How innovations in statistical approaches are helping to eliminate elephantiasis

LF patients in Tanzania (image courtesy of LCNTD)
6 October 2016

Dr Emanuele Giorgi describes how a new approach for developing predictive models to diagnosis and map lymphatic filariasis is helping in the global battle to eliminate the disease. 

The World Health Organisation estimates over 1 billion people are at risk of lymphatic filariasis (LF) infection. LF, commonly known as elephantiasis, is a devastating disease which can permanently disfigure and disable those afflicted with it. The disease is caused by a parasitic roundworm that is transmitted by mosquitos. Once inside a person, the worm lives and develops in the victim’s lymphatic system causing fluid collection and swelling.  

LF infection can be prevented by the mass distribution of medicines to people living in areas where the infections are present. Such is the effectiveness of this treatment that WHO and its partners believe that if correctly targeted and treated LF could be eliminated by 2020.

Key to this elimination is the ability to predict and target where LF actually exists. Unfortunately, the ability to build the statistical models needed to inform this mapping and targeting are currently restricted by the complexities of diagnosing LF. My research is looking at how we can develop a new approach to build a geostatistical model that will resolve this problem and will provide control programmes with the information and data they need to accurately target at risk communities.

Two camps of LF diagnosis

Diagnosis of LF falls into two camps. The first approach is to analyse patients’ blood for the presence of microfilariae, the small offspring of the adult LF worms. Microfilariae are nocturnal, which means that parasitologists have to wait until nightfall to diagnose a patient, making this a relatively difficult and costly procedure. The second approach is to look for the blood antigens that the body produces in response to the worm, this method is known as immunochromatographic rapid test (ICT). Whilst the ICT is cheaper and a lot more convenient that looking for microfilariae in the dark it can’t measure the intensity of infection which is needed to determine the risk of transmission, only microfilariae analysis will do this. So in essence you end up with two separate data sets which for the purposes of modelling and predicting where LF worms are difficult to align. 

The goal of my project is to develop novel geostatistical methods which will allow us to model the relationship between microfilariae data and ICT data and in turn use this model to predict LF prevalence in Africa.

The first law of geography

The scientific objective of a geostatistical analysis is to predict what would be the most likely value of a health outcome at any point in a geographical area of interest.  This is done by using data obtained through microfilariae and ICT prevalence surveys that are conducted by sampling a set of locations, which may correspond to villages or households. By making use of the first law of geography (“close things are more related than distant things”), geostatistical methods then allow us to predict the most likely value of LF risk at un-sampled villages or households. However, current available methods in geostatistics only allow to model ICT and MF prevalence as a pair of unrelated phenomena, without taking advantage of the potential benefits that might accrue from exploiting the strength of their association.

I have developed two different approaches for joint modelling MF and ICT data. The first uses biological knowledge about the underlying mechanism of transmission to make assumptions on the distribution of worms in the population and their reproductive rate. Using this we can then derive an explicit functional association between MF and ICT prevalence. In a second, empirical approach, this association between ICT and microfilariae is purely informed by the data. An advantage of this second framework is that it can be extended to other diseases with similar diagnosis challenges.  

Quantifying the differences between these the two approaches will help to understand to what extent the current knowledge on LF accurately describes the underlying mechanism of transmission.

The results of this research were presented at a recent Royal Statistical Society Conference, download this presentation (PDF)

Maps and data

Search for and download LF maps and Data