• Natalie A Molodecky, Isobel M Blake, Kathleen M O'Reilly, Mufti Zubair Wadood, Rana M Safdar, Amy Wesolowski, Caroline O Buckee, Ananda S Bandyopadhyay, Hiromasa Okayasu, and Nicholas C Grassly.
• Department of Infectious Disease Epidemiology, St Mary's Campus, Imperial College London, London, United Kingdom.
• PLoS Med. 2017 Jun 1; 14 (6): e1002323e1002323.

BackgroundPakistan currently provides a substantial challenge to global polio eradication, having contributed to 73% of reported poliomyelitis in 2015 and 54% in 2016. A better understanding of the risk factors and movement patterns that contribute to poliovirus transmission across Pakistan would support evidence-based planning for mass vaccination campaigns.Methods And FindingsWe fit mixed-effects logistic regression models to routine surveillance data recording the presence of poliomyelitis associated with wild-type 1 poliovirus in districts of Pakistan over 6-month intervals between 2010 to 2016. To accurately capture the force of infection (FOI) between districts, we compared 6 models of population movement (adjacency, gravity, radiation, radiation based on population density, radiation based on travel times, and mobile-phone based). We used the best-fitting model (based on the Akaike Information Criterion [AIC]) to produce 6-month forecasts of poliomyelitis incidence. The odds of observing poliomyelitis decreased with improved routine or supplementary (campaign) immunisation coverage (multivariable odds ratio [OR] = 0.75, 95% confidence interval [CI] 0.67-0.84; and OR = 0.75, 95% CI 0.66-0.85, respectively, for each 10% increase in coverage) and increased with a higher rate of reporting non-polio acute flaccid paralysis (AFP) (OR = 1.13, 95% CI 1.02-1.26 for a 1-unit increase in non-polio AFP per 100,000 persons aged <15 years). Estimated movement of poliovirus-infected individuals was associated with the incidence of poliomyelitis, with the radiation model of movement providing the best fit to the data. Six-month forecasts of poliomyelitis incidence by district for 2013-2016 showed good predictive ability (area under the curve range: 0.76-0.98). However, although the best-fitting movement model (radiation) was a significant determinant of poliomyelitis incidence, it did not improve the predictive ability of the multivariable model. Overall, in Pakistan the risk of polio cases was predicted to reduce between July-December 2016 and January-June 2017. The accuracy of the model may be limited by the small number of AFP cases in some districts.ConclusionsSpatiotemporal variation in immunization performance and population movement patterns are important determinants of historical poliomyelitis incidence in Pakistan; however, movement dynamics were less influential in predicting future cases, at a time when the polio map is shrinking. Results from the regression models we present are being used to help plan vaccination campaigns and transit vaccination strategies in Pakistan.

17 keywords...

hide…