Use of External Quality Control Material for HIV-1 RNA Testing To Assess the Comparability of Data Generated in Separate Laboratories and the Stability of HIV-1 RNA in Samples after Prolonged Storage

ABSTRACT

The National Institute of Allergy and Infectious Diseases (NIAID) AIDS Clinical Trials Group (ACTG) stores specimens from its clinical trials in a biorepository and permits the use of these specimens for nonprotocol exploratory studies, once the studies for the original protocol are concluded. We sought to assess the comparability of the data generated from real-time HIV-1 RNA testing during two clinical trials with the data generated from the retesting of different aliquots of the same samples after years of storage at −80°C. Overall, there was 92% agreement in the data generated for 1,570 paired samples (kappa statistic = 0.757; 95% confidence interval [CI], 0.716 to 0.797), where samples were tested in one laboratory using the microwell plate (MWP) version of the Roche HIV-1 Monitor test within 1 to 37 days of collection and retested in another laboratory using the Cobas version of the assay after a median of 6.7 years of storage (range, 5.7 to 8.6 years). Historical external quality control data submitted to the NIAID Virology Quality Assurance program (VQA) by client laboratories using the same two versions of the Monitor assay were used to differentiate between systematic differences in the assays to evaluate the stability of HIV-1 RNA in the stored samples. No significant loss of RNA was noted in samples containing either a low concentration (<50 copies/ml) or a high concentration (≥50 copies/ml) of HIV-1 RNA (P = 0.10 and P = 0.90, respectively) regardless of the time in storage. These data confirm the quality of the plasma samples in the ACTG biorepository following long-term storage.

INTRODUCTION

Established in 1987, the AIDS Clinical Trials Group (ACTG) is one of the largest networks of HIV-1 investigators and clinical trials units in the world that includes sites in resource-limited countries. Clinical trials conducted by the ACTG have made major contributions to antiretroviral therapy, to preventing and treating coinfections, to reducing long-term drug toxicities, to managing drug resistance, and to helping establish treatment guidelines. More than 2.5 million specimens comprised of 19 different specimen types collected for over 200 clinical trials are currently housed at a central biorepository. Once the studies for which the samples were originally collected are concluded, they become available for investigators to use for new HIV-related research (http://www.specimenrepository.org/home.html). These specimens were collected from well-characterized study participants and cohorts and were processed by trained laboratory staff using standardized protocols for processing, handling, and shipping while maintaining the cold chain throughout the process. All of these steps are critical to ensure the integrity of the stored samples.

External quality assurance programs are utilized by the ACTG to monitor processing laboratory performance for generating viable cryopreserved peripheral blood mononuclear cells and to demonstrate proficiency in the performance of virologic assays, such as HIV-1 RNA testing. The National Institute of Allergy and Infectious Diseases (NIAID) Virology Quality Assurance program (VQA) provides quality control materials to client laboratories for the purposes of proficiency testing and to assess HIV-1 RNA assay performance within a laboratory. The data generated by all laboratories are stored and can be used to evaluate performance across laboratories and assays.

The objective of the current study was to utilize external quality control data generated during the original real-time testing of HIV-1 RNA in plasma samples to assess the comparability of results obtained from the retesting of stored plasma samples from the same time points by using different versions of the Roche Monitor HIV-1 RNA assay and after long-term storage at −80°C. We sought to identify measureable sources of bias and variation across the two versions of the Roche Monitor kit while still allowing for the evaluation of HIV-1 RNA stability in the stored samples.

MATERIALS AND METHODS

Testing laboratories and VQA external quality control materials.

The ACTG uses a single centralized testing laboratory (in the United States) or multiple VQA-certified laboratories (outside the United States) for all on-study HIV-1 RNA testing in its clinical trials. The real-time data included in this analysis were generated by the ACTG central laboratory at Johns Hopkins University (JHU) 1 to 37 days after collection of the sample using the microwell plate (MWP) version of the Roche Monitor assay (Roche Molecular, Pleasantville, CA); the repeat testing was performed by the VQA laboratory using specimens from the same time points included in the real-time testing after a median storage time of 6.7 years (range, 5.7 to 8.6 years). Retesting was performed using the Cobas version of the Roche Monitor assay (1) because the MWP version had been discontinued. The MWP version of the Roche Monitor assay utilized manual extraction with amplification in a thermal cycler and manual detection using a 96-well microtiter plate coated with probes specific to the amplified HIV-1 nucleic acid and internal quantitation standards and colorimetric development of the bound amplified product (2). The Cobas version of the assay included the same manual extraction but provided automated amplification and detection of the amplified product using a bead capture system instead of a microtiter plate (3). The reportable range for the MWP and Cobas assays was 50 to 100,000 copies/ml. Estimates below the reportable range were calculated using the raw data from the MWP and Cobas assays for this analysis; calculated results below the limit of quantitation were not used for clinical management.

The VQA produces external quality control materials (QCMs) by diluting a well-characterized HIV-1 virus stock in a plasma matrix to known HIV-1 RNA concentrations. These QCMs are used by client laboratories to monitor HIV-1 RNA testing run performance or by the VQA to evaluate client laboratory proficiency in HIV-1 RNA testing (4, 5). The VQA started using a real-time format to evaluate HIV-1 RNA testing proficiency in 2000, with 31 domestic laboratories (in the United States and Puerto Rico) participating in the program. This format of proficiency testing has been ongoing since that time, but the program has changed to include 21 domestic laboratories (in the United States and Puerto Rico) and 61 international laboratories from 19 different countries around the world. VQA external quality controls were initially produced using a human-derived citrated plasma matrix but were later produced using human-derived pooled commercial serum to which EDTA was added (6 mM final concentration) to meet the EDTA requirements for new assay platforms (the Roche TaqMan and Abbott RealTime assays). The VQA 1,500-copy/ml (VQA1500) external quality controls (lots 39 to 49) were created using citrated plasma and were used by JHU during real-time testing of the samples between July 2002 and October 2005. VQA1500 external quality control lot 77 was created using serum plus EDTA and was used by the VQA in the batch testing of the samples for this study. VQA HIV-1 RNA proficiency testing panels were created using citrated plasma from May 2001 to January 2010 and have been created using the serum-EDTA matrix since May 2010. Historical run control and proficiency testing data from JHU using the MWP assay and the VQA using the Cobas assay were compared to proficiency testing data generated by all VQA client laboratories using the Roche Monitor HIV-1 MWP and Cobas assays and were included in this analysis to help define systematic differences between the data generated with the two versions of the Roche Monitor test.

Clinical samples.

The samples used for this evaluation were samples stored from two ACTG clinical trials at the Biological Research Institute (BRI) ACTG biorepository in Rockville, MD: A5095, a phase III, randomized, double-blind comparison of three protease inhibitor-sparing regimens for the initial treatment of HIV-1 infection, and A5142, a phase III, randomized, open-label comparison of lopinavir-ritonavir plus efavirenz versus lopinavir-ritonavir plus 2 nucleoside reverse transcriptase inhibitors (NRTIs) versus efavirenz plus 2 NRTIs for the initial therapy for HIV-1 infection; both studies were conducted in the United States and Puerto Rico (6, 7). All available plasma samples from trial weeks 48 to 96 from 332 randomly selected participants who remained on their initial regimen through 48 weeks of follow-up, as well as year 2 longitudinal samples from 61 additional participants with confirmed virologic failure (2 consecutive HIV-1 RNA levels of >50 copies/ml during weeks 48 to 96 of trial follow-up, as determined by the MWP assay), were collected for this evaluation. Aliquots of frozen plasma with EDTA anticoagulant (EDTA-plasma) were shipped on dry ice with temperature monitoring from BRI to the VQA laboratory. Individual aliquots from the same participant and time point were thawed, pooled, and realiquoted to ensure that homogeneous samples with a sufficient volume were available for each assay and tested by each assay without refreezing of the plasma. The electronic data generated by JHU were received from Frontier Science & Technology Research Foundation, Inc. (Buffalo, NY), the data management center for the ACTG network. The data from the real-time testing and retesting of HIV-1 RNA in the clinical trial samples were used to evaluate the stability of HIV-1 RNA in the stored samples.

Analyses.

Three separate analyses were conducted with the VQA1500 external quality control data, the historical proficiency testing data, and the clinical sample data in order to estimate the systematic differences between the MWP and Cobas versions of the assay with respect to the log₁₀ recovery (the difference between the log₁₀ measured and the log₁₀ expected HIV-1 RNA concentrations). A variance components model was fit to the VQA1500 control values for log₁₀ recovery in the presence of between-lot variation. A mixed-effects model (consisting of fixed and random effects) was fit to the log₁₀ recovery and included smooth trends (specific to each version of the assay) across nominal log₁₀ concentrations, a fixed effect to estimate the constant shift due to the control matrix, and random effects for sample lot, testing round, and run (specific for each version of the assay and the control matrix). The clinical samples that were tested on both versions of the assay were cross-tabulated according to their availability, validity, and categorized value (undetectable, <50 copies/ml, ≥50 copies/ml). Agreement between the MWP and Cobas assay categorized results was assessed with the kappa statistic (values closer to 1.0 indicate stronger agreement between the MWP and Cobas assays). The quantitative values were depicted graphically, showing Cobas versus MWP assay results as well as a Bland-Altman plot (the differences between the two platforms versus their averages). For the clinical samples in which HIV-1 RNA was detectable by both versions, a mixed-effects model was fit to the differences between the estimated log₁₀ number of copies per milliliter for the two versions and included a term for the linear trend across average concentrations and random effects for the run (the term was separate for each version's contribution to the pair). Estimates from this model were used to derive a 90% prediction interval around the estimated trend line, and the equivalence of the two versions was assessed by comparing the prediction interval to an acceptance window of ±log₁₀(5), which is equal to ±0.7, to ascertain that the pairwise differences did not exceed a 5-fold threshold (8). To assess the impact of long-term freezer storage on HIV-1 RNA testing results, the residuals from the mixed model (a residual is the difference between an observed value and the value predicted by a model, i.e., the part of the observed value not explained by the model) were plotted versus the estimated number of years of freezer storage for the sample tested by the Cobas assay to ascertain whether there was a statistically significant trend of the effect of storage time on HIV-1 RNA recovery.

RESULTS

Comparison of VQA control data generated by JHU and the VQA.

The HIV-1 RNA results obtained from the testing of different production lots of the VQA1500 external quality control by JHU and the VQA are provided in Fig. 1. Each lot-specific acceptance range is depicted by gray bars surrounding the individual run data. Average concentrations for each platform and lot-specific concentrations were estimated and predicted, respectively, from a variance components model. Since the lots were numbered chronologically, this plot shows that there was no apparent systematic trend in lot-specific deviations from the average across time. The Cobas data show a high bias relative to the nominal concentration. The trends in log₁₀ recovery across the concentrations used in the VQA HIV-1 RNA proficiency testing panels are provided in Fig. 2. These panels were created using the same citrated and serum-EDTA matrices described for the VQA1500 external quality control. Figure 2 differentiates between the data generated by the VQA and other VQA client laboratories using the Cobas version and the data generated by JHU and other VQA client laboratories using the MWP version. Matrix-specific trends are highlighted. The log₁₀ recovery varied across the linear range for data generated using both versions of the assay, but the differences between the results obtained for controls created in EDTA versus citrate matrices varied by a relatively constant amount across the reportable range of both versions of the kit.

FIG 1.

Estimated concentrations for the VQA1500 control data from all runs within a control lot. Each lot-specific acceptance range is depicted by gray bars surrounding the individual run data (pink circles). A variance components model was used to estimate lot-specific predicted concentrations (blue triangles) along with the platform-specific estimated concentrations (thick red vertical line segments).

FIG 2.

Trends in log₁₀ recovery across the concentrations used for the proficiency testing panels created using either the EDTA or citrated matrix for laboratories using either the Cobas or the MWP version of the assay. The trend for all laboratories (heavy solid lines) and for the individual laboratory (JHU or VQA; dashed lines) are shown using matrix-specific colors.

Comparison of MWP and Cobas assay results for HIV-1 RNA in plasma samples.

There were a total of 1,872 sample pairs from 350 participants for all the data generated using the Cobas (VQA) and MWP (JHU) assays (Fig. 3). The valid data for each laboratory/platform were categorized into 3 groups: samples in which HIV-1 RNA was undetectable and samples with levels either below or above the threshold of 50 copies/ml. Among the 1,570 pairs that had valid data for both assays (upper left 3-by-3 quadrant in Fig. 3), 92% of the assessments agreed with regard to being above or below a quantification threshold of 50 copies/ml (kappa statistic = 0.757; 95% confidence interval [CI] = 0.716 to 0.797) (note that the data for the 725 pairs in which HIV-1 RNA was undetectable in both assessments were included in the computation; repeating this calculation excluding the 725 negative pairs yielded 85% agreement [kappa statistic = 0.691; 95% CI = 0.642 to 0.741]). Invalid data were due to either run errors or samples in which HIV-1 RNA was present at levels above the 100,000-copy/ml cutoff and that were not diluted and retested (9, 10). Thirty-three specimens did not have any material available for retesting on the Cobas platform (VQA). Data for the 964 pairs that had quantitative values in at least one version and nonmissing data in both versions (color-coded data in Fig. 3) are depicted in Fig. 4 and 5 using the same color-coded shading used in Fig. 3. Figure 4 compares samples in which HIV-1 RNA was detectable by either version (when the analysis is restricted to samples in which HIV-1 RNA was detected and in which the results were valid in both laboratories [n = 439], Pearson correlation r = 0.944 and P < 0.001; when the analysis is restricted to samples in which HIV-1 RNA levels were ≥50 copies/ml in both laboratories, r = 0.940 and P < 0.001), while Fig. 5 provides a Bland-Altman analysis of the difference between results obtained across the two versions (in which the analysis was restricted to samples in which HIV-1 RNA was detected and in which the results were valid in both laboratories). There was a concentration bias between paired assessments across average concentrations: the value obtained with the MWP (JHU) assay tended to be lower than the value obtained with the Cobas (VQA) assay for concentrations of <1,500 copies/ml and higher than the value obtained with the Cobas (VQA) assay for concentrations of >1,500 copies/ml. The 90% prediction interval on the estimated linear trend across concentrations fell within the ±5-fold acceptance window for concentrations ranging from approximately 100 to 100,000 copies/ml. The horizontal and vertical red reference lines in Fig. 5 intersect at the difference in estimated recovery for the VQA1500 external quality control in EDTA for the Cobas (VQA) and MWP (JHU) assays, as predicted from a model based upon all VQA1500 external quality control data (Fig. 1). The point at which the trend line crosses log₁₀(1,500) provides an estimate of the analogous quantity using the clinical samples (Fig. 5).

FIG 3.

Descriptive status of the 1,872 sample pairs (tested by the MWP and Cobas assays) from the clinical trial data. The color-coded shading of the cells corresponds to the shading in Fig. 4 and 5. *, the criteria for run errors were different for the two platforms and may include some crossover between optical density (OD) ratio errors, values for out-of-sequence wells, the presence of values for all HIV-containing and quantitation standard (QS) wells outside the optical density limit, and a computed value above the 100,000-copy/ml cutoff.

FIG 4.

A graphical depiction of the data highlighted in Fig. 3 that uses the same color-coded shading used in Fig. 3. These data include those for 964 sample pairs that could be compared quantitatively (samples for which quantitative values from at least one platform were available).

FIG 5.

Bland-Altman analysis. The difference between assays for 437 clinical specimen pairs that had quantitative results from both assays is shown. The dashed black lines depict 90% limit-of-agreement (LOA; the average difference in log₁₀ recovery [solid black line] ±1.65 times the standard deviation of the difference) intervals that include all sources of variability. The acceptance window was set at ±log₁₀(5), which is equal to ±0.7, to ascertain that differences between assays were less than 5-fold. The horizontal and vertical red reference lines intersect at the difference in estimated recovery for the VQA1500 control in EDTA for the Cobas (VQA) and MWP (JHU) assays, as predicted from a model based upon all VQA1500 control results (Fig. 1). The point at which the trend line crosses log₁₀(1,500) provides an estimate of the analogous quantity obtained using the clinical trial samples. The blue dashed trend line depicts the difference between the proficiency testing control panels with samples in EDTA for the Cobas (VQA) versus MWP (JHU) assays, as predicted from the model described in the legend to Fig. 2. The symbols are as defined in the keys in Fig. 4.

Impact of storage on plasma HIV-1 RNA results.

At 50, 200, and 1,500 copies/ml, the differences between clinical samples for the VQA laboratory (Cobas) and the JHU laboratory (MWP) were −0.16, −0.12, and 0.04 log₁₀ copies/ml, respectively. Figure 6 depicts the residuals from the mixed model versus the number of years of storage prior to retesting. Data for samples with viral loads of <50 copies/ml and ≥50 copies/ml are graphed separately. There was no significant linear trend in the residuals across years of storage for samples with either a low or a high concentration (P = 0.10 and P = 0.90, respectively).

FIG 6.

These residuals factor out the random effect estimates due to run and the bias due to trend across concentrations that were estimated from the mixed model. The trend lines across years of storage in the freezer prior to the second assay, depicted in separate panels for specimens having an average below and above 50 copies/ml, were not statistically significantly different. The points on this graph use the same symbols defined in the keys in Fig. 4.

DISCUSSION

We confirmed the long-term stability of HIV-1 RNA in plasma samples collected and processed for two ACTG multicenter protocols (A5095 and A5142) and stored in the ACTG BRI biorepository. An assessment of the long-term stability of HIV-1 RNA was performed by comparing the data generated after a median of 6.7 years of storage with the HIV-1 RNA data generated in real time as the clinical trials were being conducted. These analyses were confounded by changes in the HIV-1 RNA testing platforms used, and thus, the external quality control data generated by both the original and the new testing laboratories as part of the NIAID Virology Quality Assurance program were used to determine all the sources of variation that might impact the analysis and incorporate them into a model that could be used to evaluate the stability of HIV-1 RNA in the samples.

Establishing the integrity of the stored samples provides confidence in the ability to use stored samples to address scientific questions, such as in the study conducted by Lalama et al. to evaluate the impact of different HIV-1 RNA cutoffs for evaluating virologic failure across HIV-1 RNA assay platforms (11). In that study, the virologic endpoints were evaluated by retesting stored plasma by three FDA-approved HIV-1 RNA assays, the Cobas, Roche Cobas AmpliPrep Cobas TaqMan, and Abbott RealTime HIV-1 RNA assays, and comparing the classifying confirmed virologic failures across the different platforms using either a 50- or 200-copy/ml cutoff. While there was strong agreement in all pairwise comparisons across the three assays (91 to 94%; Cohen's kappa statistic = 0.76 to 0.82) and high agreement in classifying confirmed virologic failures (Cohen's kappa statistic = 0.75 to 0.85), the relative odds of identifying a confirmed virologic failure using a cutoff of 50 copies/ml was at least four times higher with the Abbott and Roche assays than with the Cobas assay; in contract, no differences were noted if a cutoff of 200 copies/ml was used. The stability of HIV-1 RNA in the same plasma samples was evaluated by comparing the data generated on the Cobas platform after a median time of 6.7 years in storage to the original data generated using the MWP assay for the clinical trials (6, 7). Systematic differences between assays not only can contribute to differences in virologic endpoints at different HIV-1 RNA cutoffs but also can confound issues associated with evaluating stability. Our analysis shows how the use of external proficiency testing controls that monitor analytical technique and changes in assay platform performance over time can help to demonstrate sample integrity in plasma specimens properly transported to and properly stored at the ACTG BRI biorepository. To accomplish this, a combination of data was included in this analysis: (i) HIV-1 RNA viral load data from a total of 1,872 EDTA-plasma samples collected from 350 participants that were assayed using two versions of the Roche Monitor assay; (ii) HIV-1 RNA data from the VQA1500 external quality control, which was included in all assays run in both laboratories; and (iii) VQA HIV-1 RNA proficiency testing data generated by the VQA and JHU as well as the combined data from all VQA client laboratories using the MWP and Cobas assay versions. Among the 1,570 pairs of samples with valid results, there was a 92% agreement (Cohen's kappa statistic = 0.76) with regard to being above or below the 50-copy/ml cutoff, thus showing the good comparability of the data. To better understand these comparisons and to further evaluate sample stability, we next looked at the quality control data generated in each testing laboratory.

The VQA1500 external quality control and VQA HIV-1 proficiency testing panels were created using the same well-characterized virus stock, produced over time, and were included in all the testing of samples included in this analysis by both the JHU and VQA. However, when the data from the VQA1500 control were first analyzed, a systematic difference in the average log₁₀ result obtained for all the controls tested was noted (average difference = 0.13, Cobas versus MWP; P < 0.001). Subsequent analyses showed that this difference was due to the matrix (citrate versus EDTA; average difference in log₁₀ recovery = −0.26, P < 0.001; matrix versus lab interaction effect, P = 0.0012) used to produce the external HIV-1 RNA copy control (human plasma containing citrate for MWP-generated data and serum plus EDTA for Cobas-generated data). Data submitted to the VQA program for proficiency testing were used to demonstrate that while there were some systematic differences between the MWP and Cobas assays, most notably, the differences in log₁₀ recovery at the extreme ends of the reportable range, these did not explain the differences noted in the VQA1500 external quality control testing. The matrix-specific effect, on the other hand, did explain the differences that were noted in the external control testing with each kit. The VQA started producing quality control materials when the new real-time PCR assays were introduced due to the requirement within the package inserts that EDTA-plasma be used. Differences in log₁₀ recovery both in proficiency testing and in assay validations that were being submitted to the VQA for review were noted for the controls produced in the different matrices (12). The use of magnetic particles for the extraction of nucleic acids in these newer assays rather than the use of alcohol-based precipitation of extracted nucleic acids in older manual extraction methods (Roche Monitor) likely explains the differences in how EDTA impacts the PCR efficiency. EDTA is known to have a high affinity for cations, including heavy metals (13), and so it is thought that EDTA carried through the extraction may chelate manganese or magnesium and thereby impact the PCR efficiency (14). Since the real-time PCR assays are calibrated using calibrators diluted in EDTA-containing diluents, the log₁₀ recovery of specimens lacking EDTA may be affected. It is important to note, however, that while there was a constant difference between panels produced with the citrated matrix and panels produced with the EDTA matrix across the linear range of both assays, this bias noted in control testing did not impact clinical sample testing, since the matrix of the clinical trial sample did not change over time, as all protocol samples were collected in EDTA-plasma, and this bias was not included in the model used to assess stability. Inspection of the data for protocol samples with estimated HIV-1 RNA titers of 50, 200, and 1,500 copies/ml showed that the differences between the data generated using the Cobas assay and those generated using the MWP assay were −0.16, −0.12, and 0.04 log₁₀ copies/ml, respectively. These differences were most likely attributed to differences in the nucleic acid capture techniques used during the detection of the amplified product, whereby the Cobas assay utilized probes bound to magnetic microparticles to capture the amplified product and the MWP assay used probes bound to a microwell plate to capture the amplified product. The external quality control performance (VQA1500, EDTA matrix) across the kit versions was comparable to the EDTA-plasma sample recovery (Fig. 5), confirming that the external quality control performance mirrored the protocol sample performance and demonstrated the utility of this type of external control for monitoring assay performance over time.

The main goal of this study was to evaluate the stability of HIV-1 RNA in EDTA-plasma samples stored at −80°C in the ACTG specimen repository. Baleriola et al. also demonstrated the stability of HIV, hepatitis C virus, and hepatitis B virus nucleic acids in specimens stored at −70°C and −20°C for a range of times (1.2 to 9.1 years) using the same assay to assess stability at each time point (15). The current study builds on this initial report in three major ways: first, we evaluated samples that were processed as part of two large multicenter clinical trials performed by multiple laboratories using standardized processes for sample handling, shipping, and storage of specimens, which provided evidence of sample integrity in an actual biorepository; second, because the samples used for this evaluation were generated for evaluating the efficacy of treatment, a broad range of viral loads, including those in samples with viral loads near and below the lower limit of detection of the different FDA-approved assays, could be evaluated without altering the specimen (i.e., diluting the specimen) to achieve low viral titers; and third, we showed how to examine stability across assay platforms, which is important, since many of the older HIV-1 RNA assay platforms are no longer available. We fitted the regression model for qualitative data (<50 copies/ml) and quantitative data (>50 copies/ml) from all time points in the data set without including time as a predictor. If RNA was lost over time, then at each nominal concentration, the later observations would have lower values than the earlier observations. Therefore, a plot of the residuals (a residual is the difference between an observed value and the value predicted by a model, i.e., the part of the observed value not explained by the model) against time would have a negative slope. The advantage to this approach compared with an approach in which time is added to the model as a predictor is that the analysis of residuals does not require us to specify the shape of the relationship between RNA and time. Incorrect specification of that shape could lead to erroneous conclusions about changes in RNA levels over time. The HIV-1 RNA log₁₀ recovery of samples tested in this study was quite consistent, regardless of the length of time that the specimens were held in storage; no statistical trends across years of storage were noted for samples with either low viral loads (<50 copies/ml, P = 0.10) or high viral loads (≥50 copies/ml, P = 0.90). The use of external quality control material whose performance mirrors the performance of actual clinical research samples was critical in identifying systematic variation between assay platforms so that this variation could be included in the stability modes and the sources of variation could be controlled for to permit the evaluation of stability even in samples with low viral titers. Figure 5 shows how the results for VQA external quality control materials used for proficiency testing align with the results from the retesting of the stored plasma. The stability of HIV-1 based on qualitative detection versus quantitative detection near the lower limits of the HIV-1 RNA assays was evaluated separately, but the model used for this analysis was still robust enough to evaluate stability even in samples with low titers.

The stability of analytes in repository specimens should be monitored intermittently over time. While this is not a challenge per se, interpretation of the data generated at different time points from stored samples may be complicated if the original testing was done on a different assay platform. The current study shows how data from external quality control material collected for proficiency testing and run validity to assess performance over time can be used to tease out the effects caused by systematic differences between assays and to evaluate the stability of the analyte without the added complication of interpreting data generated across assay platforms.

In summary, the current study shows that the inclusion of data from external quality control material used for assay validation and proficiency testing can be used to supplement analyses that include stored samples for purposes of evaluating new scientific questions, such as the impact of assay sensitivity on virologic outcomes or the impact of long-term storage on HIV-1 RNA in plasma samples.