Epidemiology and seasonality of human parainfluenza serotypes 1‐3 in Australian children

Abstract Background Parainfluenza viruses are significant contributors to childhood respiratory illness worldwide, although detailed epidemiological studies are lacking. Few recent Australian studies have investigated serotype‐specific PIV epidemiology, and there is a paucity of southern hemisphere PIV reports. We report age‐stratified PIV hospitalisation rates and a mathematical model of PIV seasonality and dynamics in Western Australia (WA). Methods We used linked perinatal, hospital admission and laboratory diagnostic data of 469 589 children born in WA between 1996 and 2012. Age‐specific rates of viral testing and PIV detection in hospitalised children were determined using person time‐at‐risk analysis. PIV seasonality was modelled using a compartmental SEIRS model and complex demodulation methods. Results From 2000 to 2012, 9% (n = 43 627) of hospitalised children underwent PIV testing, of which 5% (n = 2218) were positive for PIV‐1, 2 or 3. The highest incidence was in children aged 1‐5 months (PIV‐1:62.6 per 100 000 child‐years, PIV‐2:26.3/100 000, PIV‐3:256/100 000), and hospitalisation rates were three times higher for Aboriginal children compared with non‐Aboriginal children overall (IRR: 2.93). PIV‐1 peaked in the autumn of even‐numbered years, and PIV‐3 annually in the spring, whereas PIV‐2 had inconsistent peak timing. Fitting models to the higher incidence serotypes estimated reproduction numbers of 1.24 (PIV‐1) and 1.72 (PIV‐3). Conclusion PIV‐1 and 3 are significant contributors towards infant respiratory hospitalisations. Interventions should prioritise children in the first 6 months of life, with respect to the observed autumn PIV‐1 and spring PIV‐3 activity peaks. Continued surveillance of all serotypes and investigation into PIV‐1 and 3 interventions should be prioritised.


| INTRODUC TI ON
Acute lower respiratory infections (ALRIs) are an important cause of childhood hospitalisation globally. Alongside respiratory syncytial virus (RSV) and influenza viruses, human parainfluenza viruses (PIVs) are increasingly understood to contribute greatly to paediatric respiratory tract infections. Though PIV-associated mortality is low in high-income countries, morbidity remains high. PIVs are the second most common cause of childhood ALRI-associated hospitalisation, after RSV; comparable in frequency to influenza. 1,2 PIVs were found to account for 6.8% of all hospitalisations for fever, ALRI or both, in children aged less than 5 years in the United States between 2000 and 2004. 3 Additionally, the healthcare costs associated with PIV hospitalisation in the United States were estimated to be over $250 million USD annually between 1998 and 2010. 4 PIVs are a group of four serologically distinct human viruses, numbered 1-4, of the Paramyxoviridae family. 5 PIVs can cause a variety of respiratory infections, from mild upper respiratory illnesses in healthy adults, to croup, bronchiolitis, and pneumonia in infants, children, and severe lower respiratory infection in the elderly, and the immunocompromised. 6 The four serotypes are known to manifest distinctively, with PIV-1 and 2 being the most common aetiologic agents of croup, whereas PIV-3 and 4 are more commonly associated with bronchiolitis and pneumonia. 6,7 There are no licensed prophylactic agents, vaccinations or therapies for any PIV serotype [8][9][10] ; development having stalled in part due to the paucity of studies on their health burden. 11 Infants and children are a particular risk group for infection with PIVs, and all serotypes are known to result in hospitalisation in young children. [12][13][14][15] Though all serotypes are common causes of childhood illness, PIV-3 and 1 are thought to be the most frequent causes of hospitalisation, with PIV-2 presenting a slightly lower health burden. [12][13][14] The epidemiology of PIVs in resource-poor settings remains poorly understood, particularly in the tropical and subtropical regions of the southern hemisphere, despite the high rates of ALRI-associated mortality which persist in these environments. 16 Studies in America and Japan have reported spring peaks in PIV-3 circulation, and autumn and winter peaks for PIV-1 and 2 across the northern hemisphere. [12][13][14][15] Although biennial peaks in PIV-1 activity, generally in odd-numbered years, are usually reported, one Korean study observed relatively indistinct PIV-1 peaks. 17 There is even less consensus on yearly PIV-2 fluctuations, with some studies reporting biennial, others reporting annual and others reporting only sporadic outbreak-like activity. 4,12,15,17 One Brazilian study identified biannual PIV-2 peaks, which have not been observed elsewhere to date. 18 Overall, reports on PIV epidemiology from outside temperate, northern hemisphere regions remain rare: the most comprehensive existing reviews of respiratory virus seasonality often lack PIV serotype specificity. 2,19 This study set out to investigate the serotype-specific epidemiology of PIV-1 and 3 in Western Australian children, reporting incidence rates of laboratory-confirmed PIV hospitalisation by age group and Aboriginal status, and seasonal parameters through dynamic transmission models.

| Study setting
Western Australia (WA) encompasses the western third of Australia, an area of approximately 2.5 million square kilometres, and, as of 2013, contained approximately 2.5 million people, around 100 000 of whom identify as being of Aboriginal and/or Torres Strait Islander origin (hereafter respectfully referred to as Aboriginal). Three quarters of the population reside in the temperate climatic region of metropolitan Perth and its surrounds. 20

| Data sources and study cohort
This analysis formed part of a larger study investigating the pathogen-specific epidemiology of respiratory infections in a populationcohort of WA births from 1996 to 2012 using probabilistically linked administrative data. Full study details are available elsewhere.

| Parainfluenza virus episodes
PathWest is WA's leading public pathology provider, and, as the state's major referral pathology laboratory, receives specimens from all public and private hospitals. Respiratory samples collected routinely from children aged 16 and under, mostly in hospitals, underwent routine viral diagnostic panels involving serology, culture, direct antigen detection and polymerase chain reaction (PCR) testing. Positive and negative results for PIV-1, 2 and 3 were available.
The WA HMDC is a state-wide dataset recording admissions to private and public hospitals, psychiatric hospitals, and day surgeries across WA. 23 Hospital admissions were linked to laboratory diagnostic records for specimens collected from the same individual within 48 hours of the hospital admission date. 1 We restricted hospital admission records to the period covering 2000-2012, as laboratory records were only available from 2000.

| Statistical analysis
We used two different primary outcome measures for this study.
The first of these, used to calculate incidence rates, was hospitali- conducted in all hospitalised children, we also assessed the rate of testing. The secondary outcome measure was all positive detections of PIV-1, 2 and 3, regardless of hospitalisation status. This measure was used to investigate seasonal dynamics using complex demodulation and dynamic transmission models.

| Incidence rates
The incidence rates of PIV-1, 2 and 3 were calculated using time to event survival analysis methods allowing for multiple PIV hospitalisations per person, as in previous analyses of this dataset for RSV. 24 Person time-at-risk was calculated from the start of the study period (1st January 2000, or from the date of birth) until the time of study exit, taken as either the 31st of December 2012, the date of death, or date of hospitalisation for PIV or for respiratory viral testing. We calculated incidence rates of testing and PIV-confirmed hospitalisations with 95% confidence intervals (95% CI) according to age and Aboriginal status. Analyses were conducted using STATA SE (version 14.1) by Stata Corp.

| Seasonality
To assess seasonality, we used data on all PIV laboratory detections for children in the birth cohort as we had no reason to believe the seasonality would differ between hospitalised and non-hospitalised children. SEIR compartmental models are a useful tool for modelling infectious diseases and, when fitted to incidence data, can inform on transmission parameters and seasonality. SEIR models categorise all individuals in a population as either susceptible (S), infected but not yet infectious (E), infectious (I) or recovered (R). We modelled transmission of each serotype of PIV using a deterministic SEIRS model with a single age class, waning immunity and seasonality in transmission (Table S1). The models were implemented and fitted in MATLAB using the ode45 differential equation solver and fitted to data using fminsearch.
Complex demodulation is an analytic approach for cyclical or seasonal time series data that extract the timing and size of peaks over time, and has been used to analyse sleep cycles, cardiovascular variability, and suicides, in addition to seasonal pathogens such as RSV. [25][26][27][28] Full details of this method are provided elsewhere. 29 F I G U R E 1 Process of dataset construction and record linkage. Databases used in birth cohort construction are in green boxes, and administrative hospital and laboratory databases are in blue boxes. Note that the number of admissions differs from the size of the birth cohort, as multiple admissions per person are permitted. The yellowshaded box highlights the analytic dataset used for incidence rate calculation, while the purple-shaded box highlights analytic dataset for seasonality and transmission models Briefly, the method deconstructs the time series into an amplitude and phase, assuming a given periodic frequency. We applied this approach to weekly PIV incidence counts for serotypes 1-3 individually, assuming a 52-week period and using a 52-week moving average filter.

| Ethics
Approvals for this study were obtained from the WA Department of Human Research Ethics Committee and the WA Aboriginal Health Ethics Committee. Data access was provided by the WA Data Linkage Branch.  Table 1 shows the frequency of PIV testing in hospitalisations in Aboriginal and non-Aboriginal children. Hospitalisations that underwent testing for PIV were 1.7-3.3 times higher in Aboriginal children than non-Aboriginal children with the per cent positivity ranging from 2% in neonates (aged less than 1 month) to 6%-7% in children aged 12-23 months (Table 1). Immunofluorescence was the most common diagnostic method used to detect PIV (52.1% of laboratory-linked hospitalisations), with 23% being detected by PCR. Croup was the most common primary diagnosis for PIV-1 (28.5%) and PIV-2 (29.5%) confirmed hospitalisations, whereas bronchiolitis was the most common primary diagnosis for PIV-3 (28.5%). Overall, 74% of all PIV-1 and 3 infections occurred in children aged less than 2 years, and PIV-3 had the lowest median age of infection (11 months), followed by PIV-2 (12 months) and PIV-1 (15 months). For each PIV serotype, incidence rates peaked in children aged from 1 to 5 months (Table S2). The highest hospitalisation rate overall was observed for PIV-3 in this age group  PIV-2 had the lowest incidence and least consistency in peak timing.

| PIV seasonality
Estimates from the mathematical modelling analysis are shown in Table 2, with plots of the model fit to data in Figures S1-S3 (Table 2).

| D ISCUSS I ON
We used a population-based dataset to describe the age-specific incidence and seasonal dynamics of PIV in children. We found that overall PIV rates were highest in children aged less than 2 years, particularly in infants aged 1-5 months, and that PIV-3 had the highest incidence of all examined serotypes for each age group, followed Global studies suggest that, of all serotypes, PIV-3 presents the most immediate public health concern for children under five, being the most frequent cause of morbidity, followed by PIV-1, then PIV-2. 3,7,12,17 Compared to influenza virus and RSV assessments from this same dataset, 1 we have shown, for children aged 1-5 months, a higher hospitalisation rate for PIV-3 alone than for influenza hospitalisations, and PIV-3 rates eclipsed influenza rates in children until the age of 2. Similarly, previous studies variously attribute a disease burden of PIVs comparable to, or greater than, influenza viruses, yet significantly lower than RSV. 2,19,31 The burden of all PIV serotypes was significantly higher in Aboriginal children, compared to non-Aboriginal children. This result has been reported for other respiratory viruses in Indigenous populations from around the world, in part due to increased risk factor exposure in marginalised, often rural, population groups. 33,34 Aside from PIV-2, our estimates of R 0 and duration of immunity for PIV were consistent with those of other respiratory pathogens. 35,36 Owing to the low sample size for PIV-2, our ability to fit a seasonal model to that serotype was limited. Though our observation of biennial PIV-1 circulation resembles that made by previous studies, Western Australian PIV-1 peaks were found to occur in even-numbered years, rather than the odd-numbered year peaks frequently reported in northern hemisphere studies. 7,12,15 This biennial peaking pattern for PIV-1 is consistent with that observed for RSV, 37,38 which may result from a degree of residual immunity in off years, but it is unclear why this is not seen in PIV-3. Though PIV-3 was present in circulation each year, PIV-3 frequency was slightly lower in even-numbered years when PIV-1 was in circulation, perhaps indicating a degree of competitive inhibition between PIV serotypes. 39 These factors may also play a role in the discrepancy in PIV-1 peak years between northern and southern hemisphere sites.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/irv.12838.

DATA AVA I L A B I L I T Y S TAT E M E N T
Datasets used in this study are available, by application, via the cor-