Measuring populations to improve vaccination coverage

Abstract

In low-income settings, vaccination campaigns supplement routine immunization but often fail to achieve coverage goals due to uncertainty about target population size and distribution. Accurate, updated estimates of target populations are rare but critical; short-term fluctuations can greatly impact population size and susceptibility. We use satellite imagery to quantify population fluctuations and the coverage achieved by a measles outbreak response vaccination campaign in urban Niger and compare campaign estimates to measurements from a post-campaign survey. Vaccine coverage was overestimated because the campaign underestimated resident numbers and seasonal migration further increased the target population. We combine satellite-derived measurements of fluctuations in population distribution with high-resolution measles case reports to develop a dynamic model that illustrates the potential improvement in vaccination campaign coverage if planners account for predictable population fluctuations. Satellite imagery can improve retrospective estimates of vaccination campaign impact and future campaign planning by synchronizing interventions with predictable population fluxes.

Figure 1. Niamey 2003–2004 brightness and measles cases.

(A) Left: map of Africa, Niger shaded grey; right: map of Niger showing three largest cities, each is a health district. (B) Reported measles cases by commune for Niamey’s 2003–2004 outbreak. Vertical dashed line indicates start of reactive immunization campaign. Inset: Pixels of Niamey: communes outlined in black, faded pixels outside communes show peripheral areas. Colors indicate communes for (B,C). (C) Solid lines: brightness within each commune’s boundaries (bright pixels in (B)). Dashed lines: brightness for each commune including associated peripheries (bright and faded pixels in (B)). Vertical line: start of reactive immunization campaign. Maps are GADM shapefiles drawn and edited in R version 1.7.1 (https://cran.r-project.org), finished in Adobe Illustrator CS3 (http://www.adobe.com/products/illustrator.html).

Figure 2. Population fluctuations within cities.

(A) Annual brightness signature for each commune of Maradi, corresponding to pixels shown on city map at left. (B) Annual brightness signature for each commune of Zinder, corresponding to pixels on city map at right.

Figure 3. Population fluctuations and vaccine coverage

. (A) Point estimates of coverage of reinforcement activities in the city from LQAS responses (left, black) and as calculated from doses distributed and city population size estimates from MSF, MoH, and the UN (left to right, faded points). Bright points and CI: Estimated coverage of reinforcement activities with CI including model estimates from posterior distribution of population flux using city population size estimates (bright points and lines). (B) Above: reported daily measles cases in Niamey. Below: estimated population fluxes of each commune by calendar day calculated from model. Vertical dashed line indicates start of reactive immunization campaign. Central solid lines indicate estimates for population flux based on posterior mean; shaded polygons indicate prediction intervals for flux based on central 95% of posterior distribution. Commune colors as in Fig. 1.

Figure 4. Vaccination time and outbreak size, commune 1, Niamey.

Boxplots of the predicted total size (central 90%) resulting from a measles outbreak if a two-week campaign vaccinating 90% of the population began on the day of the x-axis. The outbreak started on day 307, 2003; the immunization campaign began on day 105 of 2004, shown by the dashed vertical line. A campaign beginning as late as ~day 230 (grey arrow) can theoretically result in a smaller total outbreak size but best practice remains to vaccinate early in response to an outbreak.