Statistical analysis of "Plasmodium falciparum" infection dynamics

Malaria is one of the major contributors to the global burden of disease. Worldwide, there were an estimated number of 200 million malaria cases in the year 2008, with a vast majority (85%) of those being in the African Region. This has lead to an estimated number of up to one million deaths, with a similar majority (89%) happening in the African region. There are several parasite species causing malaria, but most deaths are caused by Plasmodium falciparum. Malaria remains a major challenge for scientific research: constantly the parasite evolves resistance against existing drugs, and ever new substances to cure malaria need to be found. Creating a vaccine against Plasmodium falciparum proves exceptionally difficult, because the parasite has found ways to escape the human immune response. How exactly, is poorly understood. In addition, many countries affected by the disease suffer from poverty and ineffective health infrastructure. In the 1950’s the final eradication of malaria was envisioned by the WHO: the newly discovered insecticide DDT showed very promising results in reducing the malaria burden by killing the Anopheles mosquitoes, through which malaria is transmitted, and mathematical models of malaria transmission predicted that eradication of the disease would be possible. Despite great successes in the Caribbean, parts of Asia and South-Central America, and elimination in Europe and North America during the following decades, the efforts did not succeed in tropical Africa and many parts of Asia. After that failure, malaria was a “neglected” disease for a long period. Only since recent times malaria is again high on the global health agenda. Now, the enormous progress in the life sciences during the last decades provides new tools to better understand the parasite’s natural history, and perhaps will reveal new ways of attacking it. One factor which limited the understanding of the epidemiology of the parasite was that microscopy as diagnostic tool is not able to distinguish multiple concurrent infections within one human host: people in endemic areas often harbour several infecting clones in parallel. DNA-based methods make use of genetic loci of which many different variants exist in the parasite population, e.g. merozoite surface protein 2 (msp2), to distinguish co-infecting clones. This thesis develops statistical models to analyse such data on the presence of (mostly) msp2 genotypes. In particular, data from a longitudinal study in Navrongo, Northern Ghana is used in all chapters except chapter 6, where data from Papua New Guinea is analysed. A major challenge in the analysis of this type of data is the phenomenon of imperfect detection: the parasite hides in the deep blood vessels by attaching to the capillary walls, and it can therefore not be always detected in the peripheral blood. The three parameters which are estimated by our statistical models from time-series on presence or absence of genotypes are i) the force of infection (the number of infections acquired per person and year), ii) the duration of infection for one parasite clone, and iii) the detectability (the probability of detecting a parasite, given it is present). Previous statistical methods for the analysis of longitudinal genotyping data were restricted to exponential distributions of infection duration: this means that a constant rate (per time) is assumed at which infections are cleared. The reason for this was mathematical simplicity: the age structure of the infection population within a host can be neglected because the clearance rate is constant and does not depend on the age of an infection. In other words, the same mathematical model as for radioactive decay was used. Biologically, this is a very unrealistic assumption, and to understand more about within-host dynamics of P. falciparum or immunity against it one would like to distinguish between young and old infections. This thesis develops an extension to previous statistical analysis methods and makes use of parametric survival distributions to describe infection clearance and how it depends on the age of an infection. In addition to the age of infection, the effect of host age on infection clearance is investigated: older persons have experienced more infections and are therefore more immune. Changes in infection clearance with host age can therefore be interpreted as effects of immunity. An difference between the distribution of infection durations in the Ghanaian dataset compared to artificial infections4 emerged: a large proportion of infections in the Ghanaian population are cleared quickly after inoculation. It is the first time this could be measured from field data, and the result was confirmed using a different statistical method and study design. The difference between artificial infections and the field data cannot be attributed to acquired immunity in the Ghanaian population because all age groups show a similar abundance of very short infection durations. An interaction between the multiple infections within one host in Northern Ghana appears to be the most likely explanation. The implications of this finding for our understanding of the within-host processes in falciparum malaria are discussed.