Primary care influenza‐like illness surveillance in Ho Chi Minh City, Vietnam 2013‐2015

Background Year‐round transmission of influenza has been detected in Vietnam through both national surveillance and other epidemiological studies. Understanding the demographic and clinical features of influenza‐like illness (ILI) presenting to primary care in urban Vietnam is vital to understand these transmission dynamics. Methods An observational study of patients with ILI in Ho Chi Minh City, Vietnam, was conducted between August 2013 and November 2015 in a mix of public and private primary care settings. Molecular testing for influenza A and influenza B and 12 other respiratory viruses was performed. Results A total of 1152 ILI patients were recruited. 322 and 136 subjects tested positive for influenza A and influenza B, respectively. 193 subjects tested positive for another respiratory virus; most commonly rhinovirus and parainfluenza virus 3. Influenza was detected in 81% of the 116 study weeks. Three peaks of influenza activity were detected; an H3N2 peak April‐June 2014, an influenza B peak July‐December 2014, and a mixed H3N2 and H1N1 peak March‐September 2015. Subjects recruited from private clinics were more likely to have higher income and to have reported previous influenza vaccination. Antibiotic use was common (50.3%) despite limited evidence of bacterial infection. Conclusion Influenza in southern Vietnam has complex transmission dynamics including periods of intense influenza activity of alternating types and subtypes. Broadening surveillance from hospital to the community in tropical settings is feasible and a valuable for improving our understanding of the global spread and evolution of the virus.

Influenza was detected in 81% of the 116 study weeks. Three peaks of influenza activity were detected; an H3N2 peak April-June 2014, an influenza B peak July-December 2014, and a mixed H3N2 and H1N1 peak March-September 2015.
Subjects recruited from private clinics were more likely to have higher income and to have reported previous influenza vaccination. Antibiotic  In temperate countries, predictable winter epidemics occur, but transmission dynamics in tropical settings are more irregular. For many decades, influenza was thought not to be a disease of importance in tropical countries. This has changed considerably in the last 10 years. 3 Research efforts are ongoing to increase knowledge and understanding of how tropical transmission influences global influenza transmission. Vietnam, as one of the most populous countries in the tropics, is a potentially globally important site for influenza dynamics, both for its possible contribution to global virus evolution 4,5 and as a high-risk environment for the emergence of potentially pandemic strains such as highly pathogenic avian influenza subtype H5N1.
Year-round transmission of influenza has been detected in Vietnam through both national surveillance 6,7 and phylogenetic analysis. 8 Simultaneous circulation of multiple influenza types and subtypes has been documented 6,8-11 along with patients having more than one influenza infection in one season. 9 Approximately 15%-20% of patients presenting to hospitals with influenza-like illness (ILI) have virologically confirmed influenza, 6,7 and influenza is associated with up to 14% of community-acquired pneumonia presenting to hospitals. 12 However, complementary data on community ILI incidence are limited. As national surveillance programmes 6,7 have focused largely on hospital sites, estimates from these studies are likely to underestimate true clinical burden and attack rates in the community. Seasonal influenza vaccine is not currently provided through public health services in Vietnam and there is minimal, although increasing, use in the private sector. 13 The optimal vaccination strategy for seasonal influenza in Vietnam is still to be established. 14 The objectives of this study were to describe the dynamics and demographic/clinical characteristics of ILI presenting to primary care services within an urban setting in Vietnam over a 27-month period and to estimate the proportion of ILI caused by influenza A and influenza B.

| Summary of design and conduct of the observational study
This observational study was conducted between August 2013 and November 2015 at outpatient clinics and community medical practitioners in Ho Chi Minh City (HCMC), Vietnam (private clinics were drawn from the an ongoing mHealth surveillance study 11 ). These clinics were considered representative of primary care within Vietnam. 15 Individuals were invited to join the study if they were between 10 and 70 years of age, had symptoms for <72 hours and met the ECDC definition of influenza-like illness. 16 One anterior nasal swab and one throat swab were collected at recruitment and transported in a single tube of viral transport medium to a central laboratory before being stored at −20°C within 24 hours.
Clinical characteristics of the initial ILI episode were reported by subjects to study staff. No inpatient data were collected in the event of hospitalisation. No additional objective information was available from the subjects' usual treating clinician.

| Sample analysis
Samples were stored at −20°C within 24 hours of collection and then stored at −80°C after initial PCR testing. Respiratory samples were batch tested monthly for influenza A and influenza B using standard polymerase chain reaction (PCR) techniques. 17 Samples positive for influenza A were then subtyped as H3N2 or H1N1. Testing for twelve other respiratory viruses was performed on samples which tested negative for influenza using an in-house multiplex PCR assay which has been described previously. 18

| Statistical analyses
The primary outcome was PCR-confirmed influenza A in nasal/ throat swabs. This included both mono-infections with influenza A infections and co-infections with influenza B.
Continuous variables that were normally distributed were compared with t test or ANOVA as appropriate. Tukey's HSD was used for post hoc testing following ANOVA where appropriate. Continuous variables that were non-normally distributed were compared using Mann-Whitney or Kruskal-Wallis tests, depending on the number of groups. Categorical variables were compared using Fisher's exact test, Mann-Whitney or chi-squared tests, as appropriate.
Analysis of overall prevalence of influenza as a cause of ILI was performed on the total study population; weekly prevalence was also calculated. Weekly influenza transmission intensity was categorised by percentage of ILI subjects testing influenza-positive using WHO thresholds 19 as zero (0%), low (1%-10%), medium (11%-20%), high (21%-30%) and very high (>30%). A peak of influenza activity was defined as four or more consecutive weeks where influenza transmission intensity was high or very high. Weeks when no recruitment was performed because of clinic closures were excluded from analysis; weeks during which clinics were open but no patients were recruited were included in the analysis. Interactions between influenza and non-influenza respiratory virus circulation were investigated using Pearson's correlation and cross-correlation with weekly time lags.
Demographic and clinical characteristics were compared between predefined groups based on influenza infection status. The initial analysis was planned to be performed between three groups: influenza A positive, influenza B positive and non-influenza ILI with a subsequent analysis between influenza A subtypes (H1N1 and H3N2) if numbers were sufficient. Where appropriate, analysis was stratified by age. Age was preferentially used as a continuous variable, otherwise age was categorised as per recommendations from the Consortium for the Standardisation of Influenza Seroepidemiology (CONSISE, 5-9, 10-19, 20-44, 45-65, 65+). 20 Comparison between subjects recruited at public and private clinics was also performed.
Logistic regression was used to investigate the effect of (1) household age structure (total number of household contacts, total number of household contacts in each CONSISE age class) and (2)

| Age distribution
Age of recruited study participants was right skewed with a median  Table S2). The comparison was the same using national Vietnamese or HCMC-specific age distribution data (t test, all P values >.4).

| Past medical history
At baseline, there was no significant difference in relevant past medical history (PMH) of individuals presenting with PCRconfirmed influenza A, influenza B or non-influenza ILI (Table S3).
Chronic respiratory disease was uncommon in recruited subjects (2.4%, n = 23/953) but was reported more frequently in individuals aged 45 years or over (6.3% vs 1.8%, 6/95 vs 17/858; χ 2 5.106 P value .0238). Current smoking was reported in 14.9% of the study population, all of whom were male (n = 172/1152). About 36.1% of males over the age of 20 were current smokers (n = 156/432). F I G U R E 3 Age and sex distribution by infecting influenza type. Point and lines represent expected percentage for HCMC with binomial 95% CI. Coloured bars show percentages observed in this study. Expected values were limited to age groups included in the study protocol (10-70 inclusive)

| Clinical symptoms at presentation
Most subjects presented on the second or third day of symptoms (Table 1). Those with influenza B presented later than those with influenza A or those who were influenza negative (ANOVA F 2,1149 = 15.07, P < .001); however, the difference in time to presentation was 0.28 (95% CI 0.14-0.43) of a day (ie approx. 7 hours) which is not clinically important. All subjects had mild disease, with TA B L E 1 Clinical symptoms at presentation <1% of subjects reporting that they could not carry out their normal daily activities (n = 5/1152). Rhinorrhoea and cough were reported more frequently in those infected with influenza. Antibiotic use prior to enrolment in the study was common in all groups (53.0%, n = 611/1152) but was highest in those with influenza B (66.2%, n = 90/136; χ 2 11.6, P value = .003).
Median number of household members (including the study subject) was four (IQR 3-5); however, this was heavily right skewed with the largest household having 17 members. Distribution of household size was similar for all infection groups (Kruskal-Wallis Pvalue = .2106) but different to the national urban average with fewer one-person households in the study and more households with five or more occupants (Table S5)

| D ISCUSS I ON
Influenza surveillance in Vietnam has previously centred on hospitalised patients. This observational study was designed to investigate the incidence and associated clinical and demographic features of non-severe influenza and influenza-like illness in a tropical, urban primary care setting.
Influenza was present for most of the 116 weeks of the study.
However, the same influenza strain did not persist but a mixture of high-intensity peaks of single subtypes and co-circulation of types and subtypes at variable intensities occurred. This was consistent with previous national hospital-based surveillance which demonstrated asynchronous peaks and co-circulation of different strains. 6,8 Our surveillance results were also consistent with data showing that the northern subtropical regions of Vietnam have a more predictable seasonality than the tropical central and southern regions. 23 Heterogeneity in transmission patterns within tropical region is increasingly being recognised. 24 Vietnam has one the longest and lowest "peaks" when considered nationally, and the results of this study suggest that much of this is driven by the transmission dynamics in the south of the country. This study reports higher rates of influenza positivity in ILI than are generally described globally and fewer periods where no influenza was detected. [24][25][26] Attack rates measured through clinical settings are vulnerable to differences in healthcare seeking behaviour. 27  Primary care surveillance relies heavily on syndromic presentations. The ability of current ILI definitions to discriminate between influenza and non-influenza causes of upper respiratory tract infection in tropical settings has previously been called into question. 37 The clinics in this study also participated in a trial of mobile phone reporting of ILI which demonstrated that ILI symptoms are much more consistent through the year compared to temperate countries and that peaks in ILI activity do not always correlate with peaks in influenza activity. 11 The results of our study demonstrated that although fever, cough and rhinorrhoea were reported more commonly in those who had influenza, such symptoms also occurred in more than 70% of patients with non-influenza ILI confirming their poor discriminating ability.
Another impact of the changing economic environment in Vietnam is the use of influenza vaccination. There is no public provision for influenza vaccination in Vietnam, and prior to 2009, the use in the general public was very rare with less than 1% of the population reporting previous vaccination. 13 In the present study, 5% of individuals who presented to private clinics reported receiv-