• Nov

    Cost versus accuracy: Testing rapid, low-cost epidemic detection tools

    To understand the magnitude of an epidemic we often need rapid and low cost-information. In some of the countries where Kimetrica works, security conditions, the capacities of local health authorities and the physical infrastructure, all contribute to making conventional serological and anthropometric surveys extremely costly and time-consuming. So it is natural to ask whether there are cheaper, faster alternatives. Of course the alternatives do exist, but are they reliable? How do we resolve the apparent trade off between cost and accuracy to identify appropriate and sustainable measurement solutions?

    Kimetrica staff are developing formal methods to test the accuracy of low cost alternative or “proxy” measures. Helen, our Head of Research, has recently published one such study, which is summarized below. It assesses how a simple visual test for cysticercosis compares with a full serological survey and identifies when the rough-and-ready method really works.  Eric, who leads our Data Lab, is at a much earlier stage in research into rapid alternatives to anthropometric measurement (including BMI estimation and identification of specific malnutrition related disease) using photographic imagery. Preliminary results are promising, but we need to test the method at scale before we make strong recommendations.

    The two studies share a common basic approach. Statistical models are used to compare the proxy measure with the more formal metric. There is no magic bullet. By definition, proxies provide only estimates of the underlying disease processes. But if they are “good enough” for catching an emerging epidemic or targeting responses, then they can serve a fundamental practical purpose.

    Case Study: A rapid tool for surveillance of cysticercosis 

    For centuries, farmers and pig buyers have used a very simple procedure to determine whether a pig is infected with the parasite cysticercosis. It is called lingual palpation and involves examining the underside of a pig’s tongue for cysts. Pigs infected with this parasite are unlikely to fetch high prices at market since cysts in the tongue indicate that cysts will also be found in the other muscles making the meat not only unattractive, but also of public health risk to people who consume it. Indeed, in most countries, pigs identified with cysts in their tissues at slaughter would be deemed unfit for human consumption.

    However, not all pigs infected with cysticercosis have cysts in their tongues, and relying on this technique alone to identify infected animals is flawed. For this reason, this approach is not recommended as an epidemiological or clinical tool to diagnose infection in individual pigs.

    Not to be deterred by such findings, a recent analysis by Guyatt and Fevre (doi:10.1111/tmi.12760)  investigated whether it could be used instead as a diagnostic tool for identifying geographical areas at high risk of infection. Extracting data from 26 individuals studies from 10 countries across Africa, they demonstrated that in fact there was a strong relationship between the prevalence of cysticercosis infection in pigs as assessed by lingual palpation and confirmed serological prevalence. The relationship was such that if one found at least 10 percent of pigs with cysts from lingual palpation, one could be pretty sure that at least 30 percent of pigs were infected.

    There are still debates on what constitutes a high community risk for cysticercosis, implying a need perhaps for another secondary data analysis on the relationship between prevalence of serological porcine infection and prevalence of disease in humans. Moreover, there is a clear need to validate this approach in other settings with more standardised data sets as sampling procedures for pigs was highly variable across the studies evaluated. Nonetheless, these findings suggest that there may now be a tool that could be rolled out to geographically map the prevalence of community risk from cysticercosis.

    This is extremely timely given the recent commitments of the World Health Organization to roll out control of this disease by 2020, a disease identified as the most important foodborne parasite globally by United Nations Food and Agriculture Organization. Tools are available, such as a pig vaccine and oxfendazole treatment of pig cysticercosis, but the challenge that remains is where to target control efforts. This recent work suggests that we may have an answer in screening pigs for cysts in their tongues; it is a rapid and cheap technique which requires little resources but which could yield important epidemiological information on community risk.


  • Dr Purity Nguhiu, Kenyatta University
    December 09, 2016 at 05:20 am ·
    Thanks for information Would this tool apply for areas where prevalence is low?
  • Helen Guyatt, Kimetrica
    December 13, 2016 at 10:15 am ·
    Thank you for your question. As indicated in the paper, tongue cyst prevalence is a far less reliable indicator of seroprevalence at lower prevalence: "If 10% or less of pigs have cysts on their tongues, seroprevalence can vary widely between 4% and 41%, and lingual examination becomes a far less valuable indicator of intensity of infection in the population. From the data available, it seems that this cut-off would identify seroprevalence rates of 30% or less with a negative predictive value (NPV) of 84% (16/19) (i.e. a sixth of those sites identified as having this low seroprevalence would actually have a high prevalence)." However this may change if data is collected in a more standardised format: "The data also need to be collected using standardised sampling frameworks. Various approaches to sampling villages, households and pigs were employed in the published data presented here. Studies that only sample a few households or pigs in a whole village are unlikely to be representative unless villages are highly homogenous in their levels of cysticercosis. We propose a multicountry approach in which villages with a range of pig population sizes and expected prevalences are comprehensively sampled and examined both clinically and serologically. This would help establish whether the NPV of 84% observed using this limited data set (with the possible concerns that a proportion of high-risk areas could be missed) is still applicable when the data are standardised. "

Leave a reply