This study used data routinely collected at health facilities to generate two common metrics of malaria morbidity, TPR and TCM, and compared temporal relationships between these metrics with direct estimates of malaria incidence in 5 high burden areas of Uganda. In this setting, changes in TPR were poor predictors of changes in malaria incidence, with small changes in TPR often associated with large changes in malaria incidence. In contrast, site specific changes in TCM exhibited a strong linear relationship with changes in malaria incidence, suggesting this metric could provide a useful indicator of relative changes in malaria morbidity over time within sites. However, relationships between absolute changes in TCM and absolute changes in malaria incidence varied from site to site, limiting the ability to directly translate changes in TCM to changes in malaria incidence.
Malaria surveillance is essential to monitor trends over time and space and evaluate the impact of control interventions. In settings in which transmission remains relatively high, surveillance activities focused on measures of malaria morbidity provide the most useful data for analysis of trends, stratification, and planning of resource allocation[2,13,14]. In most high endemic countries, routine health information systems involving health facilities provide the only practical, continuous, and systematic source of data on malaria morbidity. However, the utility of routine data from health facilities may be limited by incomplete or inaccurate reporting, lack of diagnostic testing in patients with suspected malaria, and poor quality laboratory diagnostics. Despite these challenges, an increased emphasis on laboratory-based confirmation of malaria and widespread availability of RDTs has improved the quality and utility of routine health facility-based data[11,15–17].
A strength of the current study was the use of high quality data from an enhanced malaria surveillance system at sentinel sites with a strong emphasis placed on complete reporting and laboratory confirmation for the diagnosis of malaria. Indeed, the fact that over 99% of patients with suspected malaria underwent diagnostic testing and over 96% of those tested had an RDT greatly reduced the potential for bias due to variations in these factors. Another strength of this study was the availability of estimates of malaria incidence from catchment areas around the health facilities. Malaria incidence provides the most direct measure of malaria burden and allows one to quantify cases over time relative to the size of the population at risk. The most accurate method of estimating malaria incidence involves prospective cohort studies, where all cases of malaria are captured from a defined study population [5,18–20]. However, cohort studies require considerable resources and are rarely undertaken as part of routine surveillance programmes. In this study, a practical and low-cost method was used to estimate malaria incidence by improving the capture of routine data on the village of residence among patients presenting to the health facilities, mapping catchment areas around the facilities, and estimating the population of these catchment areas. Indeed, although village of residence is included on the HMIS 031 standardized form, under routine circumstances this is rarely filled out and when it is filled out, fraught with inaccuracies and no way of linking this information to any meaningful population level data. Indeed, one of the key (and pain-staking) aspects of our “enhanced” surveillance system was training the staff at the MRCs to accurately fill out the village of residence, creating a novel coding system for entering this into our electronic database, and creating maps and shapefiles that would allow us to link malaria cases to our catchment areas and estimate the populations of our catchment areas. Generating direct estimates of malaria incidence provided a means of assessing the accuracy of surrogate measures of malaria morbidity, including TPR and TCM, in predicting changes over time.
TPR, defined as the number of laboratory confirmed cases per 100 suspected cases examined, has been used to define levels of endemicity, identify high burden areas, and evaluate the impact of control interventions [21–25]. However, TPR is subject to bias due to changes in the incidence of non-malaria fevers and has a complex, non-linear relationship with malaria incidence[5,7]. In addition, given that this metric is expressed as a proportion, it is commonly used as a qualitative measure as it is difficult to translate changes in TPR into meaningful quantitative measures needed to allocate resources and assess impact. In this study from 5 highly endemic areas of Uganda, temporal changes in TPR correlated poorly with changes in malaria incidence, with small changes associated with large changes in incidence. This is not surprising as when the burden of malaria is very high, TPRs can become nearly “saturated” well before malaria incidence has peaked. In a study from 15 villages in Western Uganda, the relationship between village level estimates of TPR and malaria incidence was best represented by an exponential model. In this study, the correlation between TPR and malaria incidence was poor at low transmission levels, with large changes in TPR associated with minimal changes in malaria incidence. The correlation improved among villages with higher transmission intensity where the TPRs ranged from 10-50%. However, this study did not address the other end of the spectrum when transmission intensity becomes very high and TPRs exceed 50%, as was observed in a majority of the time points for all 5 sites included in this report. Taken together, these data suggest that in Uganda TPR and malaria incidence have a non-linear relationship and correlate poorly when transmission is either relatively low or relatively high. In contrast to these data from Uganda, in a study from Yunnan Province of China annual estimates of TPR and malaria incidence had a strong linear relationship with an adjusted R2 value of 0.85 . In this study, malaria burden changed dramatically with annual TPRs declining from a high of 13% to less than 1% and malaria incidence declining from a high of 648 to 23 cases per 100,000 person years.
TCM, defined as the total laboratory confirmed cases of malaria per unit time, has also been used as a surrogate measure of malaria incidence. TCM is simple to measure, and unlike TPR, is quantitatively easy to interpret and not constrained by an upper limit. However, TCM is directly dependent on access to care and diagnositc testing and therefore highly susceptible to bias by these factors. For example, in a study from the Democratic Republic of the Congo evaluating trends in reported malaria cases between 2005 and 2014, a sharp increase in confirmed cases after 2010 was presumed to be due to the introduction and scale up in RDTs rather than a true increase in the incidence of malaria. The study presented in this report benefited from an enhanced surveillance system where almost all patients with suspected malaria underwent diagnostic testing using an RDT. Indeed, in this study with limited potential source of bias acruing from access to care and diagnositc testing, temporal changes in TCM tracked much better with changes in malaria incidence compared to temporal changes in TPR. In addition, site-specific temporal changes in TCM had a strong linear relationship with malaria incidence, meaning that within an individual health facility relative changes in TCM and malaria incidence were proportionate (e.g. a 75% increase in TCM would be associated with 3 times the increase in malaria incidence compared to a 25% increase in TCM). However, because the slopes of the linear relationships between TCM and malaria incidence varied from site to site, changes in TCM could not be directly translated into changes in malaria incidence (i.e. a 50% in TCM did not necessarily correspond with a 50% increase in malaria incidence). This is not surprising given that TCM is highly dependent on the number of patients who access a health facility, which can vary from site to site.
This study had several limitations. First, estimates of malaria incidence came from catchment areas around each MRC and could have been associated with inaccuracies in the numerator (cases of malaria per unit time) and/or the denominator (population at risk). It was assumed that all cases of malaria within the catchment areas were captured at their respective health facilities, which could have led to an underestimation of the true incidence of malaria. Population denominators came from publicly available datasets which utilized available census data and satellite imagery for mapping settlements . Errors in population estimates could have led to either an overestimation or underestimation of the true incidence of malaria. However, it is likely that potential bias in estimating malaria incidence was non-differential with respect to calendar time and therefore should not have had a significant impact on the analyses performed. Second, measurements of TPR and TCM were derived from all patients who presented to the MRCs while estimates of malaria incidence were derived only from the subset of patients who resided in the catchment areas around the MRCs. Differences between patients who did and did not reside in the catchment areas could have influenced the study findings, although in a previous study from Uganda adjustment for area of residence did not influence temporal trends in TPR . Third, this study was conducted at health facilities that were part of an enhanced malaria surveillance network where support was provided to maximize the use of laboratory testing and prevent stock-outs of essential commodities. Thus, care should be taken when generalizing results to other settings were the reporting of laboratory confirmed malaria may be affected by poor malaria case management. Finally, this study only included data from areas of Uganda with high transmission intensity and should not be generalized to lower transmission settings.