The use of routine health facility data for malaria risk stratification in mainland Tanzania

Since 2000, a renewed commitment in malaria control saw an increased investment of funding to support various malaria control interventions across Africa. This resulted in substantial reductions in the disease burden in many parts of Africa. However, progress has plateaued in recent years and ten countries in Africa currently account for 66% of the global malaria disease burden. Further donor assistance is
unlikely and a new model for improving efficiencies in resource allocations is required to maximize gains.
In line with this, a major pillar of the World Health Organization (WHO) Global Technical Strategy (GTS) 2016-2030 encourages the use of accurate and timely routine data for stratifying sub-national malaria burden to track the changes in malaria epidemiology. The WHO High Burden for High Impact initiative (HBHI) further builds on the principles of the GTS framework and re-emphasizes the use of data to shift away from a “one size fits all” to a more tailored malaria control approach to accelerate progress against malaria. Countries are called upon to use all available health information to stratify the malaria burden in order to deploy effective malaria control tools to areas in greatest need and maximize impact and efficiency. As malaria declines, the heterogeneity in its transmission increases. Many countries have had an unequal distribution of high malaria burden within their national borders, and these high burden areas continue to remain high despite substantial control investment. Identification of high transmission areas would strategically accelerate national disease burden reductions. The purpose of stratifying malaria risk is to unpack this heterogeneity for optimized planning of malaria interventions. This needs to increasingly guide development of national malaria strategic plans (NMSPs) for efficient resource allocation.
Nationally owned routine surveillance systems can provide near real-time and granular data in time and space for stratifying malaria. However, data from these sources have largely remained underutilized due to concerns over completeness and quality. As a result, the diversity of Africa’s malaria burden has relied on the use of epidemiological modelling of parasite prevalence and opportunistic, and often dated, survey malaria data. These models have guided international priority setting, but at fine scales, can misrepresent trajectories in malaria risk. Current approaches by WHO to estimate malaria burden in 30 countries of Africa involve using modelled prevalence predictions and transforming them into incidence estimates through a modelled non-linear relationship. However, the ambition is that ultimately all countries provide reliable and accurate routine data to avoid over reliance on modelled estimates.
There is an increasing use of routine data, largely as a result of factors such as the launch of the WHO universal test and treat initiative that has significantly improved testing rates, the digitization of District Health Information Software (DHIS2) system that has improved health facility (HF) reporting rates (RR) and the emphasis by WHO GTS and HBHI initiatives to use data for decisions all of which are increasing the accountability and usage of these data. Efforts to incorporate routine HF data for risk mapping are emerging although most of these efforts are driven externally due to inadequate analytical capacity within countries. The increasing use of routine data has placed data quality initiatives to become an important operational component of surveillance across countries. Global efforts have introduced surveillance assessment toolkits to ensure a well-functioning surveillance system is in place to capture quality data from the routine information system. This is all expected to further enhance the accountability at level of data collection, aggregation and entry of routine information.
In mainland Tanzania, the diversity in malaria epidemiology within the country’s border has historically been described through malaria transmission seasons, urbanization, altitude and community-based parasite prevalence. There is no evidence however, on how these early maps were used to guide malaria control decision making. Recently, a model based geospatial framework using 10 years of community- and school-survey parasite prevalence data was used to highlight the heterogeneous nature of sub-national malaria transmission intensity. Whilst this is useful and provides the country with a baseline for understanding its transmission, these statistical models based on under-powered national household sample health surveys provide only one source of data. Their sustainable updating depends heavily on donor funding to support national household or school based surveys. As such, the need to explore alternative data sources notably from routine Health Management Information System (HMIS) is important. Targeting combinations of interventions based on local epidemiological criteria, whilst referenced in previous NMSPs, had never been formally established in mainland Tanzania until 2018. In 2017, during a mid-term review (MTR), it was recognized that progress towards reducing national parasite prevalence was being made (7% in 2017), but that further gains would require a strategic redirection of limited resources to achieve a prevalence of less than 1% by 2020. The MTR was followed by a consultative process with a forum of global and national malaria experts. Recommendations from this forum National Malaria Control Program, 2018b), together with those from the WHO GTS 2016-2020, were used to consider tailoring intervention approaches to the sub-national, local context, based on epidemiological stratification. Such an approach requires a data-driven approach, maximizing survey and routine data to establish epidemiological strata at operational units of programme delivery.
The aim of the work presented in this thesis was to explore and demonstrate the potential of routine HF malaria data to inform malaria risk stratification in mainland Tanzania. The objective was to explore the added value of leveraging information from multiple malaria metrics of the routine surveillance system of Tanzania in combination with survey data to map malaria risk at different spatial resolutions and thereby support the country’s ambition towards a more tailored malaria control approach.
This was demonstrated through first conducting key informant interviews with various stakeholders to understand common encountered challenges with using such data for analytical purpose. The objective was to understand the current approaches taken for HF data processing and cleaning. The interviews highlighted some of the existing challenges and the spectrum of methodological approaches currently being used to account for it in order to produce sensible analytical outputs. The key findings of this study recommended the need for developing guidelines addressing gaps in routine data and for handling such data in a systematic manner. This is essential for increasing confidence in the data, increase the usage of routine data for decision making, and generally enhanced harmonization in the approaches taken.
A simple and pragmatic approach that made use of combinations of multiple routine malaria metrics and survey data was then utilized to support NMCP with a macro-stratification risk map at council level for sub-national tailoring of interventions. This was instrumental in translating the risk map into suitable packages of interventions. The current strategic plan makes use of this evidence and advocates for tailored interventions through emphasizing burden reduction strategies in moderate-high transmission areas, and elimination strategies in low-very low transmission areas. Importantly, the methodological approach used was well within the capacity of NMCP staff at national level as it did not require data generated through complex survey methods nor utilized complex modelling methods.
The analytics was extended to the granular level of the ward to produce a micro-stratification risk map to further improve resource allocation. As the country is currently implementing the targeted packages of interventions, the goal is to move some of the decision-making processes towards a decentralized malaria control approach where council health management teams (CHMTs) would be empowered to understand the malaria situation in their respective wards and mobilize resources to areas that most need them to further maximize impact. The resulting micro-stratification revealed malaria risk heterogeneity within 80 councils and identified wards that would benefit from community-level focal interventions, such as community-case management, indoor residual spraying and larviciding. Micro-stratification is expected to allow this profound change in health planning processes by promoting a culture of data usage and equip council level with the capacity and tools to understand and appropriately respond to the local situation.
The use of crude aggregated routine data especially at the granular level of the ward for micro-stratification came with some limitations. One of the challenges was the incomplete nature of information in space and time, resulting in lower level administrative units (7% of wards) without empirical data. To overcome sparsity of data, geo-spatial models can leverage available routine information to predict risk in areas without information as well as provide the associated levels of uncertainty. A Bayesian spatio-temporal model was therefore used on test positivity rate (TPR) to leverage routine information and fill existing spatial and temporal gaps. The exceedance/non-exceedance probabilities were used to quantify the uncertainty of the estimated risk within policy relevant thresholds of TPR. Geo-spatial modelling provided a valuable framework for enhancing the use of imperfect routine HF data for malaria micro-stratification at program-relevant administrative units.
As Tanzania moves towards transitioning decisions to lower levels, a strong and robust guidance from national to council levels needs to be continuously provided. Councils that are empowered to make such decisions would require skills for understanding the local heterogeneity and making use of their local data to drive decisions. Whether the operationalization of micro-stratification for micro-planning is feasible and politically acceptable remains to be assessed and will require close monitoring of the processes at all levels. Overall, this work has demonstrated the ability of using local routine data in driving a country-owned stratification process at different spatial resolutions. This can have immediate potential in building a culture of data usage for decision making. Efforts towards strengthening capacity at all levels of the health system remains critical.