TPR is increasingly used as an indicator of temporal trends in malaria morbidity. Ideally, changes in TPR over time will reflect true changes in malaria incidence for a population of interest. However, several factors including potential confounders such as age and area of residence, proportion of cases subjected to testing, care-seeking and utilization trends, and choice of diagnostic test may cause changes in TPR independent of true changes in the incidence of malaria. In this study, the effects of age, area of residence, and diagnostic test on TPR were investigated at two sites with different transmission intensity in Uganda. Age and area of residence demonstrated the potential to be important confounders at both sites given their independent associations with both time and TPR. However, controlling for each of them had only a small effect on the trends in TPR at the two sites. The directions of trends in the expected TPR using pLDH- and HRP-2-based RDTs were similar to trends in observed TPR, but there were differences between the values of observed and expected TPR at all time points. These differences were more pronounced at Aduku, the higher transmission site, and more pronounced using the HRP-2-based RDT.
Potential confounders are an important consideration in observational studies assessing any association, including temporal trends, which represent associations between time and an indicator of interest, in this case TPR. Any factor associated with both the exposure of interest (calendar time) and the outcome of interest (TPR) has the potential to confound temporal trends in TPR. Numerous factors may be associated with both time and malaria incidence such as weather, precipitation patterns, proportion of patients tested, care-seeking and utilization and home construction. Age and area of residence were chosen for analysis because they are well known to be associated with malaria incidence, and they can both be easily measured.
In the case of age, the expected association between younger age and higher TPR was observed. The age distribution of patients also differed significantly by calendar time, particularly at Aduku where the population receiving a malaria blood smear became gradually older over time. The comparison of temporal trends in observed TPR and age-adjusted TPR in Aduku provides a subtle demonstration of confounding in which age-adjusted TPR increased over time relative to the observed TPR due to the gradual increase in age over time and the lower TPR in older patients. Although confounding by age only had a modest effect on temporal trends in TPR in this study, the effect could be larger in other circumstances, for example a large increase in paediatric capacity at the clinic where surveillance is being conducted.
TPR was also associated with area of residence, though there were no clear patterns relative to distance from the clinic. In the case of Aduku, the TPR was lower outside the catchment area compared to near the clinic, whereas the opposite was true in Walukuba. As with the age distribution, the distribution of area of residence for patients receiving a blood smear also varied over time at both sites. However, there was no noticeable confounding of temporal trends in TPR by area of residence at either site. Nonetheless, it is easy to imagine circumstances in which confounding of temporal trends in TPR by area of residence may be important such as changes in the availability of transportation to the clinic from one area relative to another with a substantially different malaria burden. Controlling for factors such as age and area of residence with methods such as direct standardization or stratification can assure that changes in TPR over time that are due to confounding by these factors are not mistakenly ascribed to changes in malaria morbidity.
The choice of diagnostic test can also affect the interpretation of trends in TPR in two important ways. First, even when the proportion of patients with true infections stays the same, a change from one diagnostic test to another could cause a change in TPR that is exclusively due to a change in the proportion of true positive and false positive tests. Separately reporting the TPR for microscopy and RDTs, as is done in the World Malaria Report , partially addresses this problem. However, it still would not account for a substantial change in the quality of microscopy, which can be widely variable [15, 16], or a change from one RDT to another with different sensitivity and specificity . Second, the choice of diagnostic test affects the slope of the trends in TPR, and therefore the ability to distinguish a real difference in malaria morbidity. A low specificity test, which generates more false positives, will tend to obscure trends within the upper range of TPR values (closer to 1), whereas a low sensitivity test, which generates more false negatives, will obscure trends within the lower range of TPR values (closer to 0). High transmission sites such as Aduku are more likely to have a higher TPR, and are more likely to suffer from decreased specificity of RDTs, presumably due to frequent infections and persistence of parasite antigens after resolution of infection . This effect is demonstrated in the comparison between a relatively large decrease in observed TPR at Aduku between September 2009 and December 2010 and a much smaller decrease in the expected TPR with an HRP-2-based RDT (Figure2). These two effects of diagnostic test on trends in TPR can be accounted for by calculating the sensitivity and specificity of diagnostic tests periodically via comparison with a gold standard at the health facilities of interest. Using those results, the TPR can be adjusted accordingly based on the equation shown earlier.
This study has several important limitations. First, the sensitivity and specificity of RDTs at these sites using microscopy as a gold standard may have changed between the time of the study referenced for those values  and this study. Sensitivity and specificity of RDTs have been reported to vary based upon clinical and epidemiologic setting, most often related to differences in the distribution of parasite densities among infected patients, and based upon changes in storage and usage of the tests . Such a change may have affected the magnitude of differences between the values and slopes of observed and expected TPR, but the general direction of those differences likely would have been the same. Second, given the large samples sizes in this study, it is not surprising that statistically significant differences were found between potential confounders and the exposure (calendar time) and outcome (TPR) of interest. Indeed, the magnitude of these differences were of questionable relevance in terms of potential confounding and temporal trends in TPR based on adjusted analyses did not reveal differences compared to the unadjusted temporal trends that would likely be of public health importance. Third, this study was limited to two sentinel surveillance sites in Uganda, and may not be representative of many areas of the world with lower transmission intensity. TPR is not very useful as a surveillance indicator in settings with very low transmission intensity, though a TPR below 5% has been recommended as one of the criteria for readiness to shift to the elimination phase of malaria control . Finally, even in settings with malaria transmission on the order of that in these two study sites, TPR has many important limitations as an indicator of malaria burden which have been discussed elsewhere - the numerical change in TPR does not reflect either linear or proportional changes in malaria incidence in the population sampled, it is useful to estimate relative changes in malaria incidence but it cannot be used to estimate the actual incidence or compare incidence across sites, and changes in its value could be caused by changes in non-malarial fevers, the population of patients accessing the health facility, or changes in testing practices at the health facility .