Nucleic acid and antigen detection tests for leptospirosis (2024)

PMCID: PMC6699653

Marga GA Goris, Email: ln.cmumadretsma@sirog.m.

Author information Copyright and License information PMC Disclaimer

This article is an update of with doi:10.1002/14651858.CD011871.

Background

Early diagnosis of leptospirosis may contribute to the effectiveness of antimicrobial therapy and early outbreak recognition. Nucleic acid and antigen detection tests have the potential for early diagnosis of leptospirosis. With this systematic review, we assessed the sensitivity and specificity of nucleic acid and antigen detection tests.

Objectives

To determine the diagnostic test accuracy of nucleic acid and antigen detection tests for the diagnosis of human symptomatic leptospirosis.

Search methods

We searched electronic databases including MEDLINE, Embase, the Cochrane Library, and regional databases from inception to 6 July 2018. We did not apply restrictions to language or time of publication.

Selection criteria

We included diagnostic cross‐sectional studies and case‐control studies of tests that made use of nucleic acid and antigen detection methods in people suspected of systemic leptospirosis. As reference standards, we considered the microscopic agglutination test alone (which detects antibodies against leptospirosis) or in a composite reference standard with culturing or other serological tests. Studies were excluded when the controls were healthy individuals or when there were insufficient data to calculate sensitivity and specificity.

Data collection and analysis

At least two review authors independently extracted data from each study. We used the revised Quality Assessment of Diagnostic Accuracy Studies tool (QUADAS‐2) to assess risk of bias. We calculated study‐specific values for sensitivity and specificity with 95% confidence intervals (CI) and pooled the results in a meta‐analysis when appropriate. We used the bivariate model for index tests with one positivity threshold, and we used the hierarchical summary receiver operating characteristic model for index tests with multiple positivity thresholds. As possible sources of heterogeneity, we explored: timing of index test, disease prevalence, blood sample type, primers or target genes, and the real‐time polymerase chain reaction (PCR) visualisation method. These were added as covariates to the meta‐regression models.

Main results

We included 41 studies evaluating nine index tests (conventional PCR (in short: PCR), real‐time PCR, nested PCR, PCR performed twice, loop‐mediated isothermal amplification, enzyme‐linked immunosorbent assay (ELISA), dot‐ELISA, immunochromatography‐based lateral flow assay, and dipstick assay) with 5981 participants (1834 with and 4147 without leptospirosis). Methodological quality criteria were often not reported, and the risk of bias of the reference standard was generally considered high. The applicability of findings was limited by the frequent use of frozen samples. We conducted meta‐analyses for the PCR and the real‐time PCR on blood products.

The pooled sensitivity of the PCR was 70% (95% CI 37% to 90%) and the pooled specificity was 95% (95% CI 75% to 99%). When studies with a high risk of bias in the reference standard domain were excluded, the pooled sensitivity was 87% (95% CI 44% to 98%) and the pooled specificity was 97% (95% CI 60% to 100%). For the real‐time PCR, we estimated a summary receiver operating characteristic curve. To illustrate, a point on the curve with 85% specificity had a sensitivity of 49% (95% CI 30% to 68%). Likewise, at 90% specificity, sensitivity was 40% (95% CI 24% to 59%) and at 95% specificity, sensitivity was 29% (95% CI 15% to 49%). The median specificity of real‐time PCR on blood products was 92%. We did not formally compare the diagnostic test accuracy of PCR and real‐time PCR, as direct comparison studies were lacking. Three of 15 studies analysing PCR on blood products reported the timing of sample collection in the studies included in the meta‐analyses (range 1 to 7 days postonset of symptoms), and nine out of 16 studies analysing real‐time PCR on blood products (range 1 to 19 days postonset of symptoms). In PCR studies, specificity was lower in settings with high leptospirosis prevalence. Other investigations of heterogeneity did not identify statistically significant associations. Two studies suggested that PCR and real‐time PCR may be more sensitive on blood samples collected early in the disease stage. Results of other index tests were described narratively.

Authors' conclusions

The validity of review findings are limited and should be interpreted with caution. There is a substantial between‐study variability in the accuracy of PCR and real‐time PCR, as well as a substantial variability in the prevalence of leptospirosis. Consequently, the position of PCR and real‐time PCR in the clinical pathway depends on regional considerations such as disease prevalence, factors that are likely to influence accuracy, and downstream consequences of test results. There is insufficient evidence to conclude which of the nucleic acid and antigen detection tests is the most accurate. There is preliminary evidence that PCR and real‐time PCR are more sensitive on blood samples collected early in the disease stage, but this needs to be confirmed in future studies.

Reference standard

MAT is the most widely used serological test for leptospirosis. It is considered to be the reference standard, often used in combination with other serological tests (such as IgM ELISA), and with or without culture of leptospires from blood or urine.

MAT is considered an imperfect reference standard. It has a high diagnostic specificity, as the observation of seroconversion or a titre rise confirms current leptospirosis, but a negative MAT does not rule out the possibility of leptospirosis. Limmathurotsakul and colleagues used a Bayesian latent class analysis (LCA) to estimate the accuracy of MAT, which was 49.8% sensitive and 98.8% specific (Limmathurotsakul 2012). The LCA assumes that there is no reference standard, and estimates disease prevalence by taking the results of multiple tests into account (Rutjes 2007).

In another study to estimate the accuracy of MAT, Goris and colleagues selected culture‐positive people as being infected (proof of leptospirosis) and people with other known diseases and unknown disease as controls, and performed MAT on both groups (Goris 2012). In this study, the sensitivity of MAT was estimated at 81.7% and specificity of MAT was estimated at 100%.

Using a reference standard with low sensitivity to compare against the index test may result in biased estimates of specificity. However, when the case definition in the Goris 2012 study was changed to include people who were IgM ELISA positive, the sensitivity increased to 93.3% without sacrificing the specificity (Goris 2012). This indicates that combining multiple tests with high specificity as a composite reference standard can yield increased sensitivity. Therefore, we decided to include studies with only MAT as the reference standard, and studies that used other serological tests, or culturing, or both, alongside MAT as the reference standard.

Variability in MAT performance between laboratories exists and may affect test accuracy. MAT requires a panel of live Leptospira serovars (group of micro‐organisms characterised by specific set of antigens) that occur in the region, supplemented with a panel of globally standardised serovars when people present with a travelling history (Goris 2012). Determining and maintaining such panels are major, but essential, tasks; inadequate panels may lead to false‐negative results. The timing of sample collection may also influence sensitivity or specificity; antibodies are usually detectable from five to seven DPO onwards. MAT‐case definitions may vary between laboratories; a four‐fold rise in titre in paired sera or seroconversion is indicative of current infection, but some laboratories may use a high titre in a single serum sample (seropositivity) as a case definition for people who do not return for follow‐up. Seropositivity is not necessarily evidential of a current infection, since antibodies may persist after a previous infection, or cross‐reactivity with other diseases may occur (such as legionellosis, hepatitis, and autoimmune diseases) (WHO 2003). Therefore, the desirable cut‐off titres for the single‐sample MAT are higher in regions where leptospirosis and similar infectious diseases are highly prevalent.

Leptospires can be cultured from blood, CSF, dialysate fluid, and (postmortem) tissue, often within 10 DPO. Culture of urine is useful after 10 DPO. Leptospires are slow‐growing, fastidious bacteria. Cultures have to be maintained for at least four months before being regarded negative. Culturing provides evidence for leptospirosis but lacks sensitivity and does not contribute to an early diagnosis. The sensitivity of culture is estimated not to exceed 23%, according to an analysis of people with leptospirosis from 1925 to 2008 in the Netherlands (Goris 2013).

Types of studies

We included diagnostic test accuracy studies, that is, any study that evaluated the sensitivity and specificity of a nucleic acid or antigen detection test in comparison with a reference standard. In this review, we discerned three types of eligible diagnostic test accuracy studies based on their method of participant selection: the cross‐sectional study, the single‐gate case‐control study, and the two‐gate case‐control study. Their respective characteristics are summarised in Table 5 and illustrated in Figure 2.

In cross‐sectional studies, people with clinical suspicion of leptospirosis are consecutively enrolled and undergo both the index test and reference standard. In the similar single‐gate case‐control study, usually all people with positive reference standard results and a subsample of people with negative reference standard results from an original clinically suspected cohort are subsequently tested with the index test. We referred to these two study designs simply as 'single‐gate' designs (i.e. having a single inclusion criteria for clinical presentation) (Rutjes 2005). The main difference between these two designs is that the prevalence of the target condition in the single‐gate case‐control study is artificial, whereas in a cross‐sectional study, a true prevalence can be estimated.

In a two‐gate case‐control design, people with positive reference standard results and people who do not have leptospirosis are enrolled to subsequently undergo the index test. Since the participants with and without the target condition are selected from two separate cohorts, this study design is at a higher risk of bias in comparison to the single‐gate designs. The two‐gate case‐control designs can be further separated into studies in which the controls have an alternative condition resembling leptospirosis (two‐gate with alternative diagnoses controls), and studies in which the controls are healthy (two‐gate with healthy controls).

We excluded two‐gate case‐control designs with healthy controls because these studies are known to produce inflated estimates of diagnostic accuracy (Rutjes 2005).

We placed no restrictions on language and publication date. When studies met our eligibility criteria but reported insufficient data for the construction of two‐by‐two tables, we excluded them. In cases where the full‐text article was not retrievable or in case of meeting abstracts, we included the study if the abstract reported data for the construction of two‐by‐two tables. We excluded studies when both abstract and the full‐text article were not retrievable. We contacted study authors to obtain the full‐text article or study data prior to exclusion, but we excluded studies when no answer was obtained after a lengthy interval. Finally, we excluded studies with fewer than 10 participants, as they would add little value to the review.

Participants

Eligibility of participants depended on the study design.

• Cross‐sectional studies: people with clinical suspicion of leptospirosis were eligible. Compatible symptoms were, but were not limited to, fever, myalgia, headaches, malaise, conjunctival suffusion, rash, nausea/vomiting, anorexia, and cough.

• Single‐gate case‐control studies: eligible were cases with a positive reference standard result and controls derived from the same clinically suspected group as the cases, but with a negative reference standard result.

• Two‐gate case‐control studies: eligible were cases with a positive reference standard result and controls with a different known disease that resembled the clinical presentation of leptospirosis.

We excluded studies that screened asymptomatic people for leptospirosis.

Index tests

All diagnostic tests that used nucleic acid and antigen detection methods were included. Tests eligible for inclusion were, but were not limited to, PCR and its variants, that is, LAMP, ELISA, FAT, silver staining, or immunohistochemistry. We included index tests with any sample type (e.g. blood products, urine, CSF), any timing of sample collection (recorded as DPO), any variation in laboratory processing, and any threshold for tests on a continuous scale. We excluded studies that did not analyse different sample types separately, as it would be unclear which sample should be tested by the clinician in order to obtain a similar test accuracy.

Target conditions

This review was restricted to human symptomatic leptospirosis. We excluded studies of ocular and neurological manifestations of leptospirosis, as it was unclear whether MAT was a valid reference standard for these target conditions.

Reference standards

We considered several types of reference standards, which are summarised in Table 6. We elaborate the inclusion criteria separately for single‐gate and two‐gate designs.

For single‐gate designs, we considered studies that used MAT, with or without culture or other serological tests such as IgM ELISA. We included these tests alongside MAT in order to compensate for the imperfect sensitivity of MAT as a reference standard. Since these tests have high specificity, we considered any positive result from this composite reference standard as a leptospirosis case. If a study used MAT as a sole reference standard, we considered the risk of bias to be high. We excluded single‐gate designs with culture as the sole reference standard, since culture has a very low sensitivity (Goris 2013).

In two‐gate designs, the people without leptospirosis are not necessarily reference standard negatives, but they are diagnosed with an alternative condition. Hence, we only required a reference standard that ruled in leptospirosis in the case of a positive result. Reference standards considered eligible for studies with this design were those with a high specificity: MAT used alone, or culture used alone.

In order to avoid incorporation bias (the reference standard uses or incorporates the index test), we excluded studies which contained a nucleic acid or antigen detection test in the reference standard.

Electronic searches

We searched the following 16 electronic databases: the Cochrane Library (6 July 2018), MEDLINE Ovid (1946 to 6 July 2018), Embase Ovid (1974 to 6 July 2018), Web of Science (1975 to 6 July 2018), CINAHL (1937 to 6 July 2018), BIOSIS Previews (1993 to 8 February 2015 due to terminated institutional subscription), PubMed (for publications not yet included in MEDLINE; 1946 to 8 February 2015), Google Scholar, African Index Medicus (1993 to 6 July 2018), African Journals Online (from inception to 8 February 2015), LILACS (Literature in the Health Sciences in Latin America and the Caribbean, 1982 to 6 July 2018), KoreaMed (from inception to 8 February 2015), IMSEAR (Index Medicus for the South‐East Asian Region, from inception to 6 July 2018), IMEMR (Index Medicus for the Eastern Mediterranean Region, from inception to 8 February 2015), WPRIM (Western Pacific Region Index Medicus, from inception to 6 July 2018), and IndMed (from inception to 8 February 2015). For each database, we identified subject headings or free‐text terms and synonyms (or both) related to: leptospirosis, antigen, nucleic acids, PCR, LAMP, hybridisation, immunohistochemistry, silver staining, and dot blot. Appendix 1 shows the search strategies for each database.

Searching other resources

Additionally, we scanned the reference lists of included articles and we searched the World Health Organization's (WHO) International Clinical Trial Registry Platform (www.who.int/ictrp) for ongoing or unpublished trials.

Selection of studies

Two review authors (BY, MG) independently screened the titles and abstracts of all records, and excluded records with no relevance to the review question (first sift). We retrieved the full‐text of the remaining records, and three review authors (BY, MG, SdV) independently checked the full‐text articles for eligibility, using a full‐text assessment checklist, with each record being assessed by at least two review authors (second sift). Studies that were excluded during data extraction, excluded meeting abstracts and studies with irretrievable full‐texts, are listed in the Characteristics of excluded studies tables. We resolved disagreements between review authors by consensus or by consulting a senior author (ML).

Data extraction and management

From each study, two out of three review authors (BY, MG, SdV) independently extracted data by using a specially designed data collection form. The data collection form contained the following items.

• Study ID.

• Study design.

• Study region.

• Regional prevalence.

• Participants: selection methods, sex and age distribution, symptoms, risk factors.

• Index tests: threshold values, timing of sample collection (defined as DPO of symptoms where 1 DPO was 0 to 24 hours after onset of symptoms); type of sample.

• Reference standards: threshold values, timing of sample collection.

• Two‐by‐two contingency table for sensitivity and specificity calculations.

Each of the three review authors first piloted the form on two included studies to check for applicability. We resolved discrepancies between the authors by discussion and consensus. We contacted study authors for missing information.

Assessment of methodological quality

We assessed the quality of included studies using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS‐2) tool (Whiting 2011). The QUADAS‐2 tool helps quality assessment by assessment of risk of bias and applicability of results across four domains: participant selection, index test, reference standard, and flow and timing. We custom‐tailored QUADAS‐2 to the needs of our review by adding additional signalling questions where needed. We also piloted the tool on two included studies and refined it accordingly. See Appendix 2 for the signalling questions and review‐specific guidance.

Statistical analysis and data synthesis

We arranged results from each study in two‐by‐two contingency tables in which we compared people with confirmed leptospirosis (as defined by a positive MAT or other serological test or culture result) and people without leptospirosis (none of the reference standard tests were positive, or people having another disease than leptospirosis) to the binary test results from the index tests. From these tables, we calculated sensitivity and specificity for each study. As previously described, we excluded studies reporting insufficient data for the construction of two‐by‐two tables.

Some studies reported two thresholds for MAT, where the higher threshold was considered 'confirmed leptospirosis', and the lower threshold was considered 'probable leptospirosis'. In these studies, we chose the higher threshold dataset for the primary analysis. This was because we considered the specificity of the reference standard to be more important than its sensitivity. The lower threshold dataset was analysed in a sensitivity analysis.

We presented individual study results graphically by plotting estimates of sensitivity and specificity in forest plots and the summary receiver operating characteristic (SROC) space. We conducted a random‐effects meta‐analysis using the bivariate model to estimate summary values for sensitivity and specificity when little variation in threshold values was presumed. If studies used multiple thresholds for the index test, we constructed a SROC curve using the hierarchical summary receiver operating characteristic (HSROC) model. All analyses were done in SAS 9.4 (Cary Inc.).

We separately described studies that reported head‐to‐head comparisons of index tests (or index test characteristics) in the same study population, but did not perform meta‐analyses to formally compare these index tests due to the lack of a sufficient number of studies.

Investigations of heterogeneity

We assessed heterogeneity initially by visually inspecting the forest plots and the ROC plot. The following covariates were investigated as potential sources of heterogeneity.

• Timing of sample collection for the index test. We planned to analyse this based on how study authors reported the timing: as a continuous variable using medians or means, or as a categorical variable using timing intervals (e.g. 1 DPO to 4 DPO versus 5 DPO to 10 DPO).

• Prevalence in the study population (continuous variable). This was computed using two‐by‐two table data from cross‐sectional studies. If a case‐control study reported prevalence data of the original cohort, we also used these data.

• Blood sample type for the index test (categorical variable; whole blood, plasma, or serum).

• Primers or target genes for the PCR and other nucleic acid tests (categorical variable). Since two PCRs with the same target gene could use different primers, we also specified the original reference of the technique.

• Threshold of the index test, if applicable (continuous variable; e.g. threshold cycles (Ct) for the real‐time PCR).

• Real‐time PCR visualisation method (categorical variable; TaqMan probe; or SYBR green).

• Brand of the test, if applicable (categorical variable).

Sensitivity analyses

To examine the robustness of the results to the decisions we made in the review process, we conducted analyses with the following alternative decisions.

• Exclusion of studies with only abstracts.

• Exclusion of studies with high risk of bias for the 'patients' domain.

• Exclusion of studies with high risk of bias for the 'reference standard' domain.

• Exclusion of studies that used antibiotics before the index test.

• The choice of the lower MAT threshold dataset for the analysis, in studies that reported two thresholds for MAT.

Description of studies

Risk of bias

Concerns regarding applicability of results to clinical practice

Conventional polymerase chain reaction

Seventeen studies reported test accuracy data for the PCR. Fifteen studied PCR on blood products (serum: nine studies: whole blood (ethylenediaminetetraacetic acid (EDTA)): four studies; blood or serum: one study; unspecified blood product: one study) and of them, three also studied PCR on urine. Two studies included exclusively urine samples. The sensitivity of PCR on blood products ranged from 13% to 100%, and the specificity from 0% to 100% (see Figure 10). The 12 studies analysing PCR on blood products did not report the timing of sample collection for the index test (further referred to as DPO). Three studies did report the DPO: one reported a mean of five days (Vanasco 2016), one reported a range of one to five days (Zhang 1992), and one a range of one to seven days (Céspedes 2007).

The sensitivity of PCR on urine ranged from 22% to 57%, and the specificity from 56% to 100% (see Figure 11).

Real‐time polymerase chain reaction

Eighteen studies assessed the accuracy of the real‐time PCR. Sixteen studies used blood products as sample type: serum or plasma (nine studies), whole blood (six studies), serum or whole blood (three studies), blood culture samples (one study), and buffy coat samples (one study). Two studies used urine samples. Six studies each reported two sets of data: Agampodi 2012 and Riediger 2017 reported data for whole blood and serum, while Backstedt 2015, Thaipadungpanit 2011, and Woods 2018 reported data for two real‐time PCRs, each using a different target gene (rrs, lipL32, or rrs/lipL32). Waggoner 2014 reported data for a monoplex and multiplex (detecting also dengue and malaria) real‐time PCR. Ahmed 2009 reported three sets of data, each evaluating the test at different DPOs (1 DPO to 4 DPO, 5 DPO to 10 DPO, and 1 DPO to 10 DPO). Because we considered 1 DPO to 10 DPO to be the most representative time of sample collection, we included the dataset of 1 DPO to 10 DPO in the meta‐analysis. Seven studies of real‐time PCR on blood products did not report the DPO (Gonzalez 2013; Sonthayanon 2013; Backstedt 2015; Waggoner 2015; Biscornet 2017; Riediger 2017; Pakoa 2018). The other studies all reported DPOs of the index test under 10 days.

The sensitivity of real‐time PCR on blood products ranged from 0% to 100%, and the specificity ranged from 0% to 100% (Figure 14).

Two studies assessed real‐time PCR on urine (Villumsen 2012 U; Woods 2018). One study assessed real‐time PCR on blood culture samples (Villumsen 2012 BC), and one study assessed real‐time PCR on buffy coat samples (Woods 2018). Due to the lack of enough studies for urine, blood culture, and buffy coat real‐time PCR, we did not perform a meta‐analysis. The sensitivities and specificities of these PCRs are displayed in Table 10.

Nested polymerase chain reaction

Four studies reported accuracy data for the nested PCR (Merien 2005; Koizumi 2009; Blanco 2014; Gokmen 2016). All were cross‐sectional studies using serum as the sample type, and all studies used MAT as the reference standard. The reported mean timing of sample collection was 5 DPO (Merien 2005) and 7 DPO (Koizumi 2009), but was not reported for Blanco 2014 and Gokmen 2016. The sensitivity of nested PCR ranged from 0% (95% CI 0% to 13%) to 95% (95% CI 76% to 100%) and the specificity ranged from 42% (95% CI 23% to 63%) to 100% (95% CI 99% to 100%) (see Figure 16). Since only four studies were available, we did not conduct a meta‐analysis or formal assessments of heterogeneity.

Conventional polymerase chain reaction performed twice

Two studies reported data for PCR done on serum samples taken at two different times (Yersin 1998; Seng 2007). The PCR was considered positive if one of the two samples was positive. The first sample was taken at admission and the second sample approximately 14 days later. Both studies were cross‐sectional. One study used a composite reference standard (MAT and culturing; Seng 2007), while the other used only MAT (Yersin 1998). Seng 2007 reported sensitivity of 75% (95% CI 19% to 99%) and specificity of 94% (95% CI 88% to 98%), while Yersin 1998 reported sensitivity of 47% (95% CI 34% to 60%) and specificity of 96% (95% CI 87% to 100%).

Loop‐mediated isothermal amplification

Two studies using single‐gate designs evaluated the test accuracy of the LAMP (Thaipadungpanit 2011; Kitashoji 2015; Figure 17). LAMP was done on whole blood, plasma, or urine samples. The median timing of sample collection was 6.5 DPO for Kitashoji 2015 (plasma samples only, unreported for urine), but for Thaipadungpanit 2011 the timing was separately reported for cases (median 4 DPO) and non‐cases (median 6 DPO). Kitashoji 2015 reported results for LAMP on plasma (sensitivity 14%, 95% CI 9% to 22%; specificity 83%, 95% CI 76% to 89%) and LAMP on urine samples (sensitivity 14%, 95% CI 7% to 24%; specificity 91%, 95% CI 83% to 95%). Thaipadungpanit 2011 reported results for LAMP targeting rrs (Sonthayanon 2011) (sensitivity 44%, 95% CI 35% to 52%; specificity 83%, 95% CI 76% to 89%) and LAMP targeting lipL41 (Lin 2009) (sensitivity 38%, 95% CI 29% to 46%; specificity 90%, 95% CI 84% to 95%).

Enzyme‐linked immunosorbent assay

One cross‐sectional study conducted in India assessed the test accuracy of ELISA on urine samples (Chaurasia 2018), The timing of sample collection and the threshold were not reported. The study used single‐sample MAT as reference standard to classify 23 participants as cases and six as non‐cases. The ELISA was done using seven different target antigens (LipL32, Fla1, LipL41, HbpA, SphCD210, Sph2, and Sph4). For clarity, the sensitivities and specificities of these tests are shown in Table 14.

Dot‐enzyme‐linked immunosorbent assay

One two‐gate case‐control study evaluated the test accuracy of a monoclonal antibody‐based dot‐ELISA for the detection of leptospiral antigens in urine samples (Saengjaruk 2002). The study was conducted in Thailand with 42 participants, of which 25 were leptospirosis cases confirmed by culture and 17 were people with other illnesses. Timing of sample collection was not reported. The sensitivity of dot‐ELISA was 64% (95% CI 43% to 82%) and specificity was 100% (95% CI 81% to 100%).

Immunochromatography‐based lateral flow assay

One cross‐sectional study assessed the accuracy of an ICG‐based LFA using monoclonal antibodies specific to the Leptospira lipopolysaccharide (Widiyanti 2013). The study tested the urine samples of 44 participants with suspected leptospirosis, and classified 28 as cases based on MAT alone. The mean timing of sample collection was 5.5 DPO. The sensitivity of LFA was 96% (95% CI 82% to 100%) and specificity was 56% (95% CI 30% to 80%).

Dipstick assay

Widiyanti 2013 evaluated a monoclonal antibody‐based dipstick assay on urine, specific to the Leptospira lipopolysaccharide. The same urine samples were tested, with a mean timing of sample collection of 5.5 DPO. The sensitivity of the dipstick assay was 89% (95% CI 72% to 98%) and specificity was 63% (95% CI 35% to 85%).

Comparison of index tests

Several studies performed a comparison of index tests in the same patient population. Since each comparison contained only one study, we could not perform a meta‐analysis.

• PCR versus real‐time PCR: Vanasco 2016 reported a comparison of PCR and real‐time PCR, both performed on whole blood or serum samples. PCR had a sensitivity of 13% (95% CI 6% to 24%) and a specificity of 88% (95% CI 81% to 94%). Real‐time PCR had a sensitivity of 30% (95% CI 19% to 42%) and a specificity of 81% (95% CI 73% to 88%).

• PCR versus nested PCR: Blanco 2014 reported a comparison of PCR and nested PCR. PCR had a sensitivity of 14% (95% CI 4% to 33%) and a specificity of 100% (95% CI 99% to 100%). Nested PCR had a sensitivity of 86% (95% CI 67% to 96%) and a specificity of 100% (95% CI 99% to 100%).

• Real‐time PCR versus nested PCR: Merien 2005 compared real‐time PCR with nested PCR. The study reported identical results for the two tests: sensitivity was 71% (95% CI 44% to 90%) and specificity was 62% (95% CI 44% to 78%).

• PCR versus ICG‐based IFA versus dipstick assay: Widiyanti 2013 reported a comparison between PCR, ICG‐based IFA, and dipstick assay, with all three tests performed on urine samples.

  • For PCR on urine, sensitivity was 57% (95% CI 37% to 76%) and specificity was 56% (95% CI 30% to 80%).

  • For ICG‐based IFA, sensitivity was 96% (95% CI 82% to 100%) and specificity was 56% (95% CI 30% to 80%).

  • For dipstick assay, sensitivity was 89% (95% CI 72% to 98%) and specificity was 63% (95% CI 35% to 85%).

• Real‐time PCR versus LAMP: Thaipadungpanit 2011 reported a comparison of rrs (Slack 2007) andlipL32 (Stoddard 2009) real‐time PCR and rrs (Sonthayanon 2011) and lipL41 (Lin 2009) LAMP.

  • For real‐time PCR targetingrrs, sensitivity was 56% (95% CI 47% to 64%) and specificity was 89% (95% CI 83% to 94%).

  • For real‐time PCR targeting lipL32, sensitivity was 43% (95% CI 34% to 52%) and specificity was 93% (95% CI 88% to 97%).

  • For LAMP targeting rrs, sensitivity was 44% (95% CI 35% to 52%) and specificity was 83% (95% CI 76% to 89%).

  • For LAMP targeting lipL41, sensitivity was 38% (95% CI 29% to 46%) and specificity was 90% (95% CI 84% to 95%).

Summary of findings 1

Summary of findings 2

Summary of findings 3

Summary of findings 4

Interpretation of the conventional polymerase chain reaction meta‐analysis

From a meta‐analysis of 15 studies evaluating PCR on blood products collected during various stages of disease, the pooled sensitivity was 70% (95% CI 37% to 90%) and the pooled specificity was 95% (95% CI 75% to 99%). This means that in a hypothetical cohort of 1000 people, with a prevalence of 32.5% (325 diseased), 98 (95% CI 32 to 205) cases would be missed by the PCR and 35 (95% CI 6 to 168) non‐diseased people would be incorrectly diagnosed with leptospirosis. If the prior probability of an individual to have leptospirosis is 32.5%, the PPP was 87% (95% CI 53% to 97%) and the NPP was 87% (95% CI 71% to 95%). If the PCR would be used in a setting such as in the Netherlands where the prevalence is lower (9.7% in 2016; Collaborating Centre for Reference and Research on Leptospirosis in Amsterdam, unpublished data), 97/1000 people would have leptospirosis. In such a cohort, 29 (95% CI 9 to 61) diseased people would be missed and 47 (95% CI 8 to 225) non‐diseased people would be incorrectly diagnosed with leptospirosis. If the prior probability of leptospirosis in an individual is 9.7%, the PPP is 59% (95% CI 20% to 89%) and the NPP is 97% (95% CI 92% to 99%). See Table 1.

When the PCR meta‐analysis was repeated without the studies at high risk of bias for the 'reference standard' domain (i.e. only including studies with composite reference standards or two‐gate studies with MAT as reference standard, based on nine studies), the pooled sensitivity was 87% (95% CI 44% to 98%), and the pooled specificity was 97% (95% CI 60% to 100%). See Table 2. This means that in a cohort of 1000 people with 325 leptospirosis patients, 42 (95% CI 5 to 183) cases would be missed and 17 (95% CI 1 to 272) non‐diseased people would be incorrectly diagnosed with leptospirosis. In an individual with 32.5% prior probability of leptospirosis, the PPP is 94% (95% CI 41% to 100%) and the NPP is 94% (95% CI 70% to 99%). If the prevalence is 9.7%, 13 (95% CI 2 to 55) diseased people would be missed and 23 (95% CI 1 to 363) non‐diseased people would be incorrectly diagnosed with leptospirosis. If an individual has 9.7% prior probability of leptospirosis, the PPP is 78% (95% CI 13% to 99%), and the NPP is 99% (95% CI 91% to 100%).

A repeated meta‐analysis of PCR without studies at high risk of bias for the 'patients' domain did not lead to important changes in our results.

Interpretation of the real‐time polymerase chain reaction meta‐analysis

For the real‐time PCR, we estimated an SROC curve instead of a summary sensitivity and specificity, since we expected the positivity threshold (which was often not reported) to vary between studies. The median specificity of real‐time PCR on blood products was 92%. For illustrative purposes, if we were to select a point on the curve with 95% specificity, the pooled sensitivity would be 29% (95% CI 15% to 49%) at an unknown threshold (Table 3). Translating these numbers to a cohort of 1000 people of whom 325 are diseased, would mean that 230 (95% CI 167 to 276) diseased people would be missed and 34 non‐diseased people would be incorrectly diagnosed with leptospirosis. Again, in a setting with a prevalence of 9.7% this would imply that 69 (95% CI 50 to 82) diseased people would be missed and 45 non‐diseased people would be incorrectly diagnosed with leptospirosis. We did not provide PPP and NPP for real‐time PCR, as threshold values needed to produce these estimates were unknown.

When we excluded real‐time PCR studies at high risk of bias for the 'patients' domain or the 'reference standard' domain, there were no important changes in the pooled estimate.

The position of conventional polymerase chain reaction and real‐time polymerase chain reaction in the clinical pathway

Based on the properties of PCR and real‐time PCR, we examined the possible role of these tests in the diagnostic pathway for leptospirosis. Leptospirosis is a potentially life‐threatening disease, meaning that efforts should be undertaken to minimise false‐negative results. In the clinic, while most patients with suspicion of a severe infection are likely to receive broad‐spectrum antibiotics, a false‐negative result impedes the optimisation of the antibiotic therapy and assessment of the prognosis. In an outbreak setting, a missed case of leptospirosis will delay outbreak response and facilitate further dissemination of disease.

For patients presenting in the early disease stage, PCR and real‐time PCR on blood products are preferable as first‐line tests in the clinical pathway based on their ability for early detection. However, whether additional testing is needed to verify a positive or negative test result depends on test accuracy and the prevalence of leptospirosis. While PCR‐based methods have been described as sensitive tests in the literature (Budihal 2014; Hall 2014; Picardeau 2014), our results show that the sensitivity of PCR and real‐time PCR vary greatly between studies, with the CI for PCR sensitivity ranging from 37% to 90%. This can partly be explained by differences in methodological quality, but it is also likely that there is true heterogeneity, such as differences in timing of sample collection. Furthermore, the reliability of a positive or negative test result depends on the prevalence. In our review, prevalence of leptospirosis varied greatly between studies (range 3% to 85%). Consequently, whether PCR‐based methods can be used alone, or together with other follow‐up tests depends on regional considerations such as prevalence, factors that are likely to influence accuracy (e.g. timing of sampling), and downstream consequences of a positive or negative result. For example, in settings with a high prevalence, PCR and real‐time PCR may not have a high enough negative post‐test probability to confidently rule out leptospirosis. In this case, additional testing to verify negative results should be considered.

Comparison between index tests

This review did not find enough evidence to formally compare the diagnostic accuracy of included tests. Direct comparison studies (where two or more index tests are evaluated in the same patient population) are needed to draw valid conclusions about the differences in diagnostic test accuracy between tests, but such studies were lacking. Although the results of the meta‐analyses seem to imply that real‐time PCR has a lower sensitivity than PCR, this is not a valid comparison. The meta‐analysis results are composed of mostly single‐test studies and any differences between real‐time PCR and PCR could arise from other reasons than the differences in the tests themselves, such as differences in study design or spectrum of disease. Therefore, any comparison between the meta‐analytic results of PCR and real‐time PCR must be interpreted with caution.

Heterogeneity of included studies

Substantial heterogeneity, as demonstrated by the wide CIs, complicated the interpretation of our findings.

An important covariate, timing of sample collection, could not be explored in heterogeneity analysis. Three of 15 studies assessing PCR on blood products reported the timing of sample collection (range 1 DPO to 7 DPO of symptoms), and in 9/16 studies assessing real‐time PCR on blood products (range 1 to 19 DPO). However, the subgroups were either too small or the reporting of the timing variable was too heterogeneous for analysis. Two studies comparing test accuracy in patients who presented early with test accuracy in patients who presented later appear to support the hypothesis that the sensitivity of PCR and real‐time PCR is greater in the first few days of illness (Céspedes 2007; Ahmed 2009). This is consistent with the current pathophysiological understanding of leptospirosis, that leptospiraemia declines rapidly and becomes undetectable after 10 DPO (WHO 2003). However, this could also be caused by patients with a higher bacterial load presenting earlier to the clinic due to a more severe clinical presentation than those with a lower bacterial load. Moreover, one study reported that the sensitivity of real‐time PCR was not associated with timing of sample collection in patients presenting with fewer than 10 DPO (P = 0.33) (Agampodi 2012). Ultimately, more studies are needed to confirm the association between timing and test accuracy.

In the case of PCR, statistical heterogeneity may be partly explained by the prevalence of leptospirosis in the study population. Specificity was inversely correlated with prevalence (P = 0.0004). A number of explanations for this association are possible. The prevalence of alternative diagnoses may be higher in places where leptospirosis prevalence is high, causing false‐positive results on the PCR. The PCR may be detecting lower levels of infections that occur more frequently in high‐prevalence settings, that are missed by MAT, and, therefore, recorded as false‐positive results. The laboratories in high‐prevalence settings may be less well‐equipped and more often contaminated, and, therefore, allow more false‐positive results. We did not examine the inverse correlation between prevalence and sensitivity or specificity for real‐time PCR studies, since the HSROC model by default examines the association between prevalence and accuracy (alpha parameter) instead of sensitivity and specificity.

Other covariates that could possibly influence test accuracy, such as real‐time PCR threshold and specific target genes or primers used in PCR‐based methods, have not been ruled out as possible explanations for the significant heterogeneity. It is theoretically possible that the heterogeneity in sensitivity of PCR‐based methods could be explained by differences in Leptospira species, as a primer may not able to detect a particular species. However, this is not very probable as it is usual practice to account for all existing species when developing PCR or its variants (unless a new species emerges). Since all index tests were inhouse tests, there may be other potential sources of heterogeneity (e.g. use of different laboratory equipment or protocols) that cannot be measured reliably or be reported in sufficient detail. For this reason, readers should be cautious when applying summary estimates of test accuracy in their own clinical settings.

Risk of bias

A major point of attention in our review was the use of MAT as reference standard, which is considered to have an imperfect sensitivity. If the reference standard is not sensitive, the specificity of the index test is likely to be underestimated. It is furthermore not inconceivable that some index tests, for example, PCR‐based methods, may be more sensitive than MAT alone. We aimed to address this problem by including composite reference standards and rating the risk of bias as high when MAT was used as the sole reference standard, and when single samples were used (instead of paired). However, only a minority of studies used another reference standard alongside the MAT, leading to a 'high risk of bias' judgement in the majority of studies for this domain. In the case of the PCR, sensitivity increased when studies at 'high risk of bias' for the reference standard were excluded, but specificity was unchanged (Table 9). Other covariates that may be of importance, such as the cut‐off value for MAT, the use of adequate regional panels for MAT, and the differences between composite reference standards, were not taken into account in our review to avoid excessive complexity.

Another issue is the inclusion of four two‐gate case‐control studies (Samsonova 1997; Ananyina 2000; Saengjaruk 2002; de Abreu Fonseca 2006). Cases and controls in these studies are selected separately and do not reflect the spectrum of disease in the clinical population (Rutjes 2005). Another concern in these studies is the possibility of coinfections of leptospirosis and another infectious disease. In two‐gate designs, since controls are not MAT negatives but people with a condition resembling leptospirosis, coinfections with leptospirosis may be present. Treating these people as controls may underestimate the specificity of the index test. These studies have been excluded in the abovementioned sensitivity analyses as they were considered 'high risk' for patient selection (Table 9; Table 11).

Blinding of the index test result to the reference standard result interpreters (or vice versa) was largely unreported. Interpreters were blinded because either blinding methods were used, or by the virtue of their study design (e.g. blinding of the index test interpreter was not needed when the index test was done first). In the index test, eight studies reported to have used a form of blinding in their methods. However, we noted that only one study reported explicit methods for blinding (Samsonova 1997). It was unclear if the remaining studies used proper blinding methods. Considering other possible biases in our review (Figure 4; Figure 6; Figure 8), readers are advised to weigh the results against the quality of evidence.

Other index tests

For other tests included in our review (nested PCR, PCR performed twice, LAMP, ELISA, dot‐ELISA, ICG‐based LFA, and dipstick assay), we could not conduct meta‐analyses or investigations of heterogeneity due to the small number of studies.

CHBG Editorial Team:
Peer reviewers of the protocol and the review: David Brett‐Major, USA
Contact editors of the protocol and the review: Agostino Colli, Italy; Giovanni Casazza, Italy
Sign‐off editor of the protocol and the review: Agostino Colli, Italy

Peer reviewers from The UK Diagnostic Accuracy Reviews Editorial Team: anonymous.
Contact editor from The UK Diagnostic Accuracy Reviews Editorial Team: Miriam Brazzelli, UK.

Appendix 1. Search strategies

Appendix 2. QUADAS‐2 review‐specific guidance

Data tables by test

Open in a separate window