Measuring the prevalence of malaria infection in population surveys underpins surveillance and control of the parasite. During more than a century of malaria research, parasite infection has been assessed by light microscopy of blood films. This wealth of data is widely used to understand malaria epidemiology, to monitor and inform control strategy1, to map the geographical distribution of malaria over time2 and to aid development of mathematical models. Rapid diagnostic tests (RDTs) based on antigen detection are now also used for prevalence surveys. However, both techniques have limited sensitivity. Molecular detection techniques for malaria3 have a much higher sensitivity and are increasingly revealing the widespread presence of infections with parasite densities below the detection threshold of either microscopy or RDTs. These results fundamentally challenge our current view of malaria epidemiology and burden of infection. In a previous systematic review and meta-analysis we found that microscopy misses on average half of all Plasmodium falciparum infections in endemic areas compared with PCR4. There was high variability between surveys and transmission settings. It remains unclear what factors cause this variation in levels of submicroscopic infections, and to what extent such infections are relevant to current efforts to control and eliminate the parasite. From a clinical perspective, low-density infection has been associated with mild anaemia5 and adverse effects during pregnancy6, but rarely causes acute symptoms. Nevertheless, the public health importance of low-density infections may be significant, as experiments have shown that mosquitoes feeding on individuals who are parasite-negative by microscopy can become infected with malaria7 8. The probability of detecting malarial infection is a function of the density of parasites and the volume of blood examined. Parasite densities in the peripheral blood fluctuate considerably over the course of any single P. falciparum infection and may dip under the microscopic detection threshold9 due to sequestration during the second half of the 48 h life cycle and varying effectiveness of the host's immune response. The volume of blood examined during microscopy slide-reading, if 100 high-power fields are screened, is 0.1–0.25 μl (refs 10, 11, 12), whereas for PCR detection DNA is extracted from 5 to 100 μl in most commonly used protocols. On the basis of these volumes, the theoretical detection limit for standard thick film microscopy is approximately 4–10 parasites per μl, and for PCR it is 0.01–0.2 parasites per μl. In practice, a low number of parasitized red blood cells in a sample is often not sufficient to enable detection due to technical factors such as loss of parasites during staining of microscopy slides10 13 or use of single versus nested PCR protocols. Calibration against cultures with known parasite densities has shown realistic detection limits of 10–100 parasites per μl for microscopy14 and 0.05–10 parasites per μl for various PCR assays15 16. From the perspective of control agencies aiming to reduce transmission, the most important question is to what extent do submicroscopic parasite carriers, who are missed during routine surveys, contribute to sustaining transmission? To become infected with malaria, Anopheles vectors need to take up a minimum of one male and one female gametocyte in a 2- to 3-μl bloodmeal. There is still a considerable probability of this happening at parasite densities that will often be missed by microscopy (for example, 1–10 parasites per μl), both according to mathematical theory and data17, and an aggregated distribution of parasites in the blood may assist transmission at very low densities18. During the scale-up of malaria control, public health agencies must decide what screening tools to use in different populations and whether submicroscopic carriers are a priority for intervention19. With sufficient resources, submicroscopic parasites could be detected in active screening programmes20 and included in evaluations where they may alter estimates of how interventions impact the prevalence of infection. Both from a biological and a public health perspective, it is important to understand where and when submicroscopic carriage is mostly likely to occur. Here we compile and analyse epidemiological data sets to assess firstly the prevalence of submicroscopic parasite carriers, and secondly which factors cause these carriers to be more numerous in some areas and population groups. We explore the roles of immunity, anti-malarial treatment, level of malaria endemicity and technical test performance. On the basis of 106 PCR prevalence surveys, we develop an analysis tool to estimate how prevalent such carriers are likely to be in any given area. We estimate the contribution of submicroscopic parasite carriers to the onward transmission of malaria by combining survey data with human-to-mosquito transmission studies. Results Submicroscopic parasitaemia across the endemicity spectrum We compiled survey data in which P. falciparum prevalence was measured by both microscopy and by PCR in the same individuals through updating a previous systematic review4. One hundred and six surveys met our inclusion criteria for analysis, taking place in endemic populations within a defined geographic area where participants were not selected according to malaria symptoms or test results, and where nested PCR or equivalent was used for parasite detection (see also Methods and Supplementary Table S1). Submicroscopic carriers were defined as those individuals with infections detected by PCR but not by microscopy. The specificity of microscopy relative to PCR is very high (98.4% on average4), and given infrequent reporting of specificity in the included studies we assume in our analysis that slide-positive results are also PCR-positive. Microscopy detected, on average, 54.1% (95% confidence interval (CI), 50.3–58.2%) of all PCR-detected infections across the 106 surveys, but this sensitivity varied widely (Fig. 1a) as in previous analysis4. Regression analysis showed that the PCR prevalence of infection has a strong linear relationship with microscopy prevalence on the log odds scale (Fig. 1a). Stratifying by age group improved the fit to the data with microscopy sensitivity being higher in children only ( 200) so that measures were representative of an average infection. Other human infectiousness studies We searched the literature to find as many studies as possible measuring human-to-mosquito transmission from individuals in malaria-endemic areas with neither asexual parasites nor gametocytes detectable by microscopy, using PubMed and modern transmission studies17 31 as starting points and searching through the relevant literature using bibliographies and a review8. We identified three further relevant studies in addition to the malaria therapy data. One of these directly measured submicroscopic parasitaemia using quantitative nucleic acid sequence-based amplification as well as slide positivity31. Two further human-to-mosquito transmission experiments measured the infectiousness of slide-negative individuals, but their infection status was not tested by molecular methods7 30. We estimated the PCR prevalence in these study populations using the log linear model described in the main text (equation 1, Fig. 1a) and the reported slide prevalence in the study. The prevalence of submicroscopic carriage was calculated as: We assumed all infections from slide negatives arose from these submicroscopic carriers, using them as the denominator in calculating infectiousness. The contribution of slide-positives or submicroscopic carriers to the infectious reservoir was calculated as: The proportion of mosquito infections which would originate from submicroscopic infections was estimated as: Age-prevalence data We extracted the data from all studies which included children and adults and which gave a breakdown of prevalence by microscopy and PCR for at least three age groups. We fit a linear relationship between microscopy sensitivity and age, using the midpoint of the age group, and tested whether underlying population PCR prevalence was a modifying factor. Here we used log prevalence ratios, as log ORs of microscopy: PCR positivity would decline as PCR prevalence increased, even with constant sensitivity in all settings. Author contributions L.C.O. updated the systematic review, did the analysis and drafted the manuscript, T.B. reviewed the studies and collected data, J.T.G. advised and contributed to the analysis, A.L.O. contributed data. All authors contributed to interpretation of data, writing and revising the manuscript, and have seen and approved the final version. Additional information How to cite this article: Okell, L. C. et al. Factors determining the occurrence of submicroscopic malaria infections and their relevance for control. Nat. Commun. 3:1237 doi: 10.1038/ncomms2241 (2012). Supplementary Material Data sources and references Supplementary Table S1 and Supplementary References Prevalence estimation tool Spreadsheet permitting estimation of PCR prevalence from microscopy slide prevalence and vice-versa