Sustained low rotavirus activity and hospitalisation rates in the post-vaccination era in Belgium, 2007 to 2014.

In 2006, Belgium was the first country in the European Union to recommend rotavirus vaccination in the routine infant vaccination schedule and rapidly achieved high vaccine uptake (86-89% in 2007). We used regional and national data sources up to 7 years post-vaccination to study the impact of vaccination on laboratory-confirmed rotavirus cases and rotavirus-related hospitalisations and deaths. We showed that (i) from 2007 until 2013, vaccination coverage remained at 79-88% for a complete course, (ii) in children 0-2 years, rotavirus cases decreased by 79% (95% confidence intervals (CI): 68--89%) in 2008-2014 compared to the pre-vaccination period (1999--2006) and by 50% (95% CI: 14-82%) in the age group ≥ 10 years, (iii) hospitalisations for rotavirus gastroenteritis decreased by 87% (95% CI: 84-90%) in 2008--2012 compared to the pre-vaccination period (2002--2006), (iv) median age of rotavirus cases increased from 12 months to 17 months and (v) the rotavirus seasonal peak was reduced and delayed in all post-vaccination years. The substantial decline in rotavirus gastroenteritis requiring hospitalisations and in rotavirus activity following introduction of rotavirus vaccination is sustained over time and more pronounced in the target age group, but with evidence of herd immunity.


Introduction
Globally, rotavirus is the leading cause of severe acute gastroenteritis in children aged less than 5 years, resulting in substantial morbidity and mortality [1]. Most children are infected at least once with rotavirus by the age of 5 years, with severe disease occurring most commonly between the ages of 6 months and 2 years [2,3]. Before vaccine introduction in Belgium in 2006, the burden of rotavirus disease was high compared with other European countries and rotavirus was estimated to account for nearly 5,600 hospitalisations annually in children < 7 years [4]. In June 2006, the two-dose oral monovalent vaccine (Rotarix, GlaxoSmithKline Biologicals, Rixensart, Belgium) was marketed, followed by the three-dose oral pentavalent vaccine (RotaTeq, Sanofi Pasteur MSD, Lyon, France) in June 2007. Rotavirus vaccination has been recommended by the Superior Health Council (the Belgian National Immunization Technical Advisory Group) for all infants at 8 weeks of age since October 2006 and has been partially reimbursed since November 2006. The monovalent vaccine is administered at 8 and 12 weeks of age and the pentavalent vaccine at 8, 12 and 16 weeks of age. Since January 2007, rotavirus vaccination has been offered systematically during preventive consultations organised by the government agency well-baby clinics. All children between 0 and 3 years are actively invited via their parents or guardians to attend these easily accessible consultations in their local community, free of charge (including medical acts like prescribing and administering vaccines). Unlike other recommended childhood vaccines, rotavirus vaccines are only partially reimbursed on a per-prescription basis. Currently, EUR 11.8 per dose is co-paid by caregivers (usually the parents) of vaccine recipients [5]. Rotavirus vaccine introduction led to a substantial decline in rotavirus activity during the period from July 2007 to June 2008 [6] and a reduction in rotavirus-related hospitalisations in the period from June 2007 to May 2009, based on a sample of 12 hospitals in Belgium and on a study in a university hospital [7,8]. A case-control study conducted in Belgium in 2008-2010 showed that the effectiveness of two doses of the monovalent rotavirus vaccine against hospital admissions was 90% [9].
Rotavirus gastroenteritis is not mandatorily notifiable in Belgium. Surveillance is conducted through a laboratory-based sentinel network registering positive rotavirus tests, and the secondary analysis of healthcare utilisation databases (rotavirus-related hospitalisations, for which registration is obligatory). Rotavirus vaccine coverage is monitored through cluster sample surveys [10,11]. We collected all available surveillance and coverage data (one regional and seven national data sources) to study the impact of rotavirus vaccination in more detail and for a longer follow-up period. More particularly, we assessed trends in rotavirus testing and detection, hospitalisations and deaths due to rotavirus or acute gastroenteritis and rotavirus vaccination coverage. We analysed weekly rotavirus activity for up to 7 epidemiological years (1 July to 30 June) after vaccine introduction, described changes in both timing and age of rotavirus infection, obtained evidence for herd immunity, looked for changes in testing behaviour pre-and post-vaccination and estimated the coverage of the two rotavirus vaccines separately.

Methods
Data were derived from eight different databases, surveillance systems and other data sources in Belgium (Table 1). Data analysed include: vaccination coverage, number of rotavirus tests and confirmed infections, hospitalisations and deaths due to rotavirus or acute gastroenteritis. All data were processed without patient-identifying information. We assigned a random number (one, two or three tests) to weeks in which the exact number of tests/confirmed infections/hospitalisations could not be disclosed to us for privacy reasons under Belgian legislation (e.g. fewer than four tests). Analyses were performed using R [12] and SAS Enterprise Guide (version 5.1).

Vaccination coverage
Vaccination coverage was derived from three independent sources: coverage surveys, vaccination sales data and reimbursement data.
Coverage surveys: in Belgium vaccination coverage in children is estimated in the three regions (Brussels-Capital region, Flanders and Wallonia) based on cluster sample surveys [10,11]. We estimated a national coverage for 2012 as a weighted average of the three regional rates, using the population under 1 year of age of every region of the corresponding year. Population and birth statistics were retrieved from Statistics Belgium [13].
Vaccination sales data: the annual number of doses sold in Belgium was obtained from GlaxoSmithKline Biologicals s.a (Rotarix) and Sanofi Pasteur MSD    reimbursement data allowed us to derive the number of infants who had received at least one rotavirus vaccine dose and the number of infants who had completed a course of vaccination by week (i.e. two doses Rotarix or three doses RotaTeq). IMA-AIM data contain the delivery date of the vaccine (i.e. date of purchase), which is not necessarily the administration date. It was estimated that 70% of infants receive rotavirus vaccine within one week after purchase and 90% within 4 weeks of purchase (for all doses, unpublished data from Vaccinnet [14]). Annual vaccination coverage was calculated by dividing the number of infants with a complete vaccination scheme reimbursed, by the number of vaccine-eligible infants of the same year.

Hospitalisations for rotavirus and acute gastroenteritis
Rotavirus-related hospitalisations were obtained from two independent databases.

Minimal Hospital Data
The Minimal Hospital Data (MHD) are managed by the Federal Public Service of Health and are an electronic collection of anonymised records of patients admitted to all public and private hospitals in Belgium. For the period 1999-2011, we obtained the monthly number of hospitalisations with primary discharge diagnoses of: (i) rotavirus enteritis (by diagnosis code International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 008.61) and (ii) any other acute gastroenteritis not coded as rotavirus (i.e. diarrhoea of determined aetiology (bacterial (001-005 and 008.0-008.5), parasitic (006-007) and viral (008.6)), and/or diarrhoea of undetermined aetiology (presumed infectious (008.8-009.3)) [17]. Hospitalisation rates were calculated by dividing the annual number of rotavirus enteritis or acute gastroenteritis hospitalisations by the age-specific Belgian population for the corresponding years [13]. Because the age of a hospitalised person could not be derived reliably from MHD (only year of birth is registered), the Carenet-NCSF database was used to investigate the age distribution of rotavirus-related hospitalisations (see next paragraph).

Hospital database Carenet-NCSF
Carenet is designed for electronic information exchange between hospitals and health insurance companies about hospital admissions. In July 2006 Carenet covered 88% of Belgian hospital beds, in July 2009 this increased to 99%. We could only obtain Carenet data from members of the NCSF health insurance company (see above). We obtained all records (2004-2012) on hospitalised patients who were member of the NCSF for which the diagnostic field included one of the following search strings: 'rota' or ICD-9-CM code '008.61' or ICD-10 code 'A08.0' [18]. A medical clinician searched the diagnostic fields of the retrieved records manually and selected those for which rotavirus was likely to be the main reason for hospitalisation. Data extractions and analyses related to NCSF were performed at the Medical Management Department of the NCSF under the supervision of the Chief Medical Officer.
Weekly numbers of rotavirus hospitalisations from NCSF members were used to cross-validate trends observed in MHD, to investigate the age distribution of rotavirus hospitalisations and to study a possible shift in the peak number of hospitalisations after the introduction of rotavirus vaccination.

Deaths due to gastroenteritis
We calculated the death rate in children < 5 years due to gastroenteritis using the Standardized Procedures for Mortality Analysis (SPMA) website [19]. Data were available for 1987-2000 and 2003-2010. We included all intestinal infectious diseases with ICD-9 codes 001-009. From 1998 onwards ICD-10 codes A00-A09 were used. More detailed coding of mortality data were available for Flanders for the period 2000-2012 allowing the use of ICD-10 codes (A00-A09) and the specific code for rotavirus (A08.0) [20]. The annual death rate was defined as the ratio of the number of deaths to the number of people in the age group < 5 years.

General definitions and assumptions
We defined epidemiological years from the beginning of July to the end of June of the following year. We defined three periods of analysis to reflect the introduction  (Table 1).
Currently, there is no standard way to determine the onset, peak and end of a rotavirus epidemic. Indeed, the European Centre for Disease Prevention and Control (ECDC) advises that each country specifies its own definition [21]. To explore the changes in timing and the duration of the rotavirus epidemic, we defined the start and end of the rotavirus epidemic as the week in which more or less than the average weekly number of positive rotavirus tests occurred for a particular epidemiological year ('method average'). We also used the definition proposed by Tate and colleagues ('Tate method') [22], and recommended by the ECDC when the proportion of positive rotavirus laboratory tests is available. This method defines the start and end of the epidemic for each epidemiological year as the 2 first or last consecutive weeks during which the proportion of positive tests is ≥ 10%. The peak of the epidemiological year was defined as the week with the highest number of positive rotavirus tests or rotavirus hospitalisations or proportion of rotavirus positive tests.
We estimated the vaccine impact, expressed as the percentage change, by comparing the mean number of laboratory confirmed and hospitalised rotavirus cases in unvaccinated populations (pre-vaccination period) to the mean in vaccinated populations (post-vaccination period) [23]. Confidence intervals were calculated using Fieller's method for the confidence interval of the quotient of two means, assuming Gaussian distributions [24].
Vaccination sales data: in the first 7 months following vaccine introduction, vaccination coverage (two-or three-dose scheme) in infants < 1 year was 32.5% and rapidly increased to reach 89.4% in 2013, with an average of 87.5% (minimum-maximum (minmax) range: 85.5-89.4%) for 2007-2013 ( Figure 1).  Figure 1). Of these vaccinated infants, 9.3% did not complete the two-or three-dose scheme. This percentage slightly decreased from 10.8% in 2007 to 7.9% in 2012, and is larger for the three-dose vaccine (RotaTeq) than for the two-dose vaccine (Rotarix) (17.3% vs 6.8% in 2012). The proportion of infants vaccinated with RotaTeq (vs Rotarix) increased after its introduction up to 19.5% in 2010 and decreased thereafter, to reach 10.6% at the end of 2012.

Number of rotavirus tests and laboratoryconfirmed rotavirus infections
We excluded 44,284 (24.2%) reimbursed tests from the IMA-AIM database and 2,571 (6.0%) laboratoryconfirmed infections from the SLN because these were considered duplicates.
The number of laboratory-confirmed rotavirus infections in children 0-2 years of age decreased by 79.2% after widespread vaccination (Figure 1 and 2a, Table  2), whereas the number of reimbursed rotavirus tests decreased only by 31.0% ( Table 2). The proportion of positive tests in the SLN decreased from 24.8% (prevaccination period) to 7.4% (post-vaccination period) for children 0-2 years and from 45.5% to 10.1% for infants < 1 year.
The reduction of rotavirus infections was highest in infants below 1 year of age with 80.1% (95% CI: 72.1-87.7) reduction in infants 0-5 months and 85.8% (95% CI: 78.1-92.4) reduction in infants 6-11 months of age. A substantial reduction was seen in the age groups too old to be protected directly by vaccination (i.e. evidence for herd immunity, Figure 2a). In the age group of 10 years and older, the reduction was 50.0% ( Table  2).
The median age of a person tested positive for rotavirus increased from 12 months pre-vaccination to 17 months post-vaccination. In the pre-vaccination period there were more positive tests in infants 6-11 months of age (59.7%) than in 0-5 months (40.3%), while in the post-vaccination period this difference almost disappeared (51.3% in infants 6-11 months of age and 48.6% in 0-5 months). After the introduction of the rotavirus vaccination programme, a larger proportion of positive rotavirus tests occurred in children 12 months and older ( Table 2). The age distribution of the number of tests reimbursed for rotavirus did not change much following widespread vaccination ( Table 2).
The peak month of laboratory-confirmed rotavirus cases shifted from February in the pre-vaccination period to April after vaccination (Figure 3). The maximum weekly number of cases dropped by 77.8% (95% CI: 74.4-80.9%), from 633 cases pre-vaccination to 141 post-vaccination ( Figure 1). The impact of vaccination on the duration of the rotavirus epidemics depends on the method used to determine this duration: with our approach ('method average'), no clear change was observed, but with the method used by Tate et al. [22], post-vaccination epidemics were found to be 10 weeks shorter than pre-vaccination ( Figure 3). Testing behaviour (distribution of the number of rotavirus reimbursed tests over an epidemiological year) did not change after the introduction of the vaccines (results not shown).

Hospitalisations
Based on the discharge data (MHD), the overall number of rotavirus-related hospitalisations decreased by 73.3% after widespread vaccination (Table 2). During the pre-vaccination period, the mean incidence of rotavirus hospitalisations was 46.0 per 100,000 person-years (range: 41.7-53.5), compared with 12.3 per 100,000 person-years (range: 11.7-12.9) in the post-vaccine period ( Table 2). The largest reduction occurred in infants 6-11 months of age (90%, Carenet-NCSF), but a substantial reduction was also seen in persons too old to be protected directly by vaccination (Figure 2b). Before vaccination, one-third (29.2%) of children hospitalised for rotavirus were 6-11 months old. After vaccination, only one-fifth (18.3%) of rotavirus-related hospitalisations occurred in this age group. Furthermore, 43.0% of all hospitalisations occurred in infants aged 0-11 months old in the pre-vaccination period, compared with 31.4% in the post-vaccination period.
Peak number of rotavirus-related hospitalisations shifted from February to April in the post-vaccination period ( Figure 3). The width of the epidemic (based on the 'method average') did not change.

Deaths due to gastroenteritis
In the period 1987-2005, between one and seven deaths per year occurred in children < 5 years due to gastroenteritis, representing a death rate of 0.

Discussion
During the 7 years following rotavirus vaccine introduction, we established that: (i) vaccine uptake remained high; (ii) the substantial decline in both rotavirusrelated hospitalisations and laboratory-confirmed rotavirus persisted; (iii) rotavirus incidence peaked annually in spring instead of winter; (iv) the average age at infection and hospitalisation increased and (v) the number of laboratory-confirmed and hospitalised rotavirus cases decreased also in unvaccinated persons (evidence for herd immunity).
The estimated vaccination coverage was consistently high using different data sources. We found that on average 9.3% of Belgian infants did not complete their schedule, which is higher than the 2% found by a coverage survey conducted in Flanders [10]. However in Flanders, coverage and compliance with vaccinations are typically higher than in the other regions [10], and this Flemish survey considered two doses as fully vaccinated whereas a complete schedule with RotaTeq consists of three doses. Also, we found that the proportion of infants who completed the series was higher for the two-dose than the three-dose vaccine, similar to findings in the United States (US) [25][26][27].
In the pre-vaccination period, we estimate that rotavirus infections were responsible for 21.1% of hospital admissions for acute gastroenteritis, which is in line with previous European estimates (21-58% [28][29][30]). We found a substantial decrease in laboratory-confirmed cases and rotavirus-related hospitalisations and deaths in the post-vaccination period, which confirms the reduction of 87% of rotavirus hospitalisations predicted by a mathematical model assuming uptake rates similar to those for other routine infant vaccinations [31]. Furthermore, this considerable reduction is in line with the high effectiveness of the rotavirus vaccines (over 85%) [32]. A systematic review of ecological studies from eight countries reported a 49-89% decline in laboratory-confirmed rotavirus hospital admissions in children less than 5 years old within 2 years of vaccine introduction [33]. The evidence for a direct vaccinationrelated reduction is further strengthened by a lower proportion of rotavirus-positive tests in infants < 1 year following rotavirus vaccine introduction, i.e. a decrease from 45.5% to 10.1%.
In addition, the typical rotavirus seasonal peak apparent in winter and early spring before introduction of the vaccine was reduced and delayed in all postvaccination years. Although pre-vaccination this was based on data from only 3 (laboratory tests) or 2 (hospitalisations) epidemiological years, these results pointed in the same direction for both. These changes in seasonal patterns are unlikely to be due to year-toyear variations, and probably reflect a decline in virus transmission, as predicted by mathematical modelling applied to England and Wales [34]. In the Netherlands, where a rotavirus vaccination programme is absent, the peak rotavirus incidence was exceptionally low in 2013-14 [35]. The authors offered as explanations the low birth rate, mild winter, high rotavirus incidence in the previous year and the introduction of rotavirus vaccination in neighbouring countries. Also in Belgium we found the number of rotavirus positive tests in 2013-14 to be lower than in any of the previous epidemiological years, although much less pronounced than in the Netherlands. During that epidemiological year, birth rate, vaccination coverage and rotavirus testing behaviour did not change compared with the previous epidemiological years, and no exceptionally high rotavirus peak preceded 2013-14 [13]. Besides, although 2013-14 was characterised by an exceptionally warm winter, this was also the case for 2006-07 and 2007-08 [36]. Hence, the explanations proposed for the extremely low rotavirus incidence in the Netherlands in 2013-14 seem unlikely to explain the low incidence in Belgium in the same epidemiological year. Clearly, more research is needed to get insight in the cause(s) of rotavirus annual variations, both in the presence and absence of vaccination. The impact of vaccination on the average length of the rotavirus epidemic is difficult to determine due to the lack of a standard method to measure this length. According to our calculation method, the length of the yearly rotavirus epidemic was unchanged by the introduction of the vaccines. Yet, the method described by Tate and colleagues [22] suggests a 10-week decrease. Note that comparison between the two methods is difficult as only one pre-vaccination epidemiological year was available for the proportion of rotavirus tests being positive. We did not observe clear biennial increases in rotavirus activity in the post-vaccine era as observed in the US [22]. This might be due to different transmission patterns resulting from the lower speed and level of vaccine uptake in the US vs Belgium. In the US, rotavirus vaccines have been recommended for routine use since 2006 and coverage (of mainly the three-dose vaccine) increased gradually from 44% in 2009 to 73% in 2013 for a complete schedule [37], whereas vaccination coverage in Belgium increased to 79-88% within 7 months. The free of charge, low-threshold community outreach vaccination for Belgian infants (together with using predominantly the two-dose rotavirus vaccine) could lead to higher vaccination coverage and better completion rates. Another reason could be that the two vaccines differ in strain composition and may therefore exert different pressures on the circulating serotypes and overall transmission dynamics. It remains difficult, however, to explain differences in cycling patterns: modelling studies show that small changes in rotavirus transmission dynamics can lead to very different cycling patterns [38,39].
We observed an increase in the median age of confirmed rotavirus cases. This was predicted by a model applied to England and Wales, based on vaccination coverage of 91% [34]. We did not observe an increase in hospitalisations in older children, in contrast to the findings in Austria, where during the fourth year postvaccination an increase of 48% in hospitalisation rates for rotavirus was observed in children 5-9 years of age [40]. Such increase was also predicted by models assuming the probability of infection to depend on the number of previous infections, and not on age [38,39]. Paulke-Korinek and colleagues mention that the incidence increase in Austria could also be due to very high rotavirus activity in 2011 [40]. In Belgium, an increased rotavirus activity was noted in 2012-13 compared with the previous epidemiological year (Figure 2a), but it is not known if this is reflected in an increase in hospitalisations, since these data are not yet available.
In the age group older than 10 years, who were not yet vaccinated, we observed a 50.0% decrease in confirmed rotavirus cases, suggesting an indirect protection. In many countries, the reduction in rotavirus disease has indeed been broader than expected based on vaccine coverage alone [33]. The decrease in symptomatic infections in the vaccinated population most likely leads to a reduced chance of being exposed to infection for those not immunised [34].
The results of this descriptive and ecological design may reflect factors not related to immunisation, such as natural fluctuations or strain variation [41]. For instance, increased circulation of a specific rotavirus strain causing relatively mild disease could result in lower rotavirus related disease burden. However, after vaccine introduction in Belgium, G2P [4] strains, which are associated with more severe gastroenteritis [42,43] were observed to increase relative to other strains [44]. Also, the increased proportion of G2P [4] was seen more in vaccinated compared with unvaccinated children, suggesting strain-specific differences in vaccine effectiveness is playing a role in altering the genotype distribution [44]. Despite this strain shift, our study shows a strong decrease in various manifestations of the rotavirus disease burden, confirming rotavirus vaccination is highly effective in reducing disease.
Hospitalisation data might be inconsistent in relation to rotavirus coding based on irregular laboratory confirmation and the potential influence of rotavirus vaccination on coding practices. However, all findings were consistent using different independent data sources (including two parallel hospital databases), with different methods of registering diagnoses. Moreover, the proportion of positive tests decreased, reflecting lower rotavirus prevalence and we found no evidence of important changes in testing behaviour based on the number and the seasonal distribution of reimbursed tests. However, the reductions in rotavirus burden calculated using different data sources should be compared with caution, as the different data sources did not cover the same periods. Nevertheless, epidemiological years 2005 to 2010 were covered by all data sources, and the results pointed clearly in the same direction. We took a conservative approach in identifying duplicates, assuming a maximum of one episode per year. This implies an underestimation of rotavirus burden as a second infection occurred in 4% of Mexican infants by 6 months of age and nearly 30% by 1 year of age [45]. However, because the probability to be symptomatic decreases with increasing number of previous infections [45,46] and because recurrent infections occurred at a slower pace [46], we believe with our conservative approach we have not missed many episodes.

Conclusion
Rotavirus vaccination had a substantial and sustained public health impact up to 7 epidemiological years after vaccine introduction, most pronounced in the target age group but with evidence of herd immunity in unvaccinated age groups.