Letters

Effect of Poor Census Data on Population Maps

+ See all authors and affiliations

Science  05 Oct 2007:
Vol. 318, Issue 5847, pp. 43a
DOI: 10.1126/science.318.5847.43a

The Review “Large-scale spatial-transmission models of infectious disease” (S. Riley, 1 June, p. 1298) states that “[f]or humans, an accurate estimate of population density is available for the entire Earth, up to a resolution of 1 arc sec.” The differing modeling approaches and input data used in the many global human population surfaces (13) mean that the estimated spatial distribution of populations and consistency both within and between products varies markedly.

The spatial resolution of input census data is critical to the mapping accuracy (4). For many countries, contemporary census data collected at a high administrative unit level exist to facilitate “accurate,” realistic-looking population mapping (e.g., fig. S1A) (5). For the majority of low-income countries, however, such data do not exist. This is especially true for much of Africa, where census data used for the production of global products are often over a decade old and at a resolution just below national level; a simple glance at the blocky and unrealistic-looking population distributions mapped for many African countries suggests that accuracy varies substantially (e.g., fig. S1B).

The lack of high-resolution data across much of the low-income regions of the world is likely to represent a significant limit to extending the reliable application of large-scale spatial transmission models of infectious diseases.

References and Notes

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.

Response

Tatem raises a potentially important issue. The accuracy of estimates of population density varies according to the quality of available supporting census data. However, current estimates for areas with poor census data may be sufficiently accurate to be used by studies based on large-scale spatial-transmission models.

Consider the potential transmission dynamics of reemergent smallpox. The main hypothesis supported in (1) is that, for the United Kingdom, spatial disc vaccination around known cases at either 15 or 50 km would not be an efficient addition to contact tracing, isolation, and vaccination. For the Central African Republic (CAR), results from a similar study would depend on the underlying assumptions of the human population model. Specifically, visual comparison of output from the global population model (2) for the CAR and northern Democratic Republic of Congo (immediately south of the CAR) suggests that heterogeneity between major roads in the CAR is underestimated. The sensitivity of predictions of disc vaccination efficacy for the CAR would have to be tested against this frailty, just as they would have to be tested against other key assumptions such as travel behavior and pathogen transmissibility. The post-hoc adjustment of global population data required for these sensitivity analyses would present particular technical challenges. However, given the much lower population densities in the CAR compared with the United Kingdom, if accurate travel data were available, it is entirely possible that a large-scale spatial-transmission model could be used with current global human population estimates to generate robust evidence in support of disc vaccination, perhaps with disc sizes greater than 50 km.

Another example where current population density estimates for Africa may be useful is in the analysis of the effects of sexual behavior change on the incidence of HIV in Uganda and Zimbabwe at different times (3). Did behavior changes affect the evolution of the regional incidence pattern over time, or is HIV incidence locally self-sustaining? If similar sustained behavior changes occur in other countries, can we predict spatial patterns of endemicity and/or eventual eradication of sexually transmitted infections? How useful could spatial targeting of resources across the region be in minimizing overall incidence? I do not suggest for a moment that large-scale spatial-transmission models can provide rapid definitive answers to these broad questions. However, using current population density estimates to construct large-scale models with these questions in mind might be a good starting point from which more specific relevant hypotheses could be generated.

References and Notes

  1. 1.
  2. 2.
  3. 3.
  4. 4.

Related Content

Navigate This Article