Robust Analytics

An Overview of Concepts and Models

Policy advice should be robust to uncertainty. To build rigor around the idea of robustness in policy, we recognized the need for a bespoke inferential framework designed for malaria analytics, which gave rise to RAMP (Robust Analytics for Malaria Policy). An important goal of RAMP was to build an integrated computational environment to support a range of activities, including conventional statistical analysis and methods to characterize, quantify, and propagate uncertainty. As RAMP took shape, RAMP software was being developed to implemented the principles in a stable computational form to facilitate the transformation of data into robust policy advice. Today, core RAMP software includes several R packages (see ramp.malaria) to deal with malaria epidemiology, transmission dynamics, and control in the broad sense. The ramp.library includes reusable code for one of several models of human malaria infection and immuno-epidemiology.

We recognized that the complex epidemiology of malaria - defined in the narrow sense to include only parasite infection in humans and processes affecting human health or parasite transmission - merits a deeper dive. Given the complexity of malaria, we needed a mathematical framework for malaria epidemiology that could expose to scrutiny the relationship between processes and patterns in human populations exposed to malaria parasites as they age, with different intensities and with different seasonal patterns. This software package, ramp.falciparum, implements a new computational approach to malaria epidemiology (in that narrow sense) using random variables. These methods are highly mimetic, but computationally intensive and difficult to apply. To address these limitations, we use hybrid variables to build a bridge from this probabilistic approach to a set of simpler approximating models.

The epidemiology of Plasmodium falciparum malaria presents a unique set of challenges due to the complex dynamics of infection, immunity, disease and infectiousness as well as treatment and chemo-protection, diagnostics and detection. Malaria can be measured in a dozen different ways, but it has been difficult to present a simple synthesis of malaria infection and disease in terms of the metrics that are commonly used in research and clinical surveillance. An important metric is the Plasmodium falciparum parasite rate, or PfPR, defined as the average prevalence of malaria taken from a cross-sectional survey. Another metric, often measured as a covariate in research studies, is a parasite count, the number of parasites in a blood slide field counted by a light microscopist. In an old data set, collected during malariotherapy, parasite counts fluctuated substantially over the time course of an infection, but they were strongly statistically correlated with the age of infection or AoI (Henry JM, et al., 2022)¹. In malaria, the PfPR in several old studies had a characteristic shape when plotted against age. Parasite densities have been used in research settings as both a diagnostic criterion and as a correlate of disease. Malaria epidemiology exhibits patterns that differ by diagnostic method, by season, by sex, and by location.

With so many interacting factors, it has been a challenge to develop model that could deal with everything all at once. One approach to studying malaria infection has been to develop mechanistic models for the dynamics of malaria infection within a single host. The most prominent models of this type are OpenMalaria and eMod, but there have been several others. These computational models made it possible to develope comprehensive individual-based simulation models, or IBMs, for malaria policy. While these approaches have been able to replicate the patterns, the outputs of the models are usually just as complex as the data collected in field studies. A synthesis of malaria epidemiology has proven elusive.