A total of 60 studies were identified for inclusion (Fig 1 and Table 1) [19–69]. Thirteen studies were excluded due to the use of a qualitative test, 8 studies were drug resistance testing studies, 2 studies used panels as primary sample types, 2 studies were review manuscripts lacking primary data, 2 studies used spiked blood samples and 1 study used an incorrect test. comparator. There was low to moderate heterogeneity in analytical and clinical performance comparisons within technologies as well as medians and viral load distributions (Table 2).
Quality of studies
There was some risk of bias in patient selection, however, low risk of bias with the reference standard and index test (S2 Fig). Participants in most studies were not consecutively recruited or did not report the patient recruitment process and only 5% of studies reported the patient recruitment process. There was wide applicability in patient selection, index testing, and reference standard; however, there were some concerns as most of the studies (58%) were conducted in Africa; most studies (>90%) used venous blood prepared in the laboratory with a pipette; and most studies (>90%) used only one dried blood spot filter paper (Whatman 903).
Systematic review analysis
Mean bias was the most commonly reported analytical measure across all studies included in the systematic review (82%); therefore, the forest plots in each study were developed by technology (S3 Fig). Half of the studies included patients on antiretroviral therapy [21–23,29,33,35–37,39,40,42,43,45–47,51–54,56–59,62,64,65,68,69], while the remaining studies included patients who were not on antiretroviral therapy or did not report such information. Study characteristics such as sample size, viral load medians, and patient viral load distributions are summarized in Table 2.
A total of 40 studies provided 45 datasets on the 6 technologies, resulting in a total of 10,871 matched DBS and plasma viral load results [22,24,26,28,30,31,33,34,36–45,48–53,56,58,59,62–66,68,69]. Studies not included in the systematic review were due to the inability of lead authors to share data. Of these pairs, 58% were analyzed with Roche COBAS TaqMan technology [22,26,40,43,48,49,51,56,58,65,66,69]25% with Abbott RealTime HIV-1 Technology [26,28,31,37–39,42,45,58,59,63,69]10% with bioMérieux NucliSENS EasyQ technology [24,30,31,33,34,36,40,42,53,62]5% with Biocentric Generic HIV Viral Load technology [41,52,64]1% with Hologic Aptima and 1% with Siemens VERSANT HIV-1 RNA technology . About 70% of matched data points were from studies in Africa [22,24,26,28,33,34,37,39,40,42,43,48,53,56,58,59,64,65,74]of which 36% were from the Southern African Development Community region [22,26,28,42,43,56,59,64] and 24% from the East African Community region [24,33,37,48,65,69].
The distribution of viral load for the 10,871 plasma samples tested was relatively evenly distributed across all viral load ranges (Fig 2A). While approximately 41% of all plasma samples were undetectable (below the technology’s detection limit), 30% of all plasma samples studied were between detectable (at or above the detection limit technology) and 10,000 copies/ml. Moreover, by including only plasma samples from patients known to be on ART, we observed that just over 40% of patients had undetectable viral load levels (Fig 2B). Approximately 31% of plasma samples from patients on ART were between detectable and 10,000 copies/ml.
Median DBS viral loads were higher than median plasma viral loads for all but 2 technologies. Overall, the median difference was 1.03 log copies/ml (Table 3). The Abbott RealTime two-dot HIV-1, Abbott RealTime HIV-1 one-spot, Biocentric Generic HIV Charge Virale, bioMérieux NucliSENS EasyQ HIV-1, Hologic Aptima, Roche COBAS TaqMan FVE, Roche COBAS TaqMan SPEX and Siemens VERSANT HIV-1 RNA technologies showed a difference between median dried blood loads spot and plasma samples of 0.09, 0.04, 0.17, -0.30, 0.12, 0.33, 1.99 and -0.13 log copies/ml, respectively. The average bias for each technology was calculated by pooling all primary data for each technology as if it were a single study (Table 3). Mean biases between dried blood spot and plasma viral load values varied significantly by technology. The overall mean bias was 0.30 log copies/ml. The Abbott RealTime Two-point HIV-1 (−0.12 log copies/mL), Abbott RealTime The biases of the HIV-1 one-spot (0.02 log copies/ml) and Roche COBAS TaqMan FVE (0.06 log copies/ml) assays were closest to zero, while the bioMérieux NucliSENS EasyQ HIV-1 (−0.41 log copies/ml) and Roche COBAS TaqMan SPEX Test biases (1.03 log copies/ml) were farthest from zero. The Abbott RealTimThe e HIV-1 two-spot, bioMerieux NucliSENS EasyQ HIV-1 and Siemens VERSANT HIV-1 RNA technologies had negative mean biases indicating under-quantification compared to the plasma viral load result, which is expected due to the lower input sample volume. The mean positive biases of Biocentric Generic HIV Charge Virale and Roche COBAS TaqMan SPEX reflect overquantification relative to the plasma viral load result, likely due to processing and extraction chemistries resulting in amplification of total intracellular and extracellular nucleic acids .
WHO and many national clinical guidelines in resource-limited settings recommend using viral load testing as a binary result, above or below a specific threshold to identify virologic failure. We therefore compared several thresholds of virologic failure for DBS specimens (1000, 3000, 5000, 7500, and 10,000 copies/mL) to the currently suggested threshold of 1000 copies/mL for specimens of plasma in order to correctly classify patients (Table 3 and Fig. 3). Using a DBS sample threshold of 1,000 copies/ml, all 6 technologies had greater than 80% sensitivity for detecting a viral load greater than 1,000 copies/ml. At the same threshold, the specificity of detecting a viral load below 1,000 copies/ml was greater than 80% for all technologies except for Biocentric Generic HIV Charge Virale (55.16%), Hologic Aptima (73.44% ) and Roche COBAS TaqMan SPEX (43.86%). Using a higher virologic failure threshold, such as 5000 copies/mL, for dried blood spot samples reduced sensitivity and increased specificity of all technologies. Finally, HSROC curves were created for technologies where more than 4 studies were included in the meta-analysis (S4 Fig).
Fig 3. Forest plots of sensitivity and specificity of all studies included in the meta-analysis for each viral load technology using a treatment failure threshold of 1000 copies/mL.
Abbott Real Time HIV-1 two-spot (a), Abbott RealTime HIV-1 one-spot (b), Biocentric Generic HIV Charge Virale (c), bioMerieux NucliSENS EasyQ HIV-1 (d), Hologic Aptima (e), Roche COBAS TaqMan FVE (f), Roche COBAS TaqMan SPEX (g ), Siemens VERSANT HIV-1 RNA (h). Bars and red lines indicate overall metrics for each viral load technology.
Additionally, to better understand the performance of DBS specimens at lower treatment failure thresholds (below 1000 copies/mL), we compared the 6 predefined virological treatment failure thresholds: detectable, 200, 400, 500, 600, and 800 copies/mL — between dried blood spot samples and plasma for each technology and protocol (Table 4). Biocentric Generic HIV Charge Virale and Roche COBAS TaqMan SPEX technologies showed low specificity (mTwo-point HIV-1 and Roche COBAS TaqMan FVE protocols had high sensitivities and specificities at all but detectable lower thresholds. The new Abbott RealTimThe HIV-1 single-point protocol, however, showed high specificities at all lower cutoffs, but sensitivity performance was less than 85% at 200 copies/ml and detectable cutoffs. The Hologic Aptima had high sensitivities with all but detectable thresholds; however, the specificity was less than 85% at the 800 copies/ml and 200 copies/ml thresholds. Finally, the bioMérieux NucliSENS EasyQ HIV-1 presented sensitivities and specificities greater than 85% at all thresholds.
Quantitative polymerase chain reaction inherently introduces a level of variability into test results, typically +/-0.3 log copies/mL [68,69]. We therefore sought to understand whether the performance observed with each technology was within the inherent limits of assay variability. For the Abbott RealTime two-dot HIV-1, Abbott RealTime HIV-1 one-spot, Biocentric Generic HIV Charge Virale, bioMérieux NucliSENS EasyQ HIV-1, Hologic Aptima, Roche COBAS TaqMan FVE, Roche COBAS TaqMan SPEX and Siemens VERSANT HIV-1 RNA technologies, 59.28%, 68.71 %, 38.04%, 52.54%, 50.40%, 62.03%, 33.45%, 47.22% Dried Blood Spot Sample Test Results Were Within Standard Deviation of +/- 0.3 log copies/ml of the paired plasma test result, respectively (Fig 4).
Fig 4. A significant proportion of dried blood spot results fall outside the plasma result +/-0.3 log copies/ml for each technology.
Abbott Real Time HIV-1 two-spot (a), Abbott RealTime HIV-1 one-spot (b), Biocentric Generic HIV Charge Virale (c), bioMerieux NucliSENS EasyQ HIV-1 (d), Hologic Aptima (e), Roche COBAS TaqMan FVE (f), Roche COBAS TaqMan SPEX (g ), Siemens VERSANT HIV-1 RNA (h). The blue bars represent +/-0.3 log copies/ml of the plasma result, while the orange triangles represent the matched viral load result of the dried blood spot.