- Email: [email protected]
The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis
Vaccine xxx (2017) xxx–xxx
Contents lists available at ScienceDirect
Vaccine journal homepage: www.elsevier.com/locate/vaccine
The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis Phillip Alexander Kittel M.Sc. Public Health from the University of Southern Denmark, Denmark
a r t i c l e
i n f o
Article history: Received 3 August 2017 Received in revised form 2 November 2017 Accepted 15 November 2017 Available online xxxx Keywords: Rotavirus Rotavirus vaccine Interrupted time-series analysis Intervention effectiveness Segmented regression Impact Incidence Hospitalization Surveillance Quasi-experimental research design
a b s t r a c t Background: Rotavirus is a highly contagious virus causing gastroenteritis, mostly in children under the age of 5. Since 2006, two vaccines are available in Germany. In 2013, these were included into the German national vaccination schedule. The aim of this intervention is to reduce the incidence and hospitalization among children under 5 years caused by rotavirus. The effectiveness of the intervention was analyzed in this study. Methods: National surveillance data of laboratory confirmed rotavirus infections among children under the age of 5 were analyzed using interrupted time-series analysis. Weekly incidence from 2011 to 2017 and monthly hospital discharge rates from 2005 to 2015 were analyzed using a segmented generalized linear model with Poisson distribution. Results: After adjusting for seasonal effects the incidence were approximately 22% (95% CI: 13.2–30.1) lower than expected following the intervention. The hospitalizations were approximately 27% (95% CI: 14.9–39.7) lower than expected following the intervention. The long-term effects of the intervention were nearly zero. The incidence changed in trend by 0.2% (95% CI: 0.1 to ( 0.3)) and the hospitalizations by +0.2% (95% CI: 1.2–( 0.8)) following the intervention. Conclusion: After the inclusion of the vaccines into the national vaccination schedule significant immediate effects of this intervention were found. The weekly incidences and monthly hospitalization caused by the rotavirus were more than 20% lower than expected. The long-term effects of the intervention however were found to be nearly zero. This could be caused by a low vaccination rate in the German population. Ó 2017 Elsevier Ltd. All rights reserved.
1. Introduction Rotaviruses are a common cause of gastroenteritis [8, p. 483]. ‘‘Most infections occur in children under 2 years of age; by 3 years of age more than 90% of the children have been infected (. . .)” [3, p. 624]. The agent is highly contagious, 10 ingested particles are enough to cause an infection [1, pp. 300–301]. In 2006, two vaccines against rotavirus were approved and licensed for the European market [2, pp. 1946–1948]). They have since been available in Germany. An adoption of these vaccines into national vaccination schedules was recommended by the WHO in 2009 [4, pp. 1160–1161]). But it was not until 2013 that the Standing Committee on Vaccinations STIKO) in Germany included these vaccines into the national vaccination schedule for children [16], making them available free of charge. This large scale public health intervention aims at reducing the incidence of infections caused by rotavirus and especially at reducE-mail address: [email protected]
ing the number at hospitalizations among children under the age of 5 [16]. In January 2017, an overview of the rotavirus vaccination rates was published. Overall, the average vaccination rate against rotavirus in 2014 in Germany was 66%. Hence, one third of the children did not receive the vaccine [14, pp. 2–5]. Every intervention needs to be evaluated to estimate whether and to what extend it was effective. To investigate whether this intervention was effective meeting its goals, a retrospective study of observational data was conducted. Since the infection results in vomiting and diarrhea, especially the bodies of infants are exposed to a large amount of stress due to dehydration. The lack of water and electrolytes can be lifethreatening. Currently, there is no antiviral therapy available, only supportive therapy to correct fluid and electrolyte imbalances is in use [8, p. 484]. In the world’s northern hemisphere rotavirus infection follow a strong seasonal pattern with more cases in the cold seasons [4]. Since 2001, rotavirus gastroenteritis is a notifiable disease in Germany. Based on the laboratory confirmations [5, p. 957], a large
https://doi.org/10.1016/j.vaccine.2017.11.041 0264-410X/Ó 2017 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
2
P.A. Kittel / Vaccine xxx (2017) xxx–xxx
surveillance database was established. Therefore, numbers on incidences are available. In 2009, 62,207 cases were reported. 61% 37,822) of these occurred among children under the age of 5 of which 49% 18,621) were hospitalized [12]. Between 2008 and 2012 RVGE was the third most common disease among children <5 years of age [4, p. 1156]). The highest incidence occurred among infants and children under the age of 2 [11, p. 444]. These surveillance data were analyzed using an interrupted time-series analysis, which allows the researcher to draw conclusions on a casual impact of an intervention. A segmented generalized linear model with Poisson distribution was fit to perform the statistical analyses. The aim of this study is to estimate the effectiveness of the public health intervention of including them into the vaccination schedule.
changes [18, pp. 299–300]. The time series must cover the period before and after the interruption [10, p. 39]. Each segment is defined by two parameters, level and trend. The level is the y-axis-intercept at which each segment begins. The trend is the slope inclining, declining or stable) by which the observations change over the period in the segment [10, p. 39]. Changes in level and slope in the segment following a change point indicate that the intervention had an effect on the outcome. Level-changes indicate an immediate effect while slope-changes indicate a gradual/long-term effect [18, p. 300]. Since rotavirus infection follow a seasonal pattern, this needs to be taken into account when calculating the effects of the intervention. The user-written ‘‘circulation” package in STATA was used to adjust for the recurring patterns by using Fourier-terms.
2. Methods
3. Results
2.1. Study design
3.1. Incidence
The aim of vaccination programs is the reduction of incidences. The STIKO stated, that the rotavirus vaccine is recommended because other preventive measures, such as hand sanitizing, are ineffective. Moreover, the reduction of severe cases that lead to hospitalizations is a goal of the intervention [16]. Therefore, the aim of this study was to investigate the effectiveness of the German rotavirus vaccination recommendation in terms of meeting its predefined goals – reducing RVGE incidences and hospital admissions in infants and small children. The core of the study conducted to answer the research question consists of two major parts.
The time series of the incidences was divided into three segments accounting for an impact of the commercialization as well (see Fig. 1 and Table 1). The table and graph above present the results of the regression with three segments. The trend in the pre-commercialization period shows a weekly IRR of 1.0007. Every week the predicted mean rate increases by 0.07%. This result is statistically significant (see Tables 1 and 2). The level change at the date of the commercialization shows an IRR of 1.0884. A level change of approximately +8.84% at the first change point, the commercialization, occurred. This finding is statistically significant. The period following the commercialization and preceding the recommendation has an estimated weekly IRR of 0.9977. This indicates a declining trend in the predicted mean rates by approximately 0.23% per week compared to the previous segment. This finding is also statistically significant.
- (1) The change of incidences prior to and after the intervention is investigated. - (2) The change of hospital admissions prior to and after the intervention is investigated. A population based retrospective observational analysis of the RVGE incidences and hospital admission time-trends was performed. The analyses have been adjusted for seasonal effects by using the user written circulation package in Stata. 2.2. Data collection and variables The secondary quantitative data on RVGE incidences for this study was obtained from the RKI’s national surveillance database. The time span is coded weekly and reaches from 2001 to week 13 of 2017 [15]. Data on the number of hospital admission due to RVGE was provided by the German federal statistics office. The time span is coded monthly and reaches from 2005 to 2015 [17]. 2.3. Statistical analysis A powerful tool for estimating the impact of an intervention/ policy is the ITS analysis. This method is a quasi-experimental research design using observational data. It enables the researcher to draw conclusion on the causal impact of an intervention on an outcome without randomization or a case-control design [6, p. 1]. A time series is a sequence of values of a particular outcome observed continuously and ordered at equally spaced intervals e.g. weeks). This sequence is divided into two or more segments. These are defined by change points, the interruptions. At the change points, the previously established pattern of events is expected to change due to events, such as interventions or policy
Fig. 1. RVGE-cases adjusted, own creation.
Table 1 Results from adjusted analysis of RVGE-cases, own creation. Variables
IRR
p-value
(95% CI)
Trend pre-commercialization Change in level at commercialization Trend pre-recommendation Change in level at recommendation Trend post-recommendation
1.0007 1.0884 0.9977 0.7831 0.9980
0.000 0.011 0.000 0.000 0.000
1.0004 1.0194 0.9973 0.6987 0.9970
1.0010 1.1620 0.9980 0.8776 0.9990
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
3
P.A. Kittel / Vaccine xxx (2017) xxx–xxx
At the second change point, the STIKO recommendation, another level change was found. The IRR of 0.7831 shows that the level of the predicted mean rates was at approximately 78.31% of the previous predicted mean rates. This matches a drop of 21.69% of the predicted mean rates. This result is also statistically significant. Lastly, for the post-recommendation period a weekly IRR of 0.9980 was estimated. This result indicates, that the trend of the predicted mean rates in this period declines by approximately 0.20% compared to the pre-recommendation period. However, this result is not statistically significant. What is indicated by these findings is, that the slopes in each segment differ. The first segment shows an incline, the second segment shows a decline and the third segment shows a decline as well. Moreover, the levels of the predicted mean rates seem to be dropping at each change point. 3.2. Hospitalization An analysis of the hospitalization data was performed adjusting for seasonality. Again, a generalized linear model with Poisson distribution was used (see Table 2 and Fig. 2). The table and graph above show the results from the second analysis. The trend in the pre-commercialization is shown by the
weekly IRR of 1.0135. This indicates a monthly incline of the predicted mean rate by approximately 1.35%. This result is statistically significant. The first level change at the date of the commercialization is revealed by the IRR of 0.9469. This result shows a level drop of approximately 5.31% of the predicted mean rate. This result is not statistically significant since the p value is larger than 0.05 and the CI includes the value 1. The pre-recommendation period has an estimated monthly IRR of 0.9813 which indicates a declining trend. The monthly predicted mean rate decreased by approximately 1.87% compared to the pre-commercialization period. This finding is also statistically significant. The second level change at the date of the recommendation is displayed by the IRR of 0.7267. A level drop of approximately 27,33% of the predicted mean rates occurred. This finding is statistically significant. Lastly, the post-recommendation periods’ monthly IRR is 1.0022. The change trend in the last segment is estimated +0.22%. Since this must be compared to the previous segment, the trend might still be declining but less steep than before. This finding is not statistically significant, since the p-value is larger than 0.05 and the CI includes the value 1. 3.3. Summary of the findings
Table 2 Results from adjusted analysis of hospitalized RVGE-cases, own creation. Variables
IRR
p-value
(95% CI)
Trend pre-commercialization Change in level at commercialization Trend pre-recommendation Change in level at recommendation Trend post-recommendation
1.0135 0.9469 0.9813 0.7267 1.0022
0.005 0.338 0.000 0.000 0.680
1.0041 0.8469 0.9721 0.6131 0.9918
1.0230 1.0587 0.9905 0.8614 1.0127
Fig. 2. Hospitalized RVGE-cases adjusted, own creation.
Both regression findings will be used to answer this research question. The table below summarizes the findings from this chapter. As described in the theory of ITS analyses, changes in level indicate an immediate effect of an intervention, while changes in slope/trend indicate long-term effects. In the case of the STIKO recommendation of routine rotavirus vaccinations both, level and slope-changes are present. In Table 3 the findings of this study are shown. Most important is the estimated percentage. The immediate impact of the recommendation was found in both analyses. The level of predicted mean rates in the incidences as well as in the hospitalizations dropped by more than -20%. Both findings are statistically significant. Having adjusted for seasonality and excluded its influence, a causal impact of the recommendation is very likely since other factors are not present. The estimated change in slope in the incidence analysis was marginal with 0.2%. This finding is statistically significant. This finding indicates that a long-term effect of the recommendation is technically present but limited. The estimated change in trend in the hospitalization was positive by +0.22%. This finding indicates that the recommendation mitigates the declining slope from the previous segment. This finding is not statistically significant. Given the significance of the change in trend from the incidence analysis, it can be assumed that the overall long-term effect of the recommendation is nearly null. It must be kept in mind, that the trends in the pre-recommendation and the post-recommendation were declining.
Table 3 Summary of results, own creation. Results
IRR
p-value
CI (95%)
RVGE-cases per 100,000 Change in level following the recommendation Change in trend following the recommendation
Percentage (approx.)
0.7831 0.9980
0.000 0.000
(0.6987–0.8776) (0.9970–0.9990)
21.69% 0.20%
Hospitalized RVGE-cases per 100,000 Change in level following the recommendation Change in trend following the recommendation
0.7267 1.0022
0.000 0.680
(0.6131–0.8614) (0.9918–1.0127)
27.33% +0.22%
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
4
P.A. Kittel / Vaccine xxx (2017) xxx–xxx
All in all, the STIKO recommendation was found to be effective in meeting its predefined goals. A reduction of RVGE incidences and hospitalizations was found. But since the long-term effect of the intervention is nearly non-existent, the reasons for this will have to be discussed.
Furthermore, a thread to the validity of the results of an interrupted time series analysis is that another event occurred at the same time of the intervention. This thread cannot be fully eliminated but since the intervals in this study are rather narrow, bias is unlikely.
4. Discussion
5. Conclusion
From the study conducted, a moderate immediate effect of the recommendation was estimated. The fact, that the rotavirus vaccines were included into the official German vaccination schedule after the STIKO statement, lead to a significant reduction of RVGE incidences and severe infections resulting in hospital care among children under the age of 5. A reduction by more than 20% of the predicted mean weekly/monthly cases is a big progress in the prevention of the disease. The intervention can therefore be seen as effective. The predefined goals were met. As outlined, the long-term effects however are marginal, almost non-existent. The trends of both, incidence and hospitalizations have been decreasing since the commercialization of the rotavirus vaccine. This trend was only minimally influenced. As presented in the introduction, one third of the statutory insured children born in 2014 in Germany did not receive the vaccine. This might be a reason for the marginal long-term effect of the recommendation. A reason for the low uptake could be a phenomenon which can be observed in Germany and elsewhere). Skepticism towards vaccinations. Starting in the late 1990’s with false findings in a study investigating the risk of autism caused by vaccines, this fire has been fueled ever since [9]. The number of people who believe that vaccines cause more harm than they benefit is increasing. Physicians in Germany are obliged to inform patients (and in this case their parents) about the risk but also the benefits of each vaccine. Hence, the reason for the low uptake must partly be traced back to the parents. Targeted information campaigns about the specific pathogen and the vaccine could be helpful to increase the acceptance in the population. Another explanation for the marginal long-term effect might be the length of the time series itself. The average weekly number of cases is steadily decreasing and the closer the number of cases is decreasing towards zero, the lower the decline will be. Eventually, the decline has to stall, since a natural disease frequency will be reached. Generally, the quality of data can be described as adequate. The surveillance system for rotavirus infection in Germany is based on laboratory confirmed cases. Laboratories, physicians and hospitals are obliged to report every confirmed case. Moreover, a lab test is mandatory if one of the typical symptoms diarrhea or vomiting) is present [13]. Conducting a regression with Poisson distribution, two model assumptions must be tested. The first assumption is that the variance equals the expected count. Real data, as the ones used in this study, often shows a greater variance than the expected counts, biasing the standard error estimates. This is called overdispersion. The model can be modified to this by adjusting the scale [7]. When allowing for over-dispersion in regressions, the estimated IRR would stay the same. The CIs however would widen influencing the significance of the findings. The second assumption of the model is the autocorrelation, meaning that the observations are independent. In time series, this is often not fulfilled because observations which are successive are usually more similar to each other than two observations which are further apart. Seasonal changes explain most of the autocorrelation [7]. In this case, evidence of autocorrelation was found. Adjusting for seasonality mitigated the autocorrelation.
The goal of this public health intervention was to lower to the incidences and hospitalization cause by RVGE in children under the age of 5. As the findings of the study conducted show, the STIKO recommendation was an effective intervention. It does show a weak spot though. Its long-term effect is nearly zero, which is most likely due to low vaccine uptake in Germany. Disease surveillance is the key to taking tailored actions for prevention in public health and analyze the effect of such action. As this study shows, it allows the precise estimation of effects and allows researchers and policy makers to identify trends and recommend actions. Finally, the following recommendation based on this study and its findings can be given. Low vaccination rates can be combatted by launching targeted information campaigns to promote vaccines. Providing uniform information to parents in a timely manner about the vaccine which is next on their child’s schedule could encourage them to consider the vaccination. Boosting the longterm effect of the intervention is needed to tap its full potential.
Supplementary Data Supplementary data associated with the study presented can be obtained from the SurvStat website by the Robert-Koch-Institute and requested from the German Federal Statistics Office. Acknowledgment The author would like to thank the German Federal Statistics Office for the assistance provided in the data collection process. References [1] Goering R et al. Mims’ Medical Microbiology. 4th ed. Philadelphia: Elsevier; 2008. [2] Greenberg HB, Estes MK. Rotaviruses: from pathogenesis to vaccination. Gastroenterology 2009;136(6):1939–51. [3] Greenwood D, Slack R, Peutherer J. Medical Microbiology. 14th ed. Edinburgh: Churchill Livingstine; 1992. [4] Koch J, Rodeck B. Rotaviruserkrankung und ihre Impfprävention in Deutschland. Monatsschrift Kinderheilkunde 2013;161(12):1154–62. [5] Koch J et al. Background paper to the recommendation for routine rotavirus vaccination of infants in Germany. Bundesgesundheitsblatt 2013;56:957–84. [6] Kontopantelis E et al. Regression based quasi-experimental approach when randomisation is not an option: interrupted time series analysis. BMJ - Res Methods Report 2015:1–4. [7] Lopez Bernal J, Cummins S, Gasparrini A. Interrupted time series regression for the evaluation of public health interventions: a tutorial. Int J Epidemiol 2016:1–8. [8] Murray P, Rosenthal K, Kobayashi G, Pfaller M. Medical Microbiology. 3rd ed. St. Louis: Mosby; 1998. [9] Nature Publishing Group. Acase of junk science, conflict and hype. Nature Immunol 2008;9(12):1. [10] Penfold RB, Zhang F. Use of interrupted time series analysis in evaluating health care quality improvements. Acad Pediat 2013;13(6):38–44. [11] Robert-Koch-Institut, Epidemiologie der Rotavirus-Erkrankungen in Deutschland im Zeitraum von 2001 bis 2011. Epidemiologisches Bul. Issue 44 2012 441–452. [12] Robert-Koch-Institut, Rotaviren-Gastroenteritis 2013a. [Online] Available at: [accessed 9 February 2017]. [13] Robert-Koch-Institut, Falldefinitionen des Robert-Koch-Instituts zur Übermittlung von Erkrankungs- und Todesfällen und Nachweisen von Krankheitserregern. 2015 ed. Berlin: Robert-Koch-Institut. 2015.
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
P.A. Kittel / Vaccine xxx (2017) xxx–xxx [14] Robert-Koch-Institut, Imfpquoten der Rotavirus-, Masern-, HPV- und Influenza-Impfung in Deutschland. Epidemiologisches Bulletin, Issue 1 2017b, p. 1–16. [15] Robert-Koch-Institut, SurvStat@RKI 2.0,, Berlin: Robert-Koch-Institut. 2017c. [16] Impfkommission Ständige. Empfehlung zur Rotavirus-Standardimpfung von Säuglingen in Deutschland. Bundesgesundheitsblatt 2013(Issue 7):955–6.
5
[17] Statistisches Bundesamt (Destatis), DRG-Statistik, Wiesbaden: Statistisches Bundesamt (Destatis) 2017. [18] Wagner AK, Soumerai SB, Zhang F, Ross-Degnan D. Segmented regression analysis of interrupted time series studies in medication use research. J Clin Pharmacy Therap 2002;27:299–309.
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
Contents lists available at ScienceDirect
Vaccine journal homepage: www.elsevier.com/locate/vaccine
The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis Phillip Alexander Kittel M.Sc. Public Health from the University of Southern Denmark, Denmark
a r t i c l e
i n f o
Article history: Received 3 August 2017 Received in revised form 2 November 2017 Accepted 15 November 2017 Available online xxxx Keywords: Rotavirus Rotavirus vaccine Interrupted time-series analysis Intervention effectiveness Segmented regression Impact Incidence Hospitalization Surveillance Quasi-experimental research design
a b s t r a c t Background: Rotavirus is a highly contagious virus causing gastroenteritis, mostly in children under the age of 5. Since 2006, two vaccines are available in Germany. In 2013, these were included into the German national vaccination schedule. The aim of this intervention is to reduce the incidence and hospitalization among children under 5 years caused by rotavirus. The effectiveness of the intervention was analyzed in this study. Methods: National surveillance data of laboratory confirmed rotavirus infections among children under the age of 5 were analyzed using interrupted time-series analysis. Weekly incidence from 2011 to 2017 and monthly hospital discharge rates from 2005 to 2015 were analyzed using a segmented generalized linear model with Poisson distribution. Results: After adjusting for seasonal effects the incidence were approximately 22% (95% CI: 13.2–30.1) lower than expected following the intervention. The hospitalizations were approximately 27% (95% CI: 14.9–39.7) lower than expected following the intervention. The long-term effects of the intervention were nearly zero. The incidence changed in trend by 0.2% (95% CI: 0.1 to ( 0.3)) and the hospitalizations by +0.2% (95% CI: 1.2–( 0.8)) following the intervention. Conclusion: After the inclusion of the vaccines into the national vaccination schedule significant immediate effects of this intervention were found. The weekly incidences and monthly hospitalization caused by the rotavirus were more than 20% lower than expected. The long-term effects of the intervention however were found to be nearly zero. This could be caused by a low vaccination rate in the German population. Ó 2017 Elsevier Ltd. All rights reserved.
1. Introduction Rotaviruses are a common cause of gastroenteritis [8, p. 483]. ‘‘Most infections occur in children under 2 years of age; by 3 years of age more than 90% of the children have been infected (. . .)” [3, p. 624]. The agent is highly contagious, 10 ingested particles are enough to cause an infection [1, pp. 300–301]. In 2006, two vaccines against rotavirus were approved and licensed for the European market [2, pp. 1946–1948]). They have since been available in Germany. An adoption of these vaccines into national vaccination schedules was recommended by the WHO in 2009 [4, pp. 1160–1161]). But it was not until 2013 that the Standing Committee on Vaccinations STIKO) in Germany included these vaccines into the national vaccination schedule for children [16], making them available free of charge. This large scale public health intervention aims at reducing the incidence of infections caused by rotavirus and especially at reducE-mail address: [email protected]
ing the number at hospitalizations among children under the age of 5 [16]. In January 2017, an overview of the rotavirus vaccination rates was published. Overall, the average vaccination rate against rotavirus in 2014 in Germany was 66%. Hence, one third of the children did not receive the vaccine [14, pp. 2–5]. Every intervention needs to be evaluated to estimate whether and to what extend it was effective. To investigate whether this intervention was effective meeting its goals, a retrospective study of observational data was conducted. Since the infection results in vomiting and diarrhea, especially the bodies of infants are exposed to a large amount of stress due to dehydration. The lack of water and electrolytes can be lifethreatening. Currently, there is no antiviral therapy available, only supportive therapy to correct fluid and electrolyte imbalances is in use [8, p. 484]. In the world’s northern hemisphere rotavirus infection follow a strong seasonal pattern with more cases in the cold seasons [4]. Since 2001, rotavirus gastroenteritis is a notifiable disease in Germany. Based on the laboratory confirmations [5, p. 957], a large
https://doi.org/10.1016/j.vaccine.2017.11.041 0264-410X/Ó 2017 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
2
P.A. Kittel / Vaccine xxx (2017) xxx–xxx
surveillance database was established. Therefore, numbers on incidences are available. In 2009, 62,207 cases were reported. 61% 37,822) of these occurred among children under the age of 5 of which 49% 18,621) were hospitalized [12]. Between 2008 and 2012 RVGE was the third most common disease among children <5 years of age [4, p. 1156]). The highest incidence occurred among infants and children under the age of 2 [11, p. 444]. These surveillance data were analyzed using an interrupted time-series analysis, which allows the researcher to draw conclusions on a casual impact of an intervention. A segmented generalized linear model with Poisson distribution was fit to perform the statistical analyses. The aim of this study is to estimate the effectiveness of the public health intervention of including them into the vaccination schedule.
changes [18, pp. 299–300]. The time series must cover the period before and after the interruption [10, p. 39]. Each segment is defined by two parameters, level and trend. The level is the y-axis-intercept at which each segment begins. The trend is the slope inclining, declining or stable) by which the observations change over the period in the segment [10, p. 39]. Changes in level and slope in the segment following a change point indicate that the intervention had an effect on the outcome. Level-changes indicate an immediate effect while slope-changes indicate a gradual/long-term effect [18, p. 300]. Since rotavirus infection follow a seasonal pattern, this needs to be taken into account when calculating the effects of the intervention. The user-written ‘‘circulation” package in STATA was used to adjust for the recurring patterns by using Fourier-terms.
2. Methods
3. Results
2.1. Study design
3.1. Incidence
The aim of vaccination programs is the reduction of incidences. The STIKO stated, that the rotavirus vaccine is recommended because other preventive measures, such as hand sanitizing, are ineffective. Moreover, the reduction of severe cases that lead to hospitalizations is a goal of the intervention [16]. Therefore, the aim of this study was to investigate the effectiveness of the German rotavirus vaccination recommendation in terms of meeting its predefined goals – reducing RVGE incidences and hospital admissions in infants and small children. The core of the study conducted to answer the research question consists of two major parts.
The time series of the incidences was divided into three segments accounting for an impact of the commercialization as well (see Fig. 1 and Table 1). The table and graph above present the results of the regression with three segments. The trend in the pre-commercialization period shows a weekly IRR of 1.0007. Every week the predicted mean rate increases by 0.07%. This result is statistically significant (see Tables 1 and 2). The level change at the date of the commercialization shows an IRR of 1.0884. A level change of approximately +8.84% at the first change point, the commercialization, occurred. This finding is statistically significant. The period following the commercialization and preceding the recommendation has an estimated weekly IRR of 0.9977. This indicates a declining trend in the predicted mean rates by approximately 0.23% per week compared to the previous segment. This finding is also statistically significant.
- (1) The change of incidences prior to and after the intervention is investigated. - (2) The change of hospital admissions prior to and after the intervention is investigated. A population based retrospective observational analysis of the RVGE incidences and hospital admission time-trends was performed. The analyses have been adjusted for seasonal effects by using the user written circulation package in Stata. 2.2. Data collection and variables The secondary quantitative data on RVGE incidences for this study was obtained from the RKI’s national surveillance database. The time span is coded weekly and reaches from 2001 to week 13 of 2017 [15]. Data on the number of hospital admission due to RVGE was provided by the German federal statistics office. The time span is coded monthly and reaches from 2005 to 2015 [17]. 2.3. Statistical analysis A powerful tool for estimating the impact of an intervention/ policy is the ITS analysis. This method is a quasi-experimental research design using observational data. It enables the researcher to draw conclusion on the causal impact of an intervention on an outcome without randomization or a case-control design [6, p. 1]. A time series is a sequence of values of a particular outcome observed continuously and ordered at equally spaced intervals e.g. weeks). This sequence is divided into two or more segments. These are defined by change points, the interruptions. At the change points, the previously established pattern of events is expected to change due to events, such as interventions or policy
Fig. 1. RVGE-cases adjusted, own creation.
Table 1 Results from adjusted analysis of RVGE-cases, own creation. Variables
IRR
p-value
(95% CI)
Trend pre-commercialization Change in level at commercialization Trend pre-recommendation Change in level at recommendation Trend post-recommendation
1.0007 1.0884 0.9977 0.7831 0.9980
0.000 0.011 0.000 0.000 0.000
1.0004 1.0194 0.9973 0.6987 0.9970
1.0010 1.1620 0.9980 0.8776 0.9990
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
3
P.A. Kittel / Vaccine xxx (2017) xxx–xxx
At the second change point, the STIKO recommendation, another level change was found. The IRR of 0.7831 shows that the level of the predicted mean rates was at approximately 78.31% of the previous predicted mean rates. This matches a drop of 21.69% of the predicted mean rates. This result is also statistically significant. Lastly, for the post-recommendation period a weekly IRR of 0.9980 was estimated. This result indicates, that the trend of the predicted mean rates in this period declines by approximately 0.20% compared to the pre-recommendation period. However, this result is not statistically significant. What is indicated by these findings is, that the slopes in each segment differ. The first segment shows an incline, the second segment shows a decline and the third segment shows a decline as well. Moreover, the levels of the predicted mean rates seem to be dropping at each change point. 3.2. Hospitalization An analysis of the hospitalization data was performed adjusting for seasonality. Again, a generalized linear model with Poisson distribution was used (see Table 2 and Fig. 2). The table and graph above show the results from the second analysis. The trend in the pre-commercialization is shown by the
weekly IRR of 1.0135. This indicates a monthly incline of the predicted mean rate by approximately 1.35%. This result is statistically significant. The first level change at the date of the commercialization is revealed by the IRR of 0.9469. This result shows a level drop of approximately 5.31% of the predicted mean rate. This result is not statistically significant since the p value is larger than 0.05 and the CI includes the value 1. The pre-recommendation period has an estimated monthly IRR of 0.9813 which indicates a declining trend. The monthly predicted mean rate decreased by approximately 1.87% compared to the pre-commercialization period. This finding is also statistically significant. The second level change at the date of the recommendation is displayed by the IRR of 0.7267. A level drop of approximately 27,33% of the predicted mean rates occurred. This finding is statistically significant. Lastly, the post-recommendation periods’ monthly IRR is 1.0022. The change trend in the last segment is estimated +0.22%. Since this must be compared to the previous segment, the trend might still be declining but less steep than before. This finding is not statistically significant, since the p-value is larger than 0.05 and the CI includes the value 1. 3.3. Summary of the findings
Table 2 Results from adjusted analysis of hospitalized RVGE-cases, own creation. Variables
IRR
p-value
(95% CI)
Trend pre-commercialization Change in level at commercialization Trend pre-recommendation Change in level at recommendation Trend post-recommendation
1.0135 0.9469 0.9813 0.7267 1.0022
0.005 0.338 0.000 0.000 0.680
1.0041 0.8469 0.9721 0.6131 0.9918
1.0230 1.0587 0.9905 0.8614 1.0127
Fig. 2. Hospitalized RVGE-cases adjusted, own creation.
Both regression findings will be used to answer this research question. The table below summarizes the findings from this chapter. As described in the theory of ITS analyses, changes in level indicate an immediate effect of an intervention, while changes in slope/trend indicate long-term effects. In the case of the STIKO recommendation of routine rotavirus vaccinations both, level and slope-changes are present. In Table 3 the findings of this study are shown. Most important is the estimated percentage. The immediate impact of the recommendation was found in both analyses. The level of predicted mean rates in the incidences as well as in the hospitalizations dropped by more than -20%. Both findings are statistically significant. Having adjusted for seasonality and excluded its influence, a causal impact of the recommendation is very likely since other factors are not present. The estimated change in slope in the incidence analysis was marginal with 0.2%. This finding is statistically significant. This finding indicates that a long-term effect of the recommendation is technically present but limited. The estimated change in trend in the hospitalization was positive by +0.22%. This finding indicates that the recommendation mitigates the declining slope from the previous segment. This finding is not statistically significant. Given the significance of the change in trend from the incidence analysis, it can be assumed that the overall long-term effect of the recommendation is nearly null. It must be kept in mind, that the trends in the pre-recommendation and the post-recommendation were declining.
Table 3 Summary of results, own creation. Results
IRR
p-value
CI (95%)
RVGE-cases per 100,000 Change in level following the recommendation Change in trend following the recommendation
Percentage (approx.)
0.7831 0.9980
0.000 0.000
(0.6987–0.8776) (0.9970–0.9990)
21.69% 0.20%
Hospitalized RVGE-cases per 100,000 Change in level following the recommendation Change in trend following the recommendation
0.7267 1.0022
0.000 0.680
(0.6131–0.8614) (0.9918–1.0127)
27.33% +0.22%
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
4
P.A. Kittel / Vaccine xxx (2017) xxx–xxx
All in all, the STIKO recommendation was found to be effective in meeting its predefined goals. A reduction of RVGE incidences and hospitalizations was found. But since the long-term effect of the intervention is nearly non-existent, the reasons for this will have to be discussed.
Furthermore, a thread to the validity of the results of an interrupted time series analysis is that another event occurred at the same time of the intervention. This thread cannot be fully eliminated but since the intervals in this study are rather narrow, bias is unlikely.
4. Discussion
5. Conclusion
From the study conducted, a moderate immediate effect of the recommendation was estimated. The fact, that the rotavirus vaccines were included into the official German vaccination schedule after the STIKO statement, lead to a significant reduction of RVGE incidences and severe infections resulting in hospital care among children under the age of 5. A reduction by more than 20% of the predicted mean weekly/monthly cases is a big progress in the prevention of the disease. The intervention can therefore be seen as effective. The predefined goals were met. As outlined, the long-term effects however are marginal, almost non-existent. The trends of both, incidence and hospitalizations have been decreasing since the commercialization of the rotavirus vaccine. This trend was only minimally influenced. As presented in the introduction, one third of the statutory insured children born in 2014 in Germany did not receive the vaccine. This might be a reason for the marginal long-term effect of the recommendation. A reason for the low uptake could be a phenomenon which can be observed in Germany and elsewhere). Skepticism towards vaccinations. Starting in the late 1990’s with false findings in a study investigating the risk of autism caused by vaccines, this fire has been fueled ever since [9]. The number of people who believe that vaccines cause more harm than they benefit is increasing. Physicians in Germany are obliged to inform patients (and in this case their parents) about the risk but also the benefits of each vaccine. Hence, the reason for the low uptake must partly be traced back to the parents. Targeted information campaigns about the specific pathogen and the vaccine could be helpful to increase the acceptance in the population. Another explanation for the marginal long-term effect might be the length of the time series itself. The average weekly number of cases is steadily decreasing and the closer the number of cases is decreasing towards zero, the lower the decline will be. Eventually, the decline has to stall, since a natural disease frequency will be reached. Generally, the quality of data can be described as adequate. The surveillance system for rotavirus infection in Germany is based on laboratory confirmed cases. Laboratories, physicians and hospitals are obliged to report every confirmed case. Moreover, a lab test is mandatory if one of the typical symptoms diarrhea or vomiting) is present [13]. Conducting a regression with Poisson distribution, two model assumptions must be tested. The first assumption is that the variance equals the expected count. Real data, as the ones used in this study, often shows a greater variance than the expected counts, biasing the standard error estimates. This is called overdispersion. The model can be modified to this by adjusting the scale [7]. When allowing for over-dispersion in regressions, the estimated IRR would stay the same. The CIs however would widen influencing the significance of the findings. The second assumption of the model is the autocorrelation, meaning that the observations are independent. In time series, this is often not fulfilled because observations which are successive are usually more similar to each other than two observations which are further apart. Seasonal changes explain most of the autocorrelation [7]. In this case, evidence of autocorrelation was found. Adjusting for seasonality mitigated the autocorrelation.
The goal of this public health intervention was to lower to the incidences and hospitalization cause by RVGE in children under the age of 5. As the findings of the study conducted show, the STIKO recommendation was an effective intervention. It does show a weak spot though. Its long-term effect is nearly zero, which is most likely due to low vaccine uptake in Germany. Disease surveillance is the key to taking tailored actions for prevention in public health and analyze the effect of such action. As this study shows, it allows the precise estimation of effects and allows researchers and policy makers to identify trends and recommend actions. Finally, the following recommendation based on this study and its findings can be given. Low vaccination rates can be combatted by launching targeted information campaigns to promote vaccines. Providing uniform information to parents in a timely manner about the vaccine which is next on their child’s schedule could encourage them to consider the vaccination. Boosting the longterm effect of the intervention is needed to tap its full potential.
Supplementary Data Supplementary data associated with the study presented can be obtained from the SurvStat website by the Robert-Koch-Institute and requested from the German Federal Statistics Office. Acknowledgment The author would like to thank the German Federal Statistics Office for the assistance provided in the data collection process. References [1] Goering R et al. Mims’ Medical Microbiology. 4th ed. Philadelphia: Elsevier; 2008. [2] Greenberg HB, Estes MK. Rotaviruses: from pathogenesis to vaccination. Gastroenterology 2009;136(6):1939–51. [3] Greenwood D, Slack R, Peutherer J. Medical Microbiology. 14th ed. Edinburgh: Churchill Livingstine; 1992. [4] Koch J, Rodeck B. Rotaviruserkrankung und ihre Impfprävention in Deutschland. Monatsschrift Kinderheilkunde 2013;161(12):1154–62. [5] Koch J et al. Background paper to the recommendation for routine rotavirus vaccination of infants in Germany. Bundesgesundheitsblatt 2013;56:957–84. [6] Kontopantelis E et al. Regression based quasi-experimental approach when randomisation is not an option: interrupted time series analysis. BMJ - Res Methods Report 2015:1–4. [7] Lopez Bernal J, Cummins S, Gasparrini A. Interrupted time series regression for the evaluation of public health interventions: a tutorial. Int J Epidemiol 2016:1–8. [8] Murray P, Rosenthal K, Kobayashi G, Pfaller M. Medical Microbiology. 3rd ed. St. Louis: Mosby; 1998. [9] Nature Publishing Group. Acase of junk science, conflict and hype. Nature Immunol 2008;9(12):1. [10] Penfold RB, Zhang F. Use of interrupted time series analysis in evaluating health care quality improvements. Acad Pediat 2013;13(6):38–44. [11] Robert-Koch-Institut, Epidemiologie der Rotavirus-Erkrankungen in Deutschland im Zeitraum von 2001 bis 2011. Epidemiologisches Bul. Issue 44 2012 441–452. [12] Robert-Koch-Institut, Rotaviren-Gastroenteritis 2013a. [Online] Available at:
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041
P.A. Kittel / Vaccine xxx (2017) xxx–xxx [14] Robert-Koch-Institut, Imfpquoten der Rotavirus-, Masern-, HPV- und Influenza-Impfung in Deutschland. Epidemiologisches Bulletin, Issue 1 2017b, p. 1–16. [15] Robert-Koch-Institut, SurvStat@RKI 2.0,
5
[17] Statistisches Bundesamt (Destatis), DRG-Statistik, Wiesbaden: Statistisches Bundesamt (Destatis) 2017. [18] Wagner AK, Soumerai SB, Zhang F, Ross-Degnan D. Segmented regression analysis of interrupted time series studies in medication use research. J Clin Pharmacy Therap 2002;27:299–309.
Please cite this article in press as: Kittel PA. The impact of the recommendation of routine rotavirus vaccination in Germany: An interrupted time-series analysis. Vaccine (2017), https://doi.org/10.1016/j.vaccine.2017.11.041