Logo of jcmPermissionsJournals.ASM.orgJournalJCM ArticleJournal InfoAuthorsReviewers
J Clin Microbiol. 2011 Nov; 49(11): 3885–3891.
PMCID: PMC3209109

Prevalence and Density-Related Concordance of Three Diagnostic Tests for Malaria in a Region of Tanzania with Hypoendemic Malaria


Accurate malaria diagnosis has dual roles in identification of symptomatic persons for effective malaria treatment and also enumeration of asymptomatic persons who contribute to the epidemiologic determinants of transmission. Three currently used diagnostic tests, microscopy, rapid diagnostic tests (RDTs), and real-time PCR, all have different sensitivities and specificities, which are parasite density dependent. Here, we compare their concordance among 451 febrile episodes in a cohort of 2,058 children and adults followed over 6 months in a region in central Tanzania with hypoendemic malaria. Microscopy, a histidine-rich protein-based RDT, and two different real-time PCR gene probes detected Plasmodium falciparum in 20, 54, 41, and 78 episodes of fever, respectively. They had complete concordance in only 9 episodes. Real-time PCR with an 18S probe was more sensitive than with a mitochondrial probe for cytochrome b despite higher copy numbers of mitochondrial DNA. Both PCR yields were increased 4-fold by glycogen/acetate precipitation with low-speed centrifugation. Duplicate PCR increases low-density malaria detection. RDT had the highest number of unique positives, presumably from persistent antigen despite the absence of parasites, although RDT did not detect 3 parasitemias with over 1,000 parasites/μl. In a latent class analysis, real-time PCR had significantly higher sensitivity than did microscopy or RDT. Agreement between real-time PCR, RDT, and microscopy was highest in March and April, when both the P. falciparum parasite rate and parasite densities are highest. Real-time PCR is more sensitive and specific than RDT and microscopy in low-prevalence, low-parasite-density settings.


Low-prevalence, or hypoendemic, malaria poses particular diagnostic challenges because low population prevalence reduces the positive predictive value of tests. Moreover, test sensitivity suffers when parasite densities within individual infections are low (3). The proportion of low-density infections may rise with decreasing transmission (2), so hypoendemicity diminishes test sensitivity, as well as the positive predictive value (3). Malaria diagnostics, such as real-time PCR (RT-PCR), chromatographic rapid diagnostic tests (RDTs), and traditional Giemsa staining and microscopy, have performed differently across ecoepidemiological settings, leading to varying estimates of agreement between tests. Agreement between microscopy and PCR has been shown to be poor under hypoendemicity (malaria prevalence of less than 10%) (5), and agreement between RDT and RT-PCR or microscopy may also decline when the prevalence is low.

We compared the yields of RT-PCR, RDT, and microscopy for the diagnosis of Plasmodium falciparum malaria in a cohort of 2,058 adults and children from a district in central Tanzania with hypoendemic malaria followed from January to June 2009. The district has seasonal malaria, coinciding with the rainy season, and in the study year, malaria was hypoendemic. The cohort was assembled to investigate the ancillary benefits of an azithromycin mass drug administration for trachoma. In children, fevers were detected at surveillance home visits every 2 or 3 days and at regular surveys. In adults, fevers were detected at the survey visits corresponding to baseline and 1, 3, 4, and 6 months postbaseline. Febrile participants in the cohort were evaluated for P. falciparum malaria by (i) Giemsa staining and light microscopy by expert microscopists at the Amani Centre, Muheza, Tanzania; (ii) the Paracheck RDT administered in the field; and (iii) RT-PCR for the P. falciparum 18S ribosomal gene (RT-PCR18S) and for the mitochondrial multicopy (20 to 30 copies per parasite) cytochrome b gene (RT-PCRcyto) performed on blood samples collected on filter paper in the field and analyzed at Johns Hopkins Hospital (JHH).

Here, we present the application and optimization of the RT-PCR18S diagnostic test and compare RT-PCR18S to microscopy, RDT, and RT-PCRcyto. In addition to the Tanzanian field samples, we applied the RT-PCR tests to a 4-fold dilution series from six patients with P. falciparum malaria who presented at JHH and more than 20 JHH negative controls.


Field site and sampling.

The study participants were drawn from eight rural agricultural villages in Dodoma Province, central Tanzania (latitude −6.1°, longitude 36.6°). Each village has between 1,000 and 1,900 inhabitants and is approximately 25 km2 in area. Sampling occurred over 6 months, from January 2009 through July 2009, and encompassed the traditional rainy and growing seasons.

The population of the study site was enumerated by a community-wide census. Four azithromycin mass drug administration treatment villages were selected based on high (over 10%) trachoma prevalence in 1 to 9 year olds, and 4 control villages were selected. In each village, 130 families with children under 5 years old were randomly selected, and one child and one adult were randomly selected in each family. Surveys of the entire study population were carried out at baseline and months 1, 3, 4, and 6. The children were split into 3 sampling subgroups and sampled over weeks 2, 4, and 8. Contemporaneous with the surveys, staff members from the communities carried out an active surveillance program in which children were visited twice each week and examined for fevers. At all the survey visits, a blood sample was collected for RT-PCR on Whatman Protein Saver filter paper. When a person was found to have a fever, thick and thin smears were prepared for microscopy and RDT was performed, in addition to collecting the blood sample on filter paper for RT-PCR. Two experienced microscopists at the Amani Centre in Muheza, Tanzania, who were blinded to the RT-PCR and RDT read the slides from the febrile patients. For smears where the parasite count was low (<10 parasites), a third reader reconciled discordant slide readings. RDTs were interpreted in the field at the time of sampling.

Nonendemic controls.

Blood samples from 6 patients that were found to be positive for P. falciparum by the microbiology laboratory at JHH were serially diluted 10-fold four times in P. falciparum -negative blood. The initial parasitemias were 24,000, 14,000, and 4,000 plus three at 2,000 per microliter based on thin-film percentages and 4 million erythrocytes per microliter. The diluted positive blood samples and 20 negative-control blood samples taken from patients presenting to JHH from regions where malaria is not endemic were spotted onto Whatman Protein Saver filter papers and allowed to dry for 48 h.

RT-PCR procedure.

In the laboratory, five 3-mm-diameter punches containing the equivalent of 25 μl of whole blood were removed from the filter papers (4). DNA was extracted with a commercial 96-well extraction kit from Promega into 250 μl of water. A reserve sample consisting of 25 μl was stored. DNA extract (225 μl) was concentrated in the 96-well plate by addition of 20 μg of glycogen with sodium acetate, pH 5, and ethanol. The precipitate was pelleted and washed in 70% ethanol in a low-speed (3,000 × g) centrifuge for 30 min. The DNA was air dried and resuspended in 25 μl of nuclease-free water.

The 18S gene was detected in a multiplex RT-PCR that simultaneously assayed for the AMA-1 gene for Plasmodium vivax and the glpQ gene, which is conserved across Borrelia spp. Table 1 shows primers and probes. Samples were run in duplicate on manually loaded 384-well plates, and the multiplex design allowed assay of P. falciparum , P. vivax, and Borellia simultaneously. RT-PCR was carried with the Bio-Rad CFX 384 real-time PCR Detection System thermocycler (Bio-Rad, Hercules, CA) under a 40-cycle standard PCR protocol. Baseline relative fluorescence unit (RFU) values were readjusted within the Bio-Rad CFX manager software in accordance with the manufacturer's instructions. Calls for the presence or absence of parasite antigen were optimized with a panel of diluted JHH validation controls and the Tanzania microscopy data. For the 18S target, the microscopy-positive sample with the lowest relative fluorescence (RFU) from the Tanzania data was taken as a candidate for the cutoff for RT-PCR positivity. If the fluorescence from the candidate cutoff was above the highest fluorescence of the negative-control JHH validation samples, then that cutoff was used to determine positivity. If the fluorescence from the candidate was lower than that of the negative controls in the JHH validation samples, then the fluorescence from the next-higher RFU sample from the microscopy-positive Tanzania samples was compared to the fluorescence from the negative samples. Comparing Tanzania microscopy-positive samples with increasingly high fluorescence continued until a microscopy-positive Tanzania sample with fluorescence higher than that of all the negative controls was identified. When the microscopy-positive sample with fluorescence higher than that of all the negative controls was identified, that sample's fluorescence or higher fluorescence defined positivity. The cycle number at which an amplification signal crossed the 50-RFU threshold (CT) was used to quantify parasite densities.

Table 1.
Primers and probes for multiplex PCR

The reliability of the RT-PCR assay for P. falciparum was assessed by repeating the extraction process and assaying for the P. falciparum 18S gene on 10% of the samples and by repeating the RT-PCR using a cytochrome b gene-specific target.


Confidence intervals and statistical tests for differences in proportions and means were calculated in the R statistical software environment (6). Intraclass correlation coefficients for agreement between statistical tests were calculated with the “irr” package in R (10). The sensitivity and specificity for the Tanzanian field data were estimated with latent class analysis (LCA), which allows sensitivity and specificity estimation without a gold standard. In this LCA, unobserved or latent malaria infection is used to explain dependencies between the observed diagnostics. RDT, microscopy, and RT-PCR are assumed to be conditionally independent given the true unobserved infection status. LCA was estimated using the “randomLCA” package (1). Venn diagrams were made with the help of the “venneuler” package (13). All data management and statistical scripts that produced these results are available upon request.


A total of 451 febrile episodes were identified among the 2,058 participants who were followed for 6 months, 426 of which were in children. The active surveillance program for symptomatic children identified 357 instances of fever in the children, and the cross-sectional surveys found 69 instances of fever. Individuals suffered repeated episodes of fever: 55 participants had 2 instances of fever, 12 participants had 3, and 2 participants had fevers on 4 occasions. Samples from all 451 febrile episodes underwent RT-PCR with the 18S target, RDT, and microscopy; 402 of the febrile samples underwent RT-PCR with the cytochrome b target.

RT-PCR application and optimization.

In the JHH dilution series, glycogen precipitation and low-speed centrifugation allowed the detection of 62.5% (20/32) of samples with parasite densities of <1,000 parasites/μl compared to only 15.6% (5/32) prior to precipitation (P < 0.0001; Fisher's exact test). The precipitation increased the area under the curve (AUC) in a receiver-operator characteristic (ROC) curve from 63% (95% confidence interval [CI], 52% to 73%) without precipitation to 94% (95% CI, 88% to 100%) and to 100% for starting parasite densities that were greater than 1,000 parasites/μl (Fig. 1A). For samples that were correctly classified as positive in both the precipitated and nonprecipitated samples, the cycle where the RT-PCR application first crossed the threshold of detectability (CT) was 1.85 (95% CI, 1.11 to 2.59) cycles lower on average in precipitated samples than in matched nonprecipitated samples (P < 0.0001; paired t test), indicating the precipitated samples increased the DNA input to PCR by 4-fold.

Fig. 1.
ROC curves show glycogen precipitation, and using the highest of 2 RT-PCR replicates as a cutoff improves the AUC. The data are from 6 P. falciparum -positive patient samples diluted 4-fold (22 samples, with the highest 2 samples removed) and 20 P. falciparum ...

Each filter paper blood sample was run in duplicate on the RT-PCR, so multiple options were available for drawing a cutoff between positive and negative samples. Several options shown in Fig. 1B are (i) the RFU from the last RT-PCR cycle of the highest RT-PCR replicate, (ii) an average RFU from the last 5 RT-PCR cycles of the highest RT-PCR replicate, (iii) an average of RT-PCR replicates based on the average of the last 5 RT-PCR cycles from each RT-PCR replicate, (iv) a cutoff based on the CT, and (v) cutoffs based on the Bio-Rad defaults. The cutoff scheme based on the last RFU of the highest RT-PCR replicate had higher AUC in JHH validation data with the 18S (Fig. 1A) and cytochrome b targets. Using microscopy as a reference or gold standard, the last RFU of the higher RT-PCR replicate increased the AUC in the Tanzanian data compared to the other four call schemes. However, increases in the AUC from using the last cycle of the highest replicate compared to the other four call schemes were not statistically significant.

For RT-PCR18S, a cutoff for RT-PCR positivity of 650 RFU, based on the last RFU of the high RT-PCR replicate, was chosen as a cutoff because the lowest amplification signal produced by the microscopy-positive samples in the Tanzania sample was 652.35 RFU (Fig. 1, yellow dots, and 2). A cutoff of 650 RFU was consistent with 100% specificity in the JHH validation and 94% specificity in the Tanzanian samples when microscopy was used as a reference. A microcopy-positive Tanzanian sample with 350.49 RFU was observed and was considered a candidate cutoff; however, this sample was not used as a cutoff because JHH validation negative-control samples had signals as high as 503.49 RFU. Both the 18S and cytochrome b targets were run with Cy5-labeled fluorescent probes. For the cytochrombe b target, the majority of the negative controls were below the RT-PCR18S negative controls; however, a single high negative control, which may have been an outlier resulting from pipetting error, complicated the cutoff determination. Ultimately, the cytochrome b cutoff point was set at the same level as for 18S, that is, 650 RFU.

The log10 values of the Tanzanian microscopy counts were regressed on the CT values from RT-PCR18S to estimate a linear standard curve with an r2 of 0.92 (Fig. 2A). A similar standard curve of CT values from RT-PCRcyto rather than RT-PCR18S produced a standard curve with a lower r2 value (0.52) (Fig. 2B). When the Tanzania-derived, RT-PCR18S- and RT-PCRcyto-based standard curves were applied to the JHH validation samples, the RT-PCR18S curve fit the validation samples better than the RT-PCRcyto curve (mean squared error for RT-PCR18S, 1.22, versus the mean squared error for RT-PCRcyto, 1.48). References to parasite densities estimated by RT-PCR below refer to values predicted from the RT-PCR18S Tanzania-derived standard curve.

Fig. 2.
The standard curve of the 18S target performs better than the standard curve of the cytochrome b target. Shown is RT-PCR (CT) versus microscopy (log10 parasites/μl). The open circles show Tanzania microscopy, and the crosses show JHH validation ...

Duplicate RT-PCR amplification appreciably improved the detection of infections at low parasite densities (Fig. 3). Of the 41 RT-PCR18S-positive patient samples in the Tanzania group, 16 were positive on only 1 replicate, and all 16 of those single-positive samples had RT-PCR parasite densities of <200 parasites/μl. Of the 16 patient samples that were single positives, 1 sample was also RDT and microscopy positive, 1 sample was RDT positive and microscopy negative, and 1 sample was RDT negative and microscopy positive (Fig. 3B). Of the 55 RT-PCRcyto-positive samples, 29 were positive in a single replicate, with 4 RDT positive and 0 microscopy positive. In the JHH validation data, 5 of 30 correctly identified positive samples were positive on only 1 replicate.

Fig. 3.
Duplicate amplification of RT-PCR samples improves detection at low parasite densities. (A and C) Fluorescence (log10 RFU) for replicate 2 versus replicate 1 for all 451 febrile patients for RT-PCR18S (A) and RT-PCRCyto (C). Quadrant boundaries are at ...

RDT, RT-PCR18S, RT-PCRcyto, and microscopy.

RT-PCR18S captured more microscopy-positive samples than did RT-PCRcyto in both the Tanzanian (70% versus 60%, respectively) and JHH (68% versus 52%, respectively), samples (Table 2). When the microscopy read was below 1,000 parasites/μl, RT-PCR18S identified 25% of the Tanzanian samples as positive and RT-PCR18S identified 62.5% of low microscopy reads as positive in the JHH controls. In comparison, RT-PCRcyto identified none of the Tanzanian samples and 46.9% of the JHH samples as positive when microscopy classified those samples as positive but below 1,000 parasites/μl. On the other hand, CT values for RT-PCRcyto positives were lower than CT values for RT-PCR18S, indicating more starting cytochrome b DNA (mean difference in paired t test, 1.85 cycles; 95% CI, 1.11 to 2.59).

Table 2.
Positive and negative calls by diagnostic method

RT-PCR18S identified 41 (9.1%) samples as positive compared to 54 (12.0%) by RDT and 20 (4.4%) by microscopy. Despite calling more samples positive than other methods, RDT called 3 samples negative that were positive by RT-PCR18S and over 1,000 parasites/μl by microscopy. Of microscopy-positive samples, RDT identified only 50% (4/8) of the fever group (≥38°C) as positive compared to RT-PCR18S, which called 100% of the fever group positive. The RDT called negative 3 infected participants, ages 1, 4, and 4 years, with parasite densities of greater than 1,000 parasties/μl by microscopy. All 3 of the high-density infections missed by RDT had fevers greater than 38°C. An additional participant, age 3, with a fever above 38°C but with 48 parasites/μl, was incorrectly classified as negative by RDT (Table 2 and Fig. 3).

In an LCA with only microscopy and RT-PCR18S, both microscopy and RT-PCR18S showed good specificity (100%). The sensitivity for RT-PCR18S was 71% (95% CI, 49% to 88%), which is consistent with the sensitivity for the 18S probe in JHH controls (68%; 30/44) (Table 3). In an LCA that included microscopy, RDT, and RT-PCR18S or RT-PCRcyto, RT-PCR18S outperformed the other tests in terms of sensitivity (96%; 95% CI, 74% to 100%) and specificity (94%; 95% CI, 95% to 100%).

Table 3.
Sensitivity and specificity of LCA

Agreement between microscopy and RT-PCR18S was higher (intraclass correlation coefficient [ICC] = 43%) than the agreement between RT-PCR18S and RDT (ICC = 23%) and between microscopy and RDT (ICC = 25%). Agreement between RT-PCR, RDT, and microscopy was highest in March and April, when both the P. falciparum parasite rate and parasite densities are highest (Fig. 4). In January and February, when incidence and parasite densities are low, RT-PCR18S-positive participants tend to be older than in the months when incidence and parasite densities are high (Fig. 4).

Fig. 4.
In March and April, the P. falciparum (Pf) parasite rate and parasite densities by RT-PCR and microscopy are highest, and the ages of those infected are low. (A) P. falciparum parasite rates over time for positive tests. The lines indicate P. falciparum ...

A second DNA sample was extracted from 15 RT-PCR18S-positive filter papers, and 3 were positive upon the second RT-PCR run. Six samples that were RT-PCR18S negative on the first extraction tested positive on the second extraction. On both extraction 1 and extraction 2, 144 samples were negative.


Precipitating DNA to increase the final concentration 4-fold greatly enhanced the sensitivity of the RT-PCR assay and improved the AUC from 63% to 94% (P < 0.005). Classifying samples as either positive or negative based on the RFU value of the highest RT-PCR replicate from the last cycle of RT-PCR amplification performed as well as or better than other classification schemes. In particular, the RFU-based classification scheme outperformed the Bio-Rad default scheme. The Bio-Rad default defines an RFU cutoff based on a user-specified percentage of the range between the negative controls and the highest positive test. For this reason, the Bio-Rad default cutoffs are highly dependent on positive controls. In our data, RFU signals that differ from those of positive controls or a test sample recording a higher RFU than a control can render comparisons between plates problematic (data not shown). A cutoff for RT-PCR18S positivity of 650 RFU was chosen because the lowest microscopy-positive sample among the Tanzanian samples that was higher than all the negative JHH controls had a fluorescence of 652.35. A priori, we expected to set the RT-PCRcyto threshold for positivity higher than that of RT-PCR18S because cytochrome b has about 40 copies per parasite compared to 2 or 3 for 18S (9, 11), and this difference is maintained throughout asexual replication (7). However, RT-PCRcyto had lower negative controls, with the exception of a potential outlier with an RFU of >900. Both RT-PCRcyto and RT-PCR18S were run in the same Cy5 fluorescent probe to maintain parsimony and comparability between targets, both of whose thresholds for positivity were set at 650 RFU.

Based on the JHH dilution series and the Tanzania microscopy, RT-PCR18S produced a tighter standard curve (r2 = 92%) than did RT-PCRcyto (r2 = 52%). RT-PCR18S correctly classified more low-density infections and negative samples than did RT-PCRcyto, resulting in higher estimates of sensitivity (96%) and specificity (94%) for RT-PCR18S than for RT-PCRcyto (sensitivity, 87%; specificity, 84%); the difference in specificity is statistically significant (P < 0.05). In the LCA analysis, RT-PCR18S consistently showed better sensitivity and specificity than did RT-PCRcyto, microscopy, or RDT when the tests were examined in the same model. The point estimates and confidence intervals for RT-PCR18S may have been artificially inflated in models with RDT and RT-PCRcyto because of violations of the assumption of local or conditional independence. All LCA assumes that the observed measures of the latent trait, i.e., PCR, RDT, etc., are independent and conditional on the latent trait, i.e., true malaria infection. We may violate this assumption, because within the group of truly negative blood samples, those participants who recently experienced P. falciparum infection may still test positive by PCR for a short time after clearing infection and for a longer period by RDT. If RT-PCR18S correlated with another test for reasons other than true latent malaria infection, then RT-PCR18S specificity and sensitivity estimates could appear artificially high. Separating RT-PCR18S and RT-PCRcyto may minimize violation of the local independence assumption. Despite this weakness, we believe that the consistency with which RT-PCR18S outperforms other tests in the LCA supports RT-PCR18S as the superior diagnostic test for these data. On the other hand, RT-PCRcyto produced CT values that were 1.85 cycles lower than those of RT-PCR18S, and the lower CT values are consistent with higher starting DNA quantities. The cytochrome b gene is part of a small, 6-kb, exonuclear gene that codes for the Plasmodium mitochondria. During DNA isolation, a greater proportion of the DNA may be lost when starting quantities are low, which may explain RT-PCRcyto's poor performance in detection of low-density infections compared to RT-PCR18S. The 18S target may be a more reliable DNA target for RT-PCR; however, further study is needed to clarify the possible paradox of RT-PCRcyto producing a higher signal but failing to detect low-density infections.

Concordance between RT-PCR, microscopy, and RDT has varied across settings and studies (8, 9, 12). In a study published in 2009, Satoguina et al. found P. falciparum rates by RT-PCR were 25.5% but only 8.5% by microscopy and 1.1% by RDT in The Gambia and Guinea Bissau (7). RDT specificity has also been questioned, with a report of 25.9% of the OptiMal tests producing false positives. Other research has suggested the sensitivity of RDT is as high as 90% when parasite densities are >100 parasites/μl (12) or up to 89% in pregnant women (8). In our data, the prevalence by RDT was 12.0% compared to 9.1% by RT-PCR18S and 4.4% by microscopy. Despite calling fewer positives, the ICC between RT-PCR18S and microscopy was 43.2% compared to only 23.8% between RT-PCR18S and RDT and 25.8% between RDT and microscopy. The high ICC between RT-PCR18S and microscopy supports the high sensitivity (100%) and specificity (98%) estimated in the LCA. The low concordance with RDT and the other tests and the related low specificity by LCA suggest that RDT may have produced false positives.

A potential explanation for the variation in RDT performance between studies is that under hypoendemicity the positive predictive value of any test suffers. If prevalence is under 5% and a test has less than 95% specificity, then one can expect half the positive test results to be false positives. False positives reduce concordance between tests because false-positive RDTs may be independent of false positives by RT-PCR and microscopy. Furthermore, hypoendemicity may be associated with low parasite densities (2), and test sensitivity decreases with decreasing parasite densities. One would expect concordance between malaria diagnostics to be lowest in regions of hypoendemicity, where prevalence and parasite densities are lowest. Support for this explanation may lie within our data, because we found agreement between tests peaks when prevalence and parasite densities are high (Fig. 5).

Fig. 5.
Venn diagram for positive P. falciparum tests. RT-PCR18S (left), microscopy (Micro) (center), and RDT (right) are shown (13).

In many areas where malaria is endemic, patients presenting in clinics with fever are often treated for malaria without diagnostic confirmation by RDT or microscopy. Malaria overdiagnosis wastes malaria drugs and discourages clinicians from seeking correct treatments for potentially life-threatening bacterial infections. Microscopy is labor-intensive, but RDT has been suggested as a field-ready diagnostic that can confirm positive or negative clinical diagnoses when parasitemias are high. However, our results highlight the imperfections of RDTs as a prerequisite for a positive malaria diagnosis. In our data, agreement between microscopy, RT-PCR, and RDT was low, and RDT had lower sensitivity and specificity than the other tests in an LCA. Our results suggest that RDTs are useful, but requiring a positive RDT before administering antimalarial drugs could unnecessarily leave some malaria infections untreated.

Our results support the contention that RDTs have the potential for false negatives and false positives in regions of hypoendemicity. These data underscore the need to use highly sensitive molecular diagnostics in large epidemiology studies and the need for the continued development of highly sensitive, low-cost diagnostics that can be employed in less-developed countries. Although RT-PCR18S outperformed RDT, microscopy, and RT-PCRcyto, we hesitate to call RT-PCR18S a gold standard or to view RT-PCR18S as a perfect test for P. falciparum because RT-PCR18S failed to consistently detect low parasite densities (<1,000 parasites/μl).


We thank Edward Sambu and Fikirini Msuya (laboratory technologists, Amani Centre) for their assistance with training of local field staff and for reading peripheral blood smears. We thank Stephen Magesa (Director, Amani Centre) for technical support.

The study was supported in part by the Bill and Melinda Gates Foundation (National Institute of Medical Research, Supplemental Project to Partnership for the Rapid Elimination of Trachoma, PRET+ [number 48027]; Sheila West, principal investigator) and by NIAID U01 AI068613 (HIV Prevention Trials Network—Laboratory Network).


Published ahead of print on 31 August 2011.


1. Beath K.randomLCA: random effects latent class analysis. R package version 0.8-3. 2011. http://CRAN.R-project.org/package=randomLCA.
2. Bødker R., Msangeni H. A., Kisinza W., Lindsay S. W. 2006. Relationship between the intensity of exposure to malaria parasites and infection in the Usambara Mountains, Tanzania. Am. J. Trop. Med. Hyg. 74:716–723 [PubMed]
3. Brenner H., Gefeller O. 1997. Variation of sensitivity, specificity, likelihood ratios and predictive values with disease prevalence. Stat. Med. 16:981–991 [PubMed]
4. Maeno Y., et al. 2008. A dried blood sample on filter paper is suitable for detecting Plasmodium falciparum gametocytes by reverse transcription polymerase chain reaction. Acta Trop. 107:121–127 [PubMed]
5. Okell L. C., Ghani A. C., Lyons E., Drakeley C. J. 2009. Submicroscopic infection in Plasmodium falciparum-endemic populations: a systematic review and meta-analysis. J. Infect. Dis. 200:1509–1517 [PubMed]
6. R Development Core Team. 2008. A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.
7. Satoguina J., et al. 2009. Comparison of surveillance methods applied to a situation of low malaria prevalence at rural sites in The Gambia and Guinea Bissau. Malar. J. 8:274. [PMC free article] [PubMed]
8. Singer L. M., et al. 2004. Evaluation of a malaria rapid diagnostic test for assessing the burden of malaria during pregnancy. Am. J. Trop. Med. Hyg. 70:481–485 [PubMed]
9. Steenkeste N., et al. 2009. Towards high-throughput molecular detection of Plasmodium: new approaches and molecular markers. Malar. J. 8:86. [PMC free article] [PubMed]
10. Stevenson M. S., Nunes T. N., Thornton R. T. epiR: functions for analysing epidemiological data. R package version. 2010
11. Vaidya A. B., Akella R., Suplick K. 1989. Sequences similar to genes for two mitochondrial proteins and portions of ribosomal RNA in tandemly arrayed 6-kilobase-pair DNA of a malarial parasite. Mol. Biochem. Parasitol. 35:97–107 [PubMed]
12. van Dijk D. P., et al. 2009. Evaluation of the Palutop+4 malaria rapid diagnostic test in a non-endemic setting. Malar. J. 8:293. [PMC free article] [PubMed]
13. Wilkinson L.venneular: Venn and Euler diagrams. R package version 1.1-0. 2011. http://CRAN.R-project.org/package=venneuler.

Articles from Journal of Clinical Microbiology are provided here courtesy of American Society for Microbiology (ASM)
PubReader format: click here to try


Save items

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


  • MedGen
    Related information in MedGen
  • PubMed
    PubMed citations for these articles

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...