Discordant rapid HIV tests: lessons from a low-resource community

Abstract

Objectives

HIV rapid antibody tests are widely used in Africa, but dual testing sometimes produces discordant results. It is not clear if discordant rapid HIV tests should always heighten suspicion by frontline health workers that early HIV infection is present. Some studies have reported that discordant rapid tests have value for identifying early HIV infection in high HIV prevalence populations. It is not known if rapid test performance influenced this conclusion, or if this observation will hold true for low HIV prevalence populations. We therefore explored the occurrence of discordant rapid HIV tests in a low-resource community.

Methods

A cross-sectional sample of HIV status-unaware adults with recent exposure to unsafe sex was assessed using a validated risk-based tool (University of North Carolina (UNC)-Malawi Risk Screening Score) for acute HIV infection. Participants received rapid testing with Determine^™ HIV 1/2 and Uni-Gold^™ HIV assays, plus plasma HIV-1 antigen testing with the COBAS^® Ampliprep/COBAS^® Taqman^® HIV-1 assay, followed by western blot in those with detected HIV-1 antigen.

Results

Of 408 participants, 1.0% were confirmed to have established HIV infection. The discordance between rapid tests at initial screening was 2.45 and 2.94% when the two assays were used sequentially and simultaneously, respectively. Discordant rapid tests were strongly associated with risk scores > 2 [odds ratio (OR) 10.88; 95% confidence interval (CI) 2.35–50.43], and with detected HIV-1 RNA (OR 26.06; 95% CI 3.91–173.60).

Conclusions

When the sample occurrence of discordance between the first and second tests is below 5%, discordant rapid tests in an adult with sexual risk behaviour should trigger strong suspicion of early HIV infection in low HIV prevalence populations.

Introduction

Over the last decade, HIV testing has been scaled up across sub-Saharan Africa such that testing is sometimes performed in nonlaboratory settings by lay persons using point-of-care tests. Low-budget screening programmes predominantly use rapid diagnostic tests (RDTs) that are based on immunochromatographic detection of HIV antibodies. Detection of acute HIV infection (AHI) is a challenge in sub-Saharan Africa because recommended antigen-based HIV tests are not readily available [1].

Serial testing for HIV infection utilizes two different antibody tests sequentially (if the initial test is reactive), while parallel testing employs two different antibody tests simultaneously. Both testing algorithms sometimes yield discordant results between the first and second RDTs, a “reactive versus nonreactive” outcome that requires further testing [2].

Some reports have linked discordant rapid antibody results with AHI. In Malawi, Fiscus and colleagues reported that 33% of AHI cases had discordant rapid tests compared with 2% of HIV-negative participants [3]. Several studies have reported on the high rate of false positive results produced by rapid HIV testing [4–6]. It is not clear if Fiscus et al. took false positive results into consideration in arriving at the conclusion of an association between rapid test discordance and AHI. In addition, their study was clinic-based and involved patient populations with a high HIV prevalence. It is not known whether the said observation will hold true for populations with a low HIV prevalence.

We sought to determine the RDT discordance rate at first screening, and the association between discordant rapid tests and diagnostic markers of AHI during HIV screening in a relevant community with a low HIV prevalence.

Methods

Study location

A cross-sectional study was conducted from October to December 2015 in Sabo, a multi-ethnic urban-slum community in Ibadan, Oyo State, Nigeria. Oyo State recorded an HIV prevalence of 5.6% in Nigeria’s 2013 HIV/AIDS and Reproductive Health Survey [7].

Recruitment

We conducted 4 weeks of a community awareness drive, and then set up a free HIV counselling and testing station within the host community. Consenting adults (age ≥ 18 years) were recruited if their HIV status was unknown and they self-reported unsafe sex within the preceding 2 months. Unsafe sex was defined as unprotected coitus with casual or multiple partners with unknown HIV status or confirmed HIV infection. Persons with prior HIV post-exposure prophylaxis or antiretroviral treatment were excluded.

Study procedures

At enrolment, each participant received a unique study identifier, and then completed an interviewer-administered study questionnaire, which included the UNC-Malawi Risk Screening Score (UMRSS) assessment (a validated risk score algorithm in which six predictor scores are summed to calculate each participant’s AHI risk score within a possible range of 0–11 points). Powers et al.[8] reported that a risk score of ≥ 2 was 95.2% sensitive and 60.5% specific for detecting AHI. HIV rapid antibody testing was performed on finger-prick blood samples on site. All participants received HIV pre-test and post-test counselling. Each sample was tested with the Determine^™ HIV 1/2 assay (Abbott Laboratories, Japan Co., Ltd, Tokyo, Japan) and the Uni-Gold^™ HIV assay (Trinity Biotech, Bray, Ireland) according to the parallel testing algorithm. Test outcomes for the Determine assay-based serial testing algorithm were deduced during data (post hoc) analysis. We considered a test result as seropositive if both RDTs were reactive (dual-positive), as seronegative if both RDTs were nonreactive (dual negative), and as discordant if one RDT was reactive and the other nonreactive. The outcome variables were RDT discordance between Determine and Uni-Gold assays at first screening; UNC-Malawi Risk Scores; and plasma HIV-1 RNA levels.

Each participant also had 7 mL of blood collected in an ethylenediaminetetraacetic acid (EDTA) tube. Plasma was separated within 4 h of blood draw after centrifugation at 1200 g for 20 min at the University College Hospital, Ibadan (UCH), and transported to the Virology Laboratory, Nigerian Institute of Medical Research, Lagos (NIMR) for HIV-1 RNA testing using the COBAS^® Ampliprep/COBAS^® Taqman^® HIV-1 Test, version 2.0 (Roche Moelcular Systems, Branchburg, NJ, USA). This assay reports the HIV target as “detected” or “not detected”, and has an HIV-1 RNA quantification range of 20–10 000 000 HIV-1 RNA copies/mL when the HIV target is detected. A “not detected” result indicates that the assay was unable to detect HIV-1 RNA in the plasma specimen tested. Plasma samples with detected HIV-1 RNA were further subjected to western blot (WB) to decide between two possible diagnostic outcomes: acute or established HIV infection. We defined AHI as a composite outcome of (1) dual-negative or discordant RDTs at first screening, (2) detected plasma HIV-1 RNA > 20 copies/mL, and (3) negative WB. Established HIV infection was defined as a triple combination of (1) dual-positive or discordant RDTs at first screening, (2) detected plasma HIV-1 RNA > 20 copies/mL, and (3) positive WB. Participants with HIV infection were referred to the HIV clinic, UCH for clinical care.

Ethical approval

Approval was obtained from the Ethical Review Committee, Oyo State Ministry of Health; University of Ibadan/UCH Ethical Committee; and Northwestern University Institutional Review Board. Each participant provided written informed consent.

Statistical methods

Data were collected using a structured form, and entered into REDCap (Research Electronic Data Capture). Data analysis was performed with STATA 11.0 (StataCorp, College Station, TX, USA). The mean UMRSSs for participants with and without RDT discordance were compared using a t-test. Odds ratios (ORs) were used to test the associations between discordant RDT and UMRSS summated scores, and between discordant RDT and detected plasma HIV-1 RNA. The level of statistical significance was set at P ≤ 0.05.

Results

A total of 408 adults with a mean age of 37.7 ± SD14.2 years participated in the study, with 52.9% being male. The majority of participants were married (59.8%) and employed (82.4%), and 48.3% had 10 years or more of primary and secondary education.

At first screening, 408 paired tests were performed using Determine (408 tests) and Uni-Gold (408 tests) simultaneously. This yielded 394 nonreactive (dual-negative) pairs, 12 discordant pairs, and two reactive (dual-positive) pairs of results (see Table 1). The discordance rate between Determine and Uni-Gold tests was 2.94% when the two RDTs were used simultaneously. Post hoc analysis for a serial testing algorithm yielded 396 nonreactive and 12 reactive Determine (unpaired) results. Sequential application of RDT 2 (after each reactive Determine test) yielded 10 nonreactive and two reactive Uni-Gold results. The overall serial testing outcome was 10 discordant pairs and two reactive (dual-positive) pairs of results. The discordance rate between Determine and Uni-Gold was 2.45% when the two RDTs were used sequentially. There was no statistically significant difference (χ² = 0.18; P = 0.66) between the test discordance rates obtained when the RDTs were used simultaneously and sequentially.

Table 1.

Results for rapid tests, HIV-1 RNA, and western blot

Serial Number	Determine/Uni-Gold RDT result	Plasma HIV-1 RNA	Western blot	n (%) (N = 408)
1.	Dual-negative	Not detected	Not applicable	391 (95.8)
2.	Discordant (Determine reactive/Uni-Gold nonreactive)	Not detected	Not applicable	8 (2.0)
3.	Dual-negative	Detected (mean VL < 20 copies/mL)	Negative	3 (0.7)
4.	Discordant (Determine nonreactive/Uni-Gold reactive)	Not detected	Not applicable	2 (0.5)
5.	Discordant (Determine reactive/Uni-Gold nonreactive)	Detected (mean VL 39 730 copies/mL)	Positive	2 (0.5)
6.	Dual-positive (Determine reactive/Uni-Gold reactive)	Detected (mean VL 55 440 copies/mL)	Positive	2 (0.5)

RDT, rapid diagnostic test; VL, viral load; S/N serial number.

For 408 participants, the UMRSSs obtained ranged between 1 and 8 points, with a mean UMRSS of 2.56 (SD ± 1.03). For 12 participants with discordant RDT results, the mean UMRSS was 5.58 (SD ± 1.98). There was a statistically significant difference (t = 9.67; P = 0.0001) between these two means. The majority of participants (66.2%) had UMRSS ≤ 2. Discordant rapid tests were strongly associated with UMRSS > 2 [OR 10.88; 95% confidence interval (CI) 2.35–50.43]. All 408 plasma samples underwent HIV-1 RNA analysis. HIV-1 RNA was not detected in 394 (96.6%) samples with nonreactive (dual-negative) RDTs at first screening (see Table 2). Three (0.7%) samples with nonreactive (dual-negative) RDTs at initial testing had HIV-1 RNA detected (< 20 copies/mL), and were WB negative. Of 12 samples with discordant RDTs at initial testing, two (16.7%) samples had HIV-1 RNA detected (5681 and 73 779 copies/mL, respectively). Both samples were confirmed HIV positive by WB. Discordant rapid tests were strongly associated with detected plasma HIV-1 RNA (OR 26.06; 95% CI 3.91–173.60). Overall, four (out of 408) participants were confirmed to be HIV-infected, a HIV seroprevalence rate of 1.0%. No AHI case was identified.

Table 2.

Odds of discordant rapid tests according to UNC-Malawi Risk Screening Score (UMRSS) and plasma HIV-1 RNA

AHI diagnostic marker		Discordant RDTs (n)	Dual-negative RDTs (n)	Total n (N = 406)	Odds ratio (95% CI)	P-value
UMRSS	UMRSS ≤ 2	2	270	272	10.88	0.002
	UMRSS > 2	10	124	134	(2.35–50.43)
HIV-1 RNA	HIV-1 RNA not detected	10	391	401	26.06	0.008
	HIV-1 RNA detected	2	3	5	(3.91–173.60)

CI, confidence interval; RDT, rapid diagnostic test.

Discussion

A discordant rapid HIV test result arises when one of the tests is falsely negative or falsely positive. The occurrence of discordant rapid tests (Determine/Uni-Gold) in this study was approximately 3% for either parallel or serial algorithms. This operating characteristic is within the World Health Organization (WHO) recommendation that the discordance of results between first and second tests should not exceed 5% [2]. A false negative test may occur as a result of low stability of antibody expression soon after establishment of HIV infection [9]. In our study, more than half of all the discordant rapid tests involved reactive (false positive) Determine and nonreactive (true negative) Uni-Gold results. Several studies have reported a high rate of false positive results produced by rapid HIV testing [4–6]. An interim analysis of data from Ethiopia reported a 7% false positive rate using the national algorithm which relied on two concordant (out of three) RDT results for HIV diagnosis [10]. Endemic infections [11], heightened CD5 expression and early B-lymphocyte response polyclonal cross-reactivity [5] may contribute to HIV false positivity in certain settings. Researchers conducting clinical trials using RDTs in Uganda have noted that the high false positive rates observed may be attributable to misinterpretation of weakly positive bands [12]. Other reports have corroborated this observation, particularly for the Determine HIV 1/2 assay [13]. The intensity of positive bands on Determine HIV 1/2 was not coded in our study, and the possibility of observer error cannot be ruled out.

In the clinic-based study by Fiscus et al.,[3] of 1450 adults offered voluntary HIV counselling and testing in Malawi using two RDTs, approximately 40% turned out to have established HIV infection, a notably high seroprevalence rate. Discordant rapid tests occurred in seven (33.3%) of 21 individuals with AHI. Of those with reactive Determine and nonreactive Uni-Gold test results, 35% had AHI. Based on similar findings, Powers et al.[8] reported that discordant rapid test results were strongly associated with AHI (unadjusted prevalence OR 29.5; 95% CI 8.56–92.48). In developing the UMRSS, Powers and colleagues reported that a risk score of ≥ 2 was 95.2% sensitive and 60.5% specific for detecting AHI [8]. Accordingly, our study indicated that discordant rapid tests were strongly associated with UMRSS > 2, implying a suspicion of early HIV infection in adults with these attributes. No participant fulfilled all our study criteria for AHI, and this limitation constrains further inferences about the association of discordant rapid tests with detection of AHI in this sample. Our inability to identify AHI was not unexpected, as our study population generally had a low HIV prevalence. The HIV seroprevalence in our study sample was 1%, much lower than the reported prevalence of 3.2% in Nigerian urban populations [7]. Typically, AHI is difficult to identify, even among serially tested cohorts of high-risk persons. A study among high-risk Nigerians identified 16 cases of AHI in 28 655 persons screened using the same combination of RDTs that our study utilized [14]. In that cohort, four (0.63%) individuals were identified with AHI of 634 persons who were HIV antibody discordant by two RDTs at first testing. An important lesson from our study stems from the definition of AHI. We defined AHI as a composite outcome determined by dual-negative or discordant RDTs at first screening, plus detected plasma HIV-1 RNA > 20 copies/mL, and negative WB. Three plasma samples closely approached these criteria, but had detected plasma HIV-1 RNA < 20 copies/mL, with WB test negative. A result of “< 20 copies/mL” indicates that HIV-1 RNA was detected, although at a level less than the lower quantification limit of the assay used. Due to a lack of approval for follow-up HIV nucleic acid tests after the initial encounter, the clinical implications of these subthreshold nucleic acid test findings were not determined, but the findings could indicate a false positive result (as a consequence of assay variability), or very early HIV-1 infection.

We conclude that, when occurrence of discordance between Determine and Uni-Gold rapid assays is < 5% in low HIV prevalence samples, discordant tests in an adult with sexual risk behaviour should trigger strong suspicion of early HIV infection.