Overview

Methodology for Adaptive Malaria Control, Uganda

Adaptive malaria control is a methodology for managing malaria with high quality, robust analytics that includes two essential components:

This methodology draws on principles developed for adaptive management of natural resources, including a reliance on mathematical theory and models. The concept is not entirely new to malaria, but it had not been formally described as a methodology or implemented by a national malaria program.

Over the past three years, we have defined adaptive malaria control as a methodology and developed a prototype for Uganda in collaboration with the National Malaria Control Division and the Department of Health Information. Development was guided by two questions:

  1. What were the essential elements required to implement iterative, robust analytics for national malaria programs? The methodology includes a chain of algorithms from data systems to policy advice designed to do high quality analyses on demand and also to learn what was most needed to improve malaria policies. Once developed, these elements could be adapted and adopted by other national malaria programs. This website – Adaptive Malaria Control, Uganda – discusses the Uganda prototype, and it explains the algorithms, protocols, and procedures in some technical detail.

  2. What were the essential concepts for adaptive malaria control? In a companion website – RAMP – Robust Analytics for Adaptive Malaria Control, – we explore the concepts from an academic perspective. In particular, we provide more background for the notion of robust analytics for malaria policy RAMP as a bespoke inferential framework for incorporating uncertainty into malaria policy.

As we developed the prototype for Uganda, it became clear that adaptive malaria control should be integrated into professional malaria analysts. Our goal was to provide the Uganda National Malaria Control Division robust analytics, and we wanted this to be integrated into the Ministry of Health so the activities would be sustained. The systems should support routine analysis and support innovation.

The scope of adaptive malaria control includes the entire chain of activities required to support malaria programs with high quality analytics, from the development of data systems to development of malaria policy. The data systems should be developed for malaria programs to ensure consistency across analyses: differences in the way analysts handle outliers and impute missing values can lead to confusing differences in the conclusions. The data assets should be up-to-date and in a stable form to develop protocols and procedures and repeat analyses. The system should be sophisticated enough to assimilate any useful analysis or algorithms developed by academics.

Information systems and analytics for adaptive malaria control should thus have several features:

  1. Routine analyses should have well-defined protocols and procedures to ensure that they are repeated consistently over time. The algorithms, protocols, and procedures should thus be well defined, version-controlled, archived, and documented, so that they can be run iteratively, on a schedule dictated by malaria program activities.

  2. Analysts should be accountable for their analysis and the policy advice they give. The analysis should thus be transparent, and it should be possible to replicate any analysis done at any point in the past. If policy advice changes, it should be possible to explain why: did the data change? did the software change? or did the analytical procedure change? For accountability, all data, software, and the protocols and procedures defining each analysis should be archived and version controlled.

  3. The analysis should be rigorous and aligned with best practices for malaria.

  4. The analysis should be designed to characterize and quantify uncertainty. The uncertainty should be propagated through the entire analytics pipeline, including each inferential step in a structured workflow that starts with data and ends with malaria policy advice.

  5. Most elements of the system should be modifiable, making it possible to adapt as we learn more about malaria in Uganda. Development should not disrupt operations. Each step in an analytics pipeline should thus have a well-defined operating version, and new features from the development pipeline should be thoroughly tested before updating the new operating version.

  6. Over time, as we learn about malaria in Uganda, uncertainty about malaria should tend to decline. A goal for adaptive malaria control is thus to have a way of prioritizing the most relevant information needed to reduce uncertaintly about a policy.

We structured the prototype around data systems – data processing pipelines and data assets – needed by malaria programs. Next, we developed some protocols and procedures for routine analyses for weekly and monthly reporting.

The challenge of managing malaria involves several temporal epochs, as they are defined for malaria analytics: the past, the present, and the future. Many tasks call for constructing a counterfactual past, which plays a key role in scenario and strategic planning, and which is neccesary to compute the averted and avertable burden of malaria.

A core activity of malaria analysis is retrospective data analysis: all the data we have describes the past. Retrospective data analysis takes us into the familiar world of statistics, but managing malaria puts an emphasis on the present. It is one thing to look at the whole history of malaria and quite another to develop algorithms to update the assessment of malaria in real time. To manage malaria, retrospetive analyses must be redesigned around the task of assessing the malaria situation now. The study of outbreaks must support algorithms for outbreak detection – to respond to an outbreak, it must be detected as it is happening. For scenario planning, we need to evaluate many possible futures, including forecasts of the counterfactual baseline, the forecast of malaria under today’s policies, or business as usual, and under a range of alternatives that would modify the forecasted baseline in other ways.

If we want to identify key areas of uncertainty and learn about malaria, we must develop some idea of the data that is needed for malaria policy, or malaria intelligence. The data should be timely and describe malaria in every management unit, so we developed analysis for management around malaria facility data. A key limitation of facility data is that it does not come with a well-defined, active sampling frame, so there is no way to validate it. We thus set about to develop information systems that transform facility data from the DHIS2 databases, maintained by the Department of Health Information (DHI), into rigorous estimates of malaria incidence and prevalence that could be validated. The metrics we need for policy also include entomological inoculation rates and some idea of the local mix of vector species.

Malaria intelligence is designed to generate these estimates using a combination of conventional and simulation-based analytics. For example, we developed a cross walking algorithm that uses the test positivity rate from health facilities to predict the malaria parasite rate around health facilities, and then we used simulation-based algorithms to reconstruct the history of exposure, predicted entomological inoculation rates, to make it possible to predict parasite rates by location, age, sex, and time of year. We use a knowledge of mass distributions of bed nets and IRS to estimate the impact of interventions. Simulation-based analytics are then used to construct counterfactuals, to develop forecasts, and for scenario planning. All of these algorithms are developed through analytics pipelines and organized workflows to develop malaria policy ensembles from which we develop robust policy advice. Analysis of the ensembles provides a basis for identifying and prioritizing critical gaps.