Estimates of influenza‐associated hospitalisations in tropical Singapore, 2010‐2017: Higher burden estimated in more recent years

Abstract Background We previously estimated Singapore's influenza‐associated hospitalisation rate for pneumonia and influenza (P&I) in 2010‐2012 to be 29.6 per 100 000 person‐years, which corresponds to 11.2% of all P&I hospitalisations. Objectives This study aims to update Singapore's estimates of the influenza‐associated pneumonia and influenza (P&I) hospitalisation burden using the latest data from 2010 to 2017. Methods We estimated the number of P&I hospitalisations associated with influenza using generalised additive models. We specified the weekly number of admissions for P&I and the weekly influenza positivity in the models, along with potential confounders such as weekly respiratory syncytial virus (RSV) positivity and meteorological data. Results In 2010‐2017, 16.3% of all P&I hospitalisations in Singapore were estimated to be attributed to influenza, corresponding to an excess influenza‐associated P&I hospitalisation rate of 50.1 per 100 000 person‐years. Higher excess rates were estimated for children aged 0‐4 years (186.8 per 100 000 person‐years) and elderly aged ≥ 65 years (338.0 per 100 000 person‐years). Higher influenza‐associated hospitalisation rates were estimated for 2016 and 2017 (67.9 and 75.1 per 100 000 persons, respectively) years when the influenza A(H3N2) subtype was dominant. Conclusion Influenza burden in Singapore has increased since 2010. Influenza vaccination programmes should continue to be prioritised for the young and the elderly.


| INTRODUC TI ON
There is a widespread consensus that influenza imposes a considerable burden on public health, including substantial numbers of severely ill patients resulting in hospitalisations and deaths. [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19] The impact of influenza on public health varies across seasons, mainly due to the varying influenza virus types/subtypes, vaccine uptake in the community and the match between the recommended vaccine with the circulating viruses. [4][5][6][7][8] Quantifying the burden of influenza disease is important so as to assess the severity of the virus, evaluate how well public health services are coping with the additional burden or determine whether additional interventions should be considered to reduce the impact.
Estimating the burden is challenging because many who are ill typically receive empiric treatment without laboratory confirmation of influenza infection. In 2015, the World Health Organization (WHO) published a manual to introduce methods to estimate the burden associated with seasonal influenza. 20 There have also been advances in methods introduced by various studies to measure influenza burden, 21 including statistical regression models such as the generalised additive model (GAM). An advantage of using regression models is that confounding factors, such as temperature or longterm time trends, which may influence the observed outcome of choice (ie, pneumonia admissions) can be controlled for.
Singapore, being a small globally connected city-state in an equatorial region, has a climate that is high in temperature and humidity. It also experiences influenza activity that is vastly different from temperate countries, characterised by year-round influenza activity and bimodal peaks that usually occur in the beginning and middle of the year. 22 Previous local studies have highlighted the significant burden that influenza imposes on public health. One older study estimated the influenza-associated excess mortality rate to range from a low of 0.2 per 100 000 persons in 1975 to a high of 54.4 per 100 000 persons in 1957 during the pandemic. 13 More recent studies estimated the excess mortality rate for allcause deaths to be 14.8 per 100 000 person-years in 1996-2003 and the excess hospitalisation rate for pneumonia and influenza to 29.6 per 100 000 person-years in 2010-2012. 14,15 The excess hospitalisation rate was also found to be substantially greater in the very young (infants aged 0-5 months) and the very old (≥75 years).
This study aims to update Singapore's estimates of the influenzaassociated pneumonia and influenza (P&I) hospitalisation burden using the latest data from 2010 to 2017.

| Data
We obtained the weekly number of admissions for pneumonia and influenza (P&I) (ICD-9 480-487 and ICD10 J10-J18) for all acute hospitals from the inpatient administrative database. Five age groupings were analysed in our study: 0-4 years, 5-49 years, 50-64 years, ≥65 years and all ages. These age groupings were in accordance with the WHO manual for estimating influenza burden. 20 The Ministry of Health (MOH), Singapore, has a national surveillance programme for influenza. Virological surveillance is carried out on respiratory specimens from patients with ILI symptoms who visit public primary care centres (accounting for ~ 20% of all primary care consults nationwide), 23 public hospitals (accounting for ~ 75% of all hospitalisations nationwide) 24 or sentinel private primary care clinics. These specimens are sent to the National Public Health Laboratory (NPHL) to be tested for influenza and other respiratory pathogens. We used the all-age weekly influenza positivity from 2010 to 2017, derived by calculating the proportion of specimens that had been laboratory-confirmed positive for influenza, as a proxy variable for influenza activity. We also highlighted the dominant influenza type/subtype for each year based on results from another study. 25 To adjust for potential confounding, we also obtained the weekly respiratory syncytial virus (RSV) positivity data from the paediatric departments of two government acute hospitals, as well as meteorological data (weekly mean temperature and weekly mean relative humidity) from the National Environment Agency (NEA).
These data sources have been described in more detail in previous studies. 14,15

| Statistical analysis
We used a generalised additive negative binomial model to estimate the weekly number of P&I hospitalisations associated with influenza during the 8-year study period. Generalised additive models (GAMs) have been used in multiple studies for estimating influenza burden. 4,8,16,17 An advantage of the GAM is that it allows the data to suggest an appropriate functional form for the relationship between an explanatory variable and the response using a penalised likelihood function. 26 We also included a spline function to account for the non-linear long-term time trend. Each model for the five different age groupings was fitted as follows:
a Correlation between weekly all-age P&I hospitalisation numbers and weekly influenza A&B positivity using the Spearman's method.    Figure 2). 16.  6,17,29 As discussed by Abdel-Hady et al, 5 it is difficult to make like-for-like comparisons between the countries' estimates due to different outcome variables used (ie P&I vs SARI vs all-cause hospitalisations), unequal health-seeking behaviour (accessibility to healthcare at an acute hospital), non-identical methodologies (multiplication method vs statistical regression models) and different study periods. It is notable that given the year-round influenza activity in Singapore, our annual influenza positivity percentages were consistently more than 40% every year; this is more than double that reported by Chile and Oman (averages of 8.8% and 17%, respectively), countries that experience distinct influenza seasons in the winter. 5,6 Hence, it could be possible that the burden imposed by influenza on countries with year-round influenza activity is relatively greater in a calendar year, with all other things being equal.

| D ISCUSS I ON
We estimated higher overall influenza-associated P&I hospitalisation rates for the young and elderly across the study period, resulting in the J-shaped curve seen in other studies. 9,15,18 The higher rates estimated for these two groups suggest that influenza vaccination programmes should continue to be prioritised for the young aged 0-4 years and the elderly aged 65 years old and above. Vaccination has been found to be one of the most effective ways to reduce influenza burden and also a cost-effective intervention strategy for targeted age groups. 34 37 Further studies should be conducted to determine the optimal vaccination schedule in countries with yearround influenza activity, especially when there is growing evidence of waning vaccine effectiveness. 38 The estimated proportion of P&I hospitalisations attributable to influenza was the highest for the youngest age group but, in contrast, was the lowest for the oldest age group. These results suggested that while both the young and the elderly were more prone to influenza-attributed hospitalisations, 10,11,39 older people were also particularly at risk of developing severe symptoms, and requiring hospitalisations, from other pathogens.

| Limitations and strengths
There were a few limitations to our study. Firstly, we used the all-age influenza positivity variable in all models regardless of age group.
Influenza positivity percentages might vary among the different age groups, but the age-specific weekly number of respiratory specimens tested for influenza was too small for use in a time series model. This was especially so for the 0-4 years age group as parents could be more reluctant in giving consent for nasopharyngeal swabs to be taken.
Likewise, we also used RSV data from paediatrics data in our models due to lack of such routine testing in adults. Secondly, other bacterial and viral causes of pneumonia, such as rhinoviruses and streptococcus pneumonia, were not specifically accounted for in our model. Instead, we used a smoothing spline function to adjust for background variations. Lastly, our study was confined to P&I hospitalisations, which is a subset of all-cause or respiratory hospitalisations. Hence, our estimates might be an under-estimation of the true influenza burden.
The strengths of our study were Singapore's comprehensive data collection system and that our observed P&I hospitalisation numbers accounted for all acute hospitals in Singapore. The respiratory specimens tested for influenza were also collected from multiple sites around the country. This means that our data were reflective of the nation as a whole and population-based estimates were derived.

| CON CLUS ION
The overall influenza-associated P&I hospitalisation rate in