Background
Since the commitment to eradicate poliomyelitis in 1988 there has been a substantial reduction in the number of cases of poliomyelitis, and poliovirus has been eliminated from a majority of countries. India had successfully eliminated wild poliomyelitis in 2014 [
1], resulting in considerable reduction in risk within the local population and neighbouring countries. Following an absence of cases for over 2 years, wild poliomyelitis was detected in Borno State, Nigeria, in August 2016, where genetic sequencing of the virus suggests undetected transmission since at least 2013. Endemic transmission continues in Pakistan and Afghanistan, but here the incidence of wild poliomyelitis cases has reached an all-time low [
1].
Successful elimination of wild poliomyelitis from India and substantial reduction in incidence in Nigeria was the result of considerable efforts to immunise large numbers of children each year, primarily through use of supplementary immunisation activities (SIAs). Transmission of poliomyelitis in Nigeria and India prior to this time resulted in several outbreaks of poliomyelitis in countries previously free from disease. The large number of outbreaks, resulting in the re-establishment of transmission in several countries, was a cause of great concern [
2]. Since this time there has been a refocus in resources and planning of preventive SIAs, and development of country-specific response plans [
3].
A previous analysis of wild poliomyelitis outbreaks had identified that population immunity, recent outbreaks, extent of travel from infected countries and factors such as the percentage of the population aged less than 15 years were associated with poliomyelitis outbreaks [
4]. This analysis set out a framework where preventive SIAs could be planned based on inference from a statistical model that captured the different contributions of population immunity, migration from affected countries and additional factors, rather than relying solely on expert assessment of risk. To this end, a multi-agency team representing partners of the Global Polio Eradication Initiative was established in 2014 to support SIA planning by providing an assessment of wild poliomyelitis risk [
5]. In addition to this international assessment of risk, WHO regional offices make sub-national assessments of poliomyelitis risk [
6].
An inevitable consequence associated with the use of the live attenuated oral polio vaccine (OPV) is the emergence of vaccine-derived polioviruses (VDPVs), which in some situations has led to outbreaks of poliomyelitis [
7]. The first outbreak of VDPV was reported in Hispaniola in 2000. Several studies have explored the mechanisms leading to VDPV emergence; the attenuated poliovirus strains used to produce the OPV naturally mutate and these revertants with increased neurovirulence are naturally selected during replication in the gut of OPV recipients (sometimes resulting in cases of vaccine-associated paralytic poliomyelitis [
8]). Should this occur in areas of low population immunity the revertants may spread within populations, resulting in cases of paralytic poliomyelitis where the isolated virus may have >1% nucleotide sequence divergence (0.6% for serotype 2) from the original Sabin strain, indicating ongoing circulation [
7]. Within Nigeria, districts with low routine immunisation coverage and a high number of births were associated with a high probability of VDPV emergence and circulation [
9]. To eliminate the risks posed by circulating VDPVs (cVDPVs) of serotype 2, OPV of this serotype has been withdrawn from immunisation activities through a globally synchronised replacement of all trivalent OPV with bivalent OPV and inclusion of at least one dose of the inactivated poliovirus vaccine in the routine immunisation schedule [
10]. In the long-term, withdrawal of serotype 2 from the OPV will probably reduce the overall risks of cVDPV2s, but outbreaks from serotypes 1 and 3 may still occur. To minimise the risks associated with VDPVs while OPV is still in use, it is important to understand the geographical variation in risk and identify suitable measures to limit emergence.
A challenge associated with modelling of infectious diseases is integration of research findings into public health policy [
11]. A mutually-beneficial, often long-standing relationship between policy makers and epidemiological modellers is necessary for modelling inference to be successfully translated into policy, where the timeliness and interpretability of modelling results are critical to be useful. Here we describe a risk scoring method that has been used to inform an international poliomyelitis risk assessment since 2014 and can be rapidly applied to additional countries or additional time-frames. The methodology is also applied to cVDPV risk and the spatial and temporal variation in risk associated with wild and cVDPV outbreaks are compared.
Discussion
We have provided a comprehensive country-level analysis of wild and cVDPV outbreak risk for 92 countries, based on factors that describe population immunity, historical propensity, exposure and susceptibility to poliomyelitis within a country. The modelling provides a framework for assessing wild and cVDPV risk, which allows a robust comparison between countries and time-periods. The risk for wild poliomyelitis has reduced since 2014, particularly in the African continent, where countries classified as High and Medium High have reduced from 31.7% in July–December 2014 to 9.0% in July–December 2016. The risk of VDPVs has remained fairly constant, as shown by the low but consistent number of cVDPV emergent events in time and consistent geographical pattern of at-risk countries.
Wild poliomyelitis risk is associated with low population immunity, increased population movements from countries with wild poliomyelitis and factors that relate to susceptibility. Country-level estimates of routine immunisation had the strongest association with wild type outbreaks when compared to other approximate measures of population immunity, where the variance in routine immunisation was approximately double the variance of the percentage under-immunised via AFP data. Mass immunisation campaigns are designed to raise the percentage of children who have received three or more OPV doses but their impact on preventing outbreaks may be limited by other factors, such as suboptimal SIA coverage, and access to healthcare [
28]. The association of diarrhoea-associated mortality with an increased risk of wild poliomyelitis outbreaks is likely to indicate reduced efficacy of the OPV that may be attributed to wider issues associated with interference from other enteroviruses (which may result in diarrhoea) and a high incidence of intestinal pathogens and associated environmental enteropathy [
29,
31]. Additionally, the association with diarrhoeal disease may indicate an increased transmission potential owing to climatic factors, poor hygiene and lack of healthcare access [
30‐
32]. Population displacement may result in increased exposure to poliovirus but also provide an indication of a dysfunctional health system. For example estimates of displaced persons of Syrian origin rose from <1% to >13% in 2012 [
21] alongside a reduction of DPT3 estimates to <50%, and a wild outbreak was reported in July 2013 [
14].
Migration from countries with poliomyelitis cases is a clear driver for wild poliomyelitis outbreaks and can be regarded as a good predictor of risk. Data for international migration was available for only 2001 and reliable time-varying estimates are required [
20]. Migration routes are likely to change annually and there are no consistent data sources that capture these changes, consequently we opted to use expert opinion and time-varying data on population displacement within the risk assessment. Since 2013 a majority of poliomyelitis cases on the African continent were part of the West Africa-B1 lineage [
33], which suggests that migration from Nigeria had a large impact on poliomyelitis cases in the African continent. Since this time Nigeria has substantially limited transmission of wild poliovirus through heightened surveillance, improvements in population immunity, and a surge in technical capacity within the country [
34]. The recent cases in Borno State suggest gaps in surveillance for acute flaccid paralysis and immunisation against poliomyelitis within a region heavily affected by the Boko Haram insurgency [
35]. Outbreaks in the African continent within the last 2 years are likely to have been contained and the last reported case of poliomyelitis outside of Nigeria was in Somalia in August 2014. Consequently, exposure to wild poliomyelitis has substantially reduced across the African continent, but as indicated via the risk assessment many countries within Africa remain susceptible to outbreaks owing to gaps in population immunity and susceptibility indicators. While transmission of poliovirus continues, improvements in population immunity and surveillance for cases remain vital. The South Asian (SOAS) lineage of poliovirus, primarily found in Pakistan and Afghanistan, was associated with outbreaks of poliomyelitis in Syria, Iraq [
36] and China (not analysed here). There are fewer examples of cross-border transmission of the SOAS lineage, despite there being more recorded migration from Pakistan and Afghanistan to other countries when compared to Nigeria [
20]. In countries directly bordering Pakistan and Afghanistan (Iran, China, Turkmenistan, Tajikistan, India) risk mitigation consists of routine immunisation schedules including up to five OPV doses and SIAs to prevent local transmission, resulting in at least 90% of children under 2 reporting >3 OPV doses [
37]. It is therefore likely that these mitigation steps have prevented outbreaks of poliomyelitis within these countries.
The risk score has been used to inform the planning of SIAs within the AFRO, SEARO, EMRO and EURO regions. The multi-agency risk classifications are submitted to the Eradication and Outbreak Management Group of the GPEI, whose remit is to direct activities that will support poliovirus detection and interruption [
38]. The allocation of SIAs is made according to the perceived risk of wild and cVDPV outbreaks, available vaccines, resources and finances [
3]. Other steps to mitigate outbreaks include improving routine immunisation coverage, adapting the routine immunisation schedule to include more than three doses of OPV, inclusion of immunisation teams at transit points in key locations [
19], and increased surveillance for cases of AFP [
3]. All of these activities require considerable resources and it has been increasingly important to ensure that the limited resources available for polio eradication are placed where they can be most effective. The risk assessment will continue to be carried out until wild poliomyelitis has been eliminated, but the methods may require some adaptation, for example by including environmental wastewater data, serological surveys and surveillance indicators as they become available. The risk assessment has been used by the GPEI to guide prioritisation of countries in the African region for support to improve case detection, SIA quality and outbreak preparedness in 2015 and again in 2016.
cVDPV risk has factors in common with wild outbreaks such as population immunity, population displacement and a previous history of reporting cVDPV outbreaks. The origins of cVDPV outbreaks differ and this changes the associated risk factors and geographical distribution of risk. Countries where over 1 million children are born each year require suitable health systems to immunise these children against poliomyelitis and other diseases, and in these settings it is especially important to ensure at least 80% of children are immunised. In 2010, DRC, Ethiopia, Nigeria, Pakistan, India and Indonesia were reported to have more than a million births and routine immunisation below 80%, indicating large numbers of children who may be at-risk of cVDPVs. It is likely that the risk of cVDPV emergence can be reduced with improved routine immunisation and SIAs [
9], but campaigns with good coverage will be critical to halt transmission. As of April 2016, the inactivated polio vaccine and/or monovalent serotype 2 OPV will only be used in response to circulating virus of that serotype. The model could predict risk in populous countries with low routine immunisation, such as DRC and Pakistan, but had a poor ability to identify ahead of time the risk associated with outbreaks in smaller countries such as Guinea, Madagascar, South Sudan and Ukraine. Low sensitivity of the risk score may be improved by including additional information that is not currently routinely available. Expert opinion of the resilience of specific countries to cVDPVs (for example by quickly initiating an outbreak response), information from surveillance reviews, reports on healthcare access within the country, and sub-national estimates of routine immunisation to account for heterogeneity in coverage [
9], may provide additional indicators of cVDPV risk. For example, concerns about a reduction in routine immunisation services during the Ebola outbreak in West Africa were reported in early 2015 [
39], and DPT3 coverage in Ukraine had been below 80% since 2008 [
40,
41]. A systematic way to incorporate these changes in risk is required to improve the accuracy of the risk assessment.
The inherent risk of cVDPVs associated with OPV use means that the gradual withdrawal of the OPV is essential, and this process started in April 2016 by removing the Sabin 2 strain from routine and SIAs and inclusion of the inactivated polio vaccine in routine immunisation [
10]. This step will strongly influence the temporal and geographical distribution of VDPV risk as it removes direct exposure to Sabin 2 poliovirus. Long-term, circulation of VDPV2 will reduce as emergence of VDPV progenitors decline. Surveillance for cases of AFP and poliovirus in wastewater samples will continue to be important in countries free from wild poliomyelitis as detection and response of cVDPVs and its progenitors is essential to prevent disease. Further research is required to understand how the distribution and incidence of cVDPVs of each serotype will change in the absence of serotype 2, and the analysis presented here will be updated to inform the surveillance and response activities associated with cVDPVs.
The use of a risk score based on a regression model is advantageous when compared to using the regression model alone. The risks for additional countries and time periods can be rapidly calculated and the reasons for why specific countries have been allocated a score are transparent, which can be advantageous when engaging with stakeholders with limited statistical modelling experience. There was a small reduction in the H-measure from using a risk score instead of the regression model, but we considered this a minimal reduction. Measures of the variation in uncertainty for each assessment are not available when using the risk score, whereas the prediction from the regression analysis includes estimates of uncertainty. For the purposes described here, statistical uncertainty of risk is not formerly considered when used to prioritise SIAs, and so the risk classification approach is the preferred option.
Acknowledgements
The authors gratefully acknowledge all those involved in the surveillance for acute flaccid paralysis and immunisation of children against poliomyelitis. We acknowledge the organisations that have made data publically available for incorporation into this risk assessment. We thank Margarita Pons-Salort for her comments on the manuscript.