ISS. However, not all individuals fulfilled its criteria. Therefore, in order to validate the accuracy of RSV detection through the ISS, we decided to test three different case definitions (Table 1): (i) EU ILI, (ii) a modified EU case definition for ARI (ARI-like), and (iii) only one respiratory symptom of the three included in ARIlike/ILI case definitions. ARI-like and only one respiratory symptom case definitions were created based on the clinical characteristics that were reported by the sentinel ISS.

Data analysis
Retrospective analyses were performed on data from individuals included in the study. Demographic characteristics were described using percentages. Associations were assessed using the chi-squared test. For the sudden onset of symptoms, each symptom and the three case definitions, we calculated the odds ratios (OR) and 95% confidence intervals (CI) to evaluate their association with laboratory-confirmed RSV by bivariate logistic regression. Moreover, we calculated the relative risk ratio (RRR) and 95% CI for laboratory-confirmed RSV and influenza using multinomial analysis for three groups: (i) RSV-positive cases, (ii) influenza-positive cases and (iii) a reference group that included negative cases for any respiratory virus. Case definition performance characteristics, including sensitivity, specificity, and area under the receiver operator characteristic curve (AUC), were evaluated according to the age group [18]. Tests with p value < 0.05 were considered statistically significant. The data analysis was performed using STATA 12.

Selection of study patients and demographic characteristics
During the influenza seasons between 2010 and 2018, 7,085 individuals were tested for any respiratory viral agent ( Figure). Of these, 562 were excluded from this study due to missing clinical characteristics. Remarkably, myalgia and malaise were missing in children aged ≤ 14 years. were significantly associated (p < 0.0001) with RSV-positive cases whereas fever or feverishness (RRR: 0.6; 95% CI: 0.4-0.8) and myalgia (RRR: 0.5; 95% CI: 0.3-0.7) were negatively associated (Table 4). With the exception of malaise and sore throat, which were not significant, and shortness of breath (RRR: 0.7; 95% CI: 0.6-0.8), all clinical characteristics, including the EU ILI, the ARI-like and the one respiratory symptom case definitions, were significantly associated with influenza-positive cases.

Performance of the influenza-like illness, acute respiratory infection-like and one respiratory symptom case definitions
When testing the performance of the case definitions, one respiratory symptom had the highest sensitivity for detection of RSV with 99.5%, whereas the ILI and ARI-like case definitions revealed the highest specificities with 21.1% and 20.8%, respectively (Table 5). AUC was 0.51 in all case definitions.
Concerning the analysis by age group, for children < 5 years old the ILI and ARI-like case definitions presented the lowest sensitivity (56.5%) and the highest specificity (27.7% and 26.4%, respectively). Only one respiratory symptom as case definition showed the highest sensitivity (100%) in all age groups, with the exception of the 5-14 years' age group. This definition had the lowest specificity with a range from 2.5% to 3.3% depending on the age group. The ILI and ARIlike case definitions presented the highest AUC in the 15-64 years' age group (0.56 for both definitions) and the lowest in the children < 5 years old (0.42 for ILI and 0.41 for ARI-like).

Discussion
Although many studies regarding RSV have focused on children aged under 5 years and elderly adults, little is known regarding RSV among patients of all ages. This study evaluated the clinical characteristics and the performance of three different case definitions for the diagnosis of RSV infection based on epidemiological and laboratory data from 2010 to 2018 in Portugal.
In comparison with other age groups, laboratory-confirmed respiratory infection frequency in children aged  Table. b Cases with missing age or sex were excluded from the analyses by age or sex respectively. Age was missing for seven RSV-positive and 89 RSV-negative cases; sex was missing for two RSV-positive and 19 RSV-negative cases. c Observed frequencies were higher than expected among these age groups: 0-4 and ≥ 65 years (p < 0.0001).

Table 3
Clinical characteristics associated with laboratory-confirmed respiratory syncytial virus cases detected through the sentinel influenza surveillance system using bivariate logistic regression, Portugal   ≤ 14 years was significantly higher than what would be expected if respiratory virus infection and age were independent. Moreover, RSV infection was significantly associated with two age groups, 0-5 years and ≥ 65 years, which is consistent with RSV infection being considered as one of the most common causes of respiratory tract infection during childhood as well as among elderly adults or adults with underlying medical conditions [19].
The clinical characteristics associated with RSVpositive cases varied across age groups, which suggested that a different case definition according to age might be suitable. Among respiratory symptoms, shortness of breath was significantly associated with being RSV-positive among children aged ≤ 14 years, whereas cough was most common (94.7%) and significantly associated with RSV infection overall. Nonetheless, no significant difference in the frequency of cough, either positive or negative, was found between age groups when we stratified by age, which differed from previous studies that found cough associated with RSV infection in children under 5 years old [12,[20][21][22].
Among the systemic symptoms, fever has always been controversial as to its inclusion in the case definition for RSV infection [23]. In our study, fever or feverishness was very frequently observed among RSV-positive cases ≤ 14 years old (95.7-100%). However, no significant difference between this age group and others, either positive or negative, was found, which was in agreement with Nyawanda et al. [12] but not with other studies [14,20,21]. Moreover, among the 15-64 and ≥ 65 year olds, the observed frequencies (53.3-65.4%) of this symptom were similar to previous studies conducted in adults [24,25] and negatively associated with RSV-positive cases.
Overall in the current study, good clinical predictors for laboratory-confirmed RSV infection were respiratory symptoms including cough and shortness of breath whereas systemic symptoms were negatively associated. These findings were also in agreement with the results from the multinomial analysis. In this analysis, taking respectively as a reference persons testing negative for any respiratory virus, several considerable differences were found in the presentation of influenza and RSV, with all clinical characteristics considered in the EU ILI case definition significantly associated with influenza-positive cases with the exception of malaise, sore throat and shortness of breath. Therefore, it may be easier to diagnose influenza than RSV infection based on the considered clinical characteristics and case definitions in this study.
Much of what is known about RSV globally comes from surveillance systems that use case definitions intended for influenza, which are not optimised for RSV detection. In our study, we demonstrated that the EU ILI case definition, which is the one used in the current Portuguese sentinel ISS, was not accurate for RSV detection, and in fact, it was only a good clinical predictor in the 15-64 years' age group. Furthermore, none of the evaluated three case definitions was able to discriminate among RSV-positive and negative cases with sufficiently high sensitivity and specificity to be clinically reliable because all these definitions presented an AUC of 0.51. However, the optimal case definition should be determined considering the epidemiological context, available resources, study population and the objective for which it is being used. Only one respiratory symptom showed the highest sensitivity with 99.5%, which may be very good to better understand burden, seasonality and mortality associated with RSV infection. In fact, a very high proportion of hospitalisations are due to respiratory symptoms [21]. For vaccine effectiveness studies, a highly specific case definition would be more suitable [26,27] and the ILI and ARI-like case definitions showed higher specificities (21.1% and 20.8%, respectively) than only one respiratory symptom, albeit they were low in comparison with previous studies in hospitalised children [12].
In interpreting our results, the following limitations should be taken into consideration. Firstly, missing symptoms of myalgia and malaise were mostly among individuals aged under 14 years. Secondly, the number of persons of this age group was very low in comparison with other age groups, with only 21 children under 1 year old included in the study, thus not allowing further age-stratified analyses in the 0−14 year-old age group. Thirdly, the analysis did not differentiate between inpatients and outpatients and, indeed, most collected information was from outpatients, whose underlying risk factors were unknown. Finally, data were collected based on EU ILI case definition and thereby, individuals presenting only one respiratory symptom might have not been selected to test for any respiratory virus resulting in a selection bias. Therefore, these findings may not be representative for inpatient respiratory disease surveillance and the most severe spectrum of cases with RSV infection. However, we believe that this study may contribute to decide the best option for RSV surveillance in Portugal, especially at a community level, and may help support public health strategies and interventions at a European and global level regarding the prophylaxis and treatment options.
In Portugal, the most feasible option would be to include RSV surveillance within the existing sentinel ISS. We demonstrated that the current EU ILI case definition was not suitable to detect RSV infections and indeed, a more sensitive case definition to avoid the underestimation of RSV burden disease and a different case definition according to age would be more appropriate. Moreover, hospital-based surveillance, especially with a focus on paediatric patients, including paediatric intensive care units, should be specifically enhanced for RSV.