Estimation of the total number of paralytic polio cases based on Tebbens et al. (2011)

Our articles and data visualizations rely on work from many different people and organizations. When citing this entry, please also cite the underlying data sources. This entry can be cited as:

Sophie Ochmann and Max Roser (2018) - "Estimation of the total number of paralytic polio cases based on Tebbens et al. (2011)". Published online at OurWorldInData.org. Retrieved from: 'https://ourworldindata.org/estimating-total-global-paralytic-polio-cases' [Online Resource]
Published on April 14, 2018.

We thank Professor Radboud Duintjer Tebbens for helpful explanations and feedback on this appendix page.

In this document we explain the methodology of estimating the total number of paralytic polio cases from the number of documented polio cases.1

The methodology we follow here was introduced by Tebbens et al. (2011).2

In their paper, however, the authors conducted their estimations only for the time period 1980 to 2009 and they only included countries that were recipients of GPEI support. We extended their analysis and applied their methodology to a global dataset of countries until 2016.

The WHO surveillance system standards

With the establishment of the Global Polio Eradication Initiative (GPEI) in 1988, polio-monitoring and -reporting procedures became standardized around the world. The WHO's surveillance system relies on the detection of acute flaccid paralysis (AFP), which refers to paralytic symptoms more generally. Polio is just one of several causes for paralytic symptoms. Other reasons include trauma or Guillain-Barré syndrome (GBS).

Having to report all cases of AFP in any child under 15 years of age means that a country with a good monitoring system in place should detect at least one AFP case per 100,000 under-15-year-olds. If the number of AFP cases is less than 1 per 100,000 under-15-year-olds then there is reason to suspect that paralytic polio cases are underreported as well.3 The recording of AFP can therefore be used as a proxy measure for the quality of polio monitoring. A low rate of AFP cases suggests that polio cases are underreported as well.

The WHO system prescribes that each case of AFP should be followed by a diagnosis to find its cause. Within 14 days of the onset of AFP two stool samples should be collected 24 to 48 hours apart and these need to be sent to a GPEI accredited laboratory to be tested for the poliovirus.4 Since 2000, the WHO reports these surveillance quality indicators on a yearly basis for each country as the "Non polio AFP Rate" and "% Adequate Stool Collection", respectively.

Tebbens et al.'s (2011) estimation of total paralytic polio cases

These WHO data were used for the Tebbens et al. (2011) model on the economic benefits of polio's eradication to generate an estimate of the actual total number of paralytic polio cases based on the reported number of paralytic polio cases. Key for this estimation is the correction factor that captures the share of not-reported paralytic polio cases.

The authors used the following algorithm to determine which correction factor to use for all countries and years and we follow their approach:

  • Prior to 1996 from when the time the WHO surveillance data is available, Tebbens et al. assumed a correction factor of 7 for all countries.
  • From 1996 onwards, they continued to assume a correction factor of 7 if the non-polio AFP rate was smaller than 1 or if the percentage of adequate stool collection was below 60%.
  • For a country to have a lower correction factor, it would need to both improve its documentation of non-polio AFP cases and increase its share of adequate stool collection. If that was achieved but at least one of the two indicators was still relatively low (i.e. if the non-polio AFP rate was below 2 or the adequate stool collection percentage below 80%), the authors applied a correction factor of 2.
  • Only if both the non-polio AFP rate was larger than 2 and the adequate stool collection share was above 80% did Tebbens et al. use a correction factor of 1.11 (i.e., 90% completeness of reporting).
  • However, the authors apply two small exceptions. In footnote a of Table 1 of their paper, Tebbens et al. (2011) justify their choice of a 1.11 correction factor for China from 1989 to 1992 and for Oman in 1988.

Tebbens et al. then multiply this correction factor with the reported number of paralytic polio cases to arrive at their estimate of the true total number of paralytic polio cases. For their paper, they only used a sample of 104 countries that were recipients of the Global Polio Eradication Initiative's support. Furthermore, due to their publication date in 2011, they also only applied their model until 2009.

Results

We have applied the algorithm proposed by Tebbens et al. (2011), but taken a global dataset of reported paralytic polio cases and applied the algorithm to data for the period up to 2016. Because the WHO only provides the surveillance system data from 2000 onwards on their website, we used a correction factor of 7 for all countries up to 2000 and only from then onwards started using the algorithm outlined above.

The results of our estimation is visualized as the red time-series in the chart below.

Robustness check

The downside of our approach is that we multiplied the number of polio cases reported in countries with well-funded health systems also by 7 prior to 2000, even though they probably had a good reporting system in place. The United States, for example, has never had a surveillance system that reported non-polio AFP rates but is probably nevertheless good at detecting polio cases. After all, polio was declared eradicated from the United States in 1979 already so any cases documented after that were cases imported with people from other countries traveling back to the US.

On the other hand polio has also been rare in these countries in the period we are interested in here.

To see whether our model overestimates the total number of paralytic polio cases by making the assumption of a correction factor for all countries prior to 2000, we estimated two more models.

Robustness check (1) – using reported cases after being certified polio-free

For the model depicted in yellow in the chart above we used the estimated number of paralytic polio cases following Tebbens et al.'s (2011) algorithm for the years prior to a country's polio-free certification. This year is shown for each country in this world map here. From the year of certification onwards we here used the number of polio cases officially recorded by the WHO instead.

Robustness check (2) – excluding cases after being certified  polio-free

This model, which is shown in green in the chart above, can be seen as the lower bound for the estimated total cases for polio. We again used the estimated number of paralytic polio cases following Tebbens et al.'s (2011) algorithm for the years prior to a country's polio-free certification. But from the year of certification onwards we deleted all reported cases.

These two robustness checks show that the different assumptions for how to treat countries with good health systems and countries that eliminated polio early make very little difference for the total estimated number of global polio cases.