This is the first study estimating the spatio-temporal impact of RV vaccination on RV and AGE hospitalisations. The number of averted hospitalisations by RV vaccination was increasing in space and time in the Valencia Region during the study period in children <3 years. Overall, ~1866 hospital admissions for AGE (potentially attributable to RV) were averted during 2007–2016. Despite the low-medium vaccine coverage (~50%) in 2015-2016, relevant 36% and 20% reductions were estimated in RV and AGE hospitalisations respectively. It should be noted that ~8606 hospitalisations would have possibly been averted during the whole study period if all children had been vaccinated. Direct benefits of vaccination were observed in the reduction of hospitalisation rates for RV (86%) and GEA (47%) in vaccinated children. These results are in accordance with the vaccine effectiveness estimated in the Valencia Region previously (9). Regarding the spatio-temporal results, substantial variability was seen in RV vaccine coverage and hospitalisation risk for RV and AGE among health departments and health care districts. Spatio-temporal clusters were clearly distinguished. These patterns could be explained by climatic, environmental, sociodemographic, or economic differences, or by the different admission policies of health departments.
Although other impact studies reported relevant reductions in both RV and AGE hospitalisations in children <5 years following RV vaccination (4), (6), (6) (31), (32), (7), (13) , (14), (33), only two of them showed a coverage-dependent response (8), (34). Moreover, many of them were time-trend ecological studies comparing hospitalisation data in pre and post-vaccine populations and a historical pre-vaccine group (7), (13) , (14), (33). Even though this is the most commonly used method, it has been associated with potential confusion bias (15), (16). The reported impact of the vaccination could be due to other secular trends caused by, changes in reporting, in medical practices, in health seeking behaviour, etc (35). Besides, vaccine impact based on hospitalisation data is prone to confounding, because hospitalisations rates are closely related to changes in the quality, access and use of the health care system which often occur simultaneously with introduction of new vaccines (17).
On the other hand, few spatial and spatio-temporal models have studied RV and AGE dynamics and none of them included the vaccination status of the population. Spatial variation in RV hospitalisations explained by sociodemographic characteristics of the population has previously been shown in studies conducted in Germany and New Zealand (23), (24). Other studies in the USA and Brazil found that spatio-temporal variation in birth rate can lead to secular changes in the RV pattern (21), (25). Finally, a study conducted in Bhutan showed that rainfall and temperature explain much of the spatio-temporal dynamics of diarrhoea (possibly due to RV infection in approximately 23% of cases) (31). The studies developed in Germany and New Zealand were based in aggregated data over time, however, caution should be taken when interpreting this analysis because the area-specific risk may be overestimated or underestimated. Furthermore, none of these standard models considered spatio-temporal dependence; however, what occurs in a health care district is intimately related to what occurs in the adjacent one and is also related to what happened previously (36).
The present study analysed the impact of RV vaccination on RV and AGE hospitalisations from a different point of view. We developed a sophisticated spatio-temporal model that allowed us to estimate the RV vaccination impact in terms of adverted hospitalisations according to the number of children vaccinated. The spatio-temporal approach improves the commonly used methodologies to estimate the RV vaccine impact and its interpretation as follows. First of all, this analysis showed the evolution of the impact of RV vaccination and the risk of hospitalisation for RV/AGE in the Valencia Region at the health care district level over time. Second, adjusting by spatial variables such us health care district and health department in the analysis, several potentially attributable biases can be controlled. Those biases could have been caused by economic inequalities, environmental factors, socio-demographic differences or even possible changes in hospitalisations-admission policies (37) (38) (21) (39). Moreover, the hospitalisation rate for any cause of each health care district was included to adjust the confusion caused by hospital attraction or other secular trends (17). Finally, the Bayesian approach used allowed us to adequately capture dependencies among health areas and the potential relationship of data over time that cannot be easily modelled in classical statistics (40) (41).
Nevertheless, some limitations of our study should be highlighted. First of all, RV vaccines are not included in the official immunisation schedule, which may suggest differences between rotavirus vaccinees and non-vaccinees in terms of socioeconomic conditions and health-seeking behaviour. Therefore, socioeconomic factors might be an important confounder of our results and admissions at private hospitals should also be considered in future studies.
Secondly, although the positive predictive value of the rotavirus ICD-9-CM code identifying acute gastroenteritis attributable to rotavirus using MBDS resulted in 90% (9), different immunochromatographic methods with different sensitivities and specificities could have been used in the different hospitals during the study period (42). In fact, based on the difference found in the number of hospitalisations prevented for AGE and RV (1866 vs. 1142), ~40% of underdiagnosis in RV hospitalisations was detected in the present study. Thirdly, health care district and health department could have varied over time; but only the last updated information was available. Fourthly, children who were unable to receive RV vaccines according to manufacturer recommendations (i.e. immunocompromised children) were not excluded from the analysis due to the lack of information.
Finally, it should be noted that both vaccines (RV1 and RV5) were used concurrently until 2010. But, RV5 was the only rotavirus vaccine available in Spain between 2010 and 2016. Therefore, results will have a limited value for estimating the impact of RV1.