Malaria was placed at the forefront of the global public health agenda in 2000 when the Roll Back Malaria Partnership announced a commitment to halving malaria cases by 2015 [28]. A decade later, a 21% reduction in the malaria burden (clinical cases) had been achieved, with elimination [1] and even eradication now being pursued [28, 29]. Underpinning these aims is the need for early, accurate diagnosis and effective treatment of infections [30], to reduce onward transmission and prevent the development of severe malaria [1]. Ensuring that diagnostics continue to perform well, through regular assessments in all transmission settings, is critical for successful disease management, surveillance, and reaching targets such as elimination [31]. The detection of Plasmodium antigens using RDTs are widely used for malaria diagnosis. They are cheap, do not require advanced sample preparation or specialized training, and rapidly generate results [13]. This combination has transformed malaria case management from an emphasis on symptomatic presumptive diagnosis to one focused on diagnostic confirmation [3]. RDTs are now also used in active case detection (ACD) to identify infections in the community both as an intervention and as a means to assess changes in prevalence [32]. However, the sensitivity and specificity of RDTs varies significantly between commercial providers [33–35]. We therefore compared RDT and LM diagnostic performance against PET-PCR as a gold standard across a transmission gradient in Zambia with samples collected in a nationally representative MIS in 2018.
While both diagnostics performed well, overall LM had a higher diagnostic accuracy (91.3%) than RDTs (84.6%), supporting the historical use of LM as the gold standard diagnostic [1]. In general, this discrepancy was due to RDTs recording a high number of false positives. This tendency to over-estimate malaria prevalence can be seen in Fig. 3b where RDT prevalence is consistently higher (~ 10%) than PET-PCR prevalence, as well as in the low RDT PPV (range = 36–75%) compared to LM PPV (range = 71–90%). It is unclear from this study whether this relationship would hold below 15% true prevalence, but it seems unlikely that in near elimination settings there would be an ~ 10% overestimate. This over-estimation has been documented in other studies [36], and is most likely due to persistent antigenaemia post-infection, or to a lesser extent cross reactivity with other infections or autoantibodies [37, 38]. To investigate this observation further, we looked at the proportion of false positives and true negatives self-reporting having had a fever or taking drugs for a fever in the past two weeks. In these two groups, 33.3% (n = 421) of false positives vs 17.8% (n = 356) of true negatives reported fever / drugs taken in past two weeks. The elevated reporting in the false positive group potentially suggests that this group had a higher prevalence of malaria infection that either resolved by itself or was cleared with an antimalarial drug, either of which would lead to a false positive outcome. While not desirable, at the population level unnecessarily treating an additional 10% of the population is arguably better than missing 10% of the infected population. While this would translate to the consumption of additional treatments, current antimalarials have excellent safety profiles so is unlikely to unduly affect the cost-benefit analysis. Furthermore, if treatments are long-lived they could act as a prophylaxis. More worryingly, is the chance of false negative diagnostic results. In this case, an infection, and therefore the chance to break the chain of transmission, is missed. Both diagnostics perform well in this regard (Fig. 2), but RDTs have a noticeably lower FN rate (3.68%) compared to LM (6.81%). Similarly, LM prevalence was consistently below the PCR prevalence suggesting that this diagnostic is more likely to miss infections (Fig. 3b).
A key question this study set out to address was how diagnostic performance varies across a wide transmission / prevalence range, and in the main, the performance was consistent (Fig. 4). However, it does appear that sensitivity and PPV reduces as prevalence decreases, and more markedly with RDTs vs LM. This may be due to the diagnostics inherent LOD, i.e., the concentration of the target analyte that must be present for a diagnostic to be able to detect it. For symptomatic infections, that are generally characterized by high parasitaemias, a high LOD is unlikely to affect diagnostic performance. However, in ACD where asymptomatics with very low-density infections are likely to be encountered, the LOD could be critical in determining the success or failure of ACD as an intervention. As expected, the sensitivity of both LM and RDTs reduced as parasitaemia decreased (Fig. 6) suggesting that in settings where there is a high proportion of infections below the diagnostics LOD, e.g. pre-elimination, more sensitive diagnostic tools will be required to find the last asymptomatic infections [39, 40].
Overall, our findings from this study suggest that, to more accurately monitor malaria transmission dynamics, countries in pre-elimination settings require more sensitive diagnostic tools to detect asymptomatic and low parasitaemia cases. Molecular techniques such as nested polymerase chain reaction, quantitative real-time PCR, and ligase detection reaction fluorescent microsphere assay have been developed and have shown greater sensitivity for a broad range of parasitaemias [41, 42]. However, few molecular methods are commonly used in malaria endemic countries as they require a healthy capital budget, advanced laboratories, and skilled workforce. There have been many attempts to develop simplified molecular tools for malaria diagnosis appropriate for low-resource countries [43], e.g. loop-mediated isothermal amplification (LAMP) is a molecular point-of-care test with high specificity and sensitivity (5 parasites/µl of blood) well below the LOD for LM/RDT [44, 45]. Serological diagnostic assays (serosurveillance tools) that detect active or latent infection as well as past exposure may help to assess the malaria burden at a community level more accurately especially in low transmission settings where infections are rare, and surveys therefore require large sample sizes to confidently calculate prevalence [46–48].
This study had a number of limitations. Firstly, the MIS sampling design limited enrollment to children and we therefore could not evaluate diagnostic performance outside this age group. Moreover, there was a marked variation in sample size per province with overestimation of samples from some high transmission areas. This study also was not able to address treatment status of study subjects and other non-febrile illnesses among children to assess longevity of Plasmodium antigens after treatment and contribution of non-malaria illness on RDT false positivity.