Meta-Analysis: Accuracy of Rapid Tests for Malaria in Travelers Returning from Endemic Areas

Context What is the accuracy of rapid tests for detecting malaria? Contribution This meta-analysis of 21 studies in nonimmune travelers with suspected malaria summarizes the accuracy of various rapid tests compared with expert microscopic examination or polymerase chain reaction tests. Histidine-rich protein-2 (HRP-2) tests had high sensitivities (range, 88% to 99%) and specificities (range, 95% to 100%) for detecting Plasmodium falciparum malaria. Positive 3-band HRP-2 test results detected malaria better than 2-band HRP-2 tests. Negative HRP-2based test results excluded malaria better than parasite lactate dehydrogenasebased tests. Cautions Few studies evaluated 3-band HRP-2 tests for malaria species other than P. falciparum. The Editors Imported malaria is an increasing problem and causes considerable morbidity in the United States and other industrialized countries. Few travelers use recommended chemoprophylaxis and insect protection measures (1), which means that malaria needs to be ruled out in febrile patients who have recently returned from an endemic country. Microscopic examination of thick blood smears is the diagnostic gold standard, but it is accurate only if performed by experienced microscopists. In a Canadian survey (1), community-based microscopic diagnosis provided incorrect species identification in 64% of cases. Polymerase chain reaction (PCR) is a highly sensitive alternative to microscopy, but the infrastructure and expertise required preclude its routine use in many health care settings (2). The time until results from microscopic examinations become available can substantially delay provision of appropriate therapy to patients with malaria who are treated in centers without expertise in tropical medicine (1). Such delays may be detrimental, particularly in infections caused by Plasmodium falciparum, which can be rapidly fatal (3, 4). Tests that allow bedside diagnosis of malaria have been introduced in recent years (5). By using immunochromatographic methods, these tests detect parasite antigens in lysed blood from a fingerprick blood sample and can be performed in approximately 15 minutes without special equipment (2). Two-band tests target histidine-rich protein-2 (HRP-2) and detect P. falciparum only, while the more recent 3-band tests also detect other malaria parasites (P. vivax, P. malariae, and P. ovale) by using HRP-2 and aldolase combined or parasite lactate dehydrogenase (LDH) as targets. Licensing authorities in Canada and some European countries, but not the United States, have approved rapid diagnostic tests. Currently, it is unclear whether rapid tests are an accurate alternative to established laboratory-based methods in industrialized countries and whether diagnostic accuracy differs among different types of tests. We performed a systematic review and meta-analysis of test accuracy studies to determine the ability of different rapid tests to rule out malaria in nonimmune individuals with suspected malaria. Methods Identification and Selection of Studies We searched MEDLINE, EMBASE, CAB Health, and CINAHL (1988 to September 2004), combining free-text terms (plasmodi* OR malaria* OR falcipar* OR vivax* OR ovale*) with a search strategy for diagnostic studies described elsewhere (6); hand-searched conference proceedings; checked references of obtained papers; and contacted experts and manufacturers. Details on the search strategy are available from the authors. We included diagnostic accuracy studies in nonimmune individuals with suspected malaria (that is, travelers returning from malaria-endemic areas), which compared rapid tests with microscopic examination or PCR as the reference standard and presented 2 2 contingency tables or data allowing their construction. We excluded studies in which more than 10% of individuals were immune and studies that determined the accuracy of self-administered rapid tests. We applied no language restrictions (7). If more than 1 test based on the same target antigen had been evaluated in the same patient population for diagnosis of the same parasite, we avoided duplication of data by applying predefined rules: We gave precedence to 3-band tests over 2-band tests and to the evaluation that was based on more patients if only 2-band tests had been evaluated. If sample sizes were identical, we selected the more recently developed test. Data Extraction and Quality Assessment Two observers independently extracted data on test characteristics, study setting, and patients and assessed the components of methodologic quality that may be associated with bias in test accuracy studies (8): prospective design; relevant clinical population (as opposed to diagnostic casecontrol study); enrollment of a series of consecutive patients with suspected malaria; all patients undergoing reference testing; performance and interpretation of the index test without knowledge of the reference test results; and performance and interpretation of the reference test without knowledge of the index test results. We used a piloted, standardized data extraction sheet and resolved disagreements by consensus. Definition of Reference Test and Main Outcome The reference standard consisted of a combination of results from microscopic examination and PCR, with patients considered positive if 1 of the 2 test results was positive. We gave PCR results precedence over microscopic examination if combined results were unavailable. In some studies, additional tests, such as quantitative buffy coat, were performed when microscopy results were unclear. We considered patients with pure gametocytemia to be positive. We defined a priori the likelihood ratio for a negative test result (negative likelihood ratio) for P. falciparum infection as the primary measure of accuracy. The negative likelihood ratio indicates how much less likely it is to find a negative result in individuals with malaria as compared with those without infection (9). Conversely, the positive likelihood ratio specifies how much more likely it is to find a positive test result in individuals with malaria as compared with individuals without malaria. In our main analysis, we considered individuals infected with a Plasmodium species other than P. falciparum to be negative. We excluded individuals infected with P. falciparum in analyses restricted to 3-band tests and calculated the negative likelihood ratio for infections with P. vivax and for infections with P. malariae or P. ovale. We also calculated positive likelihood ratios and plotted the sensitivity of tests against 1 specificity. Statistical Analysis We used a random-effects model to combine estimates of likelihood ratios (10). We planned analyses stratified by the antigen targeted by index tests (HRP-2 or parasite LDH) and by test generation (2-band tests, first-generation 3-band tests, or second-generation 3-band tests). We also analyzed subgroups of relevant studies to determine the association between the negative likelihood ratio and parasite density. For the diagnosis of P. falciparum, we examined funnel plots (11) of negative likelihood ratios by plotting likelihood ratios against total sample size (12, 13). In addition, we used meta-regression models (14) to quantify the association of the negative likelihood ratio with the type of target antigen, components of methodologic quality, and study size. We calculated the I2 statistic, which describes the percentage of total variation across studies that is due to heterogeneity rather than chance (15): I2 = 100% (Q df)/Q, where Q is Cochran's heterogeneity statistic and df is the degrees of freedom. Mild heterogeneity will account for less than 30% of the variation, and pronounced heterogeneity will account for substantially more than 50%. Four studies included direct comparisons of HRP-2 and parasite LDH tests (16-19). We separately compared negative likelihood ratios of HRP-2 and parasite LDH tests for each of these studies and subsequently pooled these comparisons by using a random-effects model. Finally, we determined pretest probabilities for individuals returning from different continents on the basis of prevalences found in test accuracy studies and malaria surveillance data from the Centers for Disease Control and Prevention (CDC) (20). We used the programs meta and metareg in Stata, version 8.2 (Stata Corp., College Station, Texas), for random-effects meta-analysis and meta-regression analysis. Role of the Funding Source The Swiss Federal Office of Public Health had no role in the design, conduct, or reporting of the study or in the decision to submit the manuscript for publication. Results Literature Search We screened 5689 citations and identified 308 potentially eligible reports. We excluded 165 articles on the basis of their abstracts and obtained the full-text articles for the remaining 143 reports. Of these, we excluded 118 reports. The most common reasons for exclusion were immune study populations and ineligible study designs (Figure 1). We analyzed 22 published reports (2, 16-19, 21-37) and 3 unpublished reports (38); Hernandez E, unpublished; de Monbrison F, unpublished) on 21 studies and 5747 individuals with suspected malaria. Figure 1. Identification of 21 eligible test accuracy studies. Characteristics of Studies and Tests Table 1 presents the characteristics of included studies. Three studies were unpublished (38; Hernandez E, unpublished; de Monbrison F, unpublished), 3 were published in French (17, 22, 24), 1 was published in Italian (34), and the remaining studies were published in English. Four studies were from North America (21, 25, 36, 37), 1 study was from Australia (19), and the other studies were from Europe. Travel destinations were available for 3 Canadian studies, which reported that 58% to 64% of individuals had traveled in Africa, 18% to 21% in Asia, and 14% to 19% in Central and South America, and from 1 French study (18), which reported that 75% of patients had returned from Af