CDC reports on two alternative HIV testing algorithms

Amy Carpenter Aquino

August 2019—For HIV testing, a three-step algorithm that differs from the one recommended since 2014 can potentially reduce the number of tests performed and speed up the availability of viral load results, according to a CDC analysis presented at the HIV Diagnostics Conference in March. The CDC also evaluated a two-step algorithm that begins with the BioPlex 2200 HIV Ag-Ab differentiation assay and ends with the Aptima HIV-1 Quant assay.

The currently recommended CDC/APHL algorithm calls for three steps: HIV-1/2 antigen-antibody screening, followed by an HIV-1/2 antibody differentiation assay, and an HIV-1 nucleic acid test, if needed, for indeterminate or nonreactive results in the second step.

“At this point, the 2014 recommended algorithm has been implemented pretty widely in laboratories across the United States,” said Marc Pitasi, MPH, epidemiologist, Division of HIV/AIDS Prevention, CDC, in his presentation at the conference. Implementation challenges remain, however, mostly involving the use of a supplemental differentiation assay as step two and accessibility, resource requirements, and potential delays in turnaround time related to using qualitative NAT in step three. “Given those challenges, some laboratories might benefit from an alternative approach.”

One approach is to use a quantitative HIV nucleic acid test for HIV-1 viral load as step two, followed by antibody differentiation testing as step three if the viral load result is target not detected. (See “An alternative laboratory testing algorithm.”)

“Our objective was to provide data that could help demonstrate the potential feasibility of this alternative algorithm,” Pitasi said of the revised three-step algorithm. The CDC evaluated its performance against the 2014 algorithm using specimens that had been tested with five HIV-1/2 antigen-antibody combination screening assays and two HIV-1 viral load quantitative assays.

Specimens were collected from more than 6,000 patients of unknown HIV status but at high risk for HIV infection who were seeking testing at two clinics in Los Angeles between 2003 and 2005. Since specimen volume was limited, screening tests were performed on serum, and viral load tests were performed on plasma.

The five antigen-antibody screening tests used were the Abbott Architect HIV Ag/Ab Combo, Bio-Rad GS HIV Combo Ag/Ab EIA, Siemens Advia Centaur HIV Ag/Ab Combo, Bio-Rad BioPlex 2200 HIV Ag-Ab, and Alere Determine HIV-1/2 Ag/Ab Combo. All specimens with reactive results for any of the antigen-antibody screening tests were included in the analysis.

Specimens were followed through the alternative algorithm using a viral load quantitative test—Roche Amplicor HIV-1 Monitor or Hologic Aptima HIV-1 Quant—as step two. (Neither is FDA approved for diagnosis.) Specimens with no detected viral load were then tested with the Bio-Rad Geenius HIV 1/2 Supplemental Assay.

Specimens were classified according to the 2014 recommended algorithm. Five specimens showed early infection, defined as nonreactive/indeterminate Ab Geenius results and reactive Aptima qualitative assay results; 152 specimens showed established infection, defined as reactive Geenius results; and 38 specimens had false-positive screening results, defined as nonreactive/indeterminate Ab Geenius results and nonreactive Aptima qualitative assay results.

“We calculated the percentage of specimens that was correctly classified by antigen-antibody screening followed by viral load testing, relative to the recommended diagnostic algorithm individually for each screening and viral load test, and for any combination of these tests,” Pitasi said. Sensitivity for detecting early infection was between 53 and 60 percent for the Abbott, Siemens, and Bio-Rad BioPlex and Combo Ag/Ab assays, while the Alere assay showed a slightly lower sensitivity of 40 percent. “The confidence intervals were quite wide,” he noted.

For established infection, sensitivity and specificity were uniformly high among all five antigen-antibody screening tests.

Looking at viral load assay performance, the Roche Amplicor assay correctly classified all of the specimens with early infection detected by the antigen-antibody screening tests. In other words, they had quantified viral loads, Pitasi said. The Roche assay also correctly classified 95 percent of the 152 specimens with established infection. The proportion of detectable viral load was similar regardless of which screening test was used.

“Finally, all of the specimens [38] that were false reactive on any screening test were correctly classified, which in this case means they had a Roche viral load result of target not detected,” he said.

The Aptima Quant also correctly classified all of the specimens with early infection and all 110 specimens with established infection. Because of limited specimen volume, false-positive specimens were not tested with the Aptima Quant.

Eight specimens with established infection had undetectable virus on the Roche viral load test. “If the alternative algorithm were used here, these would go on to test with Geenius as the third step,” he said. There are no Aptima quantitative assay results on those eight specimens.

“Following the alternative algorithm, we saw that each of the eight specimens first was positive on every screening test, with the exception of the one specimen that was false-negative on the [Alere] Determine assay but positive on the other four screening tests.”

In step two, Roche Amplicor gave results of target not detected, which was considered a false-negative for diagnostic purposes.

The eight specimens then proceeded to testing on the Geenius assay as step three and showed HIV-1 positive results. Since all specimens tested positive on the Geenius assay, “each of these specimens would have been detected using the alternative algorithm,” with the possible exception of the specimen that tested false-negative on the Alere Determine assay. “With that same exception, they also would have all been detected using the recommended algorithm,” he added.

In shifting the differentiation supplemental testing to step three, however, only those specimens would require differentiation supplemental tests, compared with 152 tests using the recommended algorithm, Pitasi said.

“Keep in mind that all infected persons, of course, would go on to get a viral load eventually. We’re just shortening that process.”

Among the 38 specimens that had false reactive results on any of the five screening tests, eight specimens were false reactive on more than one screening test. “Using the recommended algorithm, all of these specimens would have proceeded to Geenius, followed by the Aptima qualitative assay, where they would have been resolved as negative,” Pitasi said. “Using the alternative algorithm, these specimens would have proceeded first to a viral load test, which yielded the result of target not detected for all eight.”

One specimen had a result of HIV-1 indeterminate on the Geenius assay and would likely require additional testing relative to the recommended algorithm. Cross-contamination or other sources of error could have explained that result, he suggested.

Of the remaining 30 of 38 specimens, four had Geenius assay results of HIV-2 indeterminate, which might require additional testing if the alternative algorithm were used. “Of these four specimens, three were false reactive on Abbott Architect, and one was false reactive on BioPlex,” he said, noting that all four specimen results had relatively low signal-to-cutoff values.

Pitasi noted the study’s limitations. Tests were performed on stored specimens, collected between 2003 and 2005; antigen-antibody screening and Ab supplemental testing were performed between 2008 and 2015 and viral load testing was performed in 2017. For the eight specimens not detected on viral load testing, “there could have been RNA degradation.” All positive specimens in the sample had high viral loads (more than 1,000 copies), so there was no investigation of the sensitivity of the alternative algorithm in lower viral load specimens. Limited specimen volume meant there was no repeat testing of initial reactive results for most of the antigen-antibody screening tests, which could have increased the number of false reactive specimens. Last, the Geenius tests were performed using software version 1.1, which was updated in 2017 to raise the cutoff value of the HIV-2 gp140 band.

“In closing, this alternative algorithm performs well overall with the vast majority of early and established infection confirmed by viral load testing at the second step,” Pitasi said.

Using the alternative algorithm would have averted antibody supplemental testing in more than 75 percent of the specimens examined, he said. “However, of the 46 specimens that did go on to Geenius, five of them had indeterminate results, including results of HIV-2 indeterminate, which might have required additional testing.”

Although the alternative algorithm would likely reduce the overall number of specimens with difficult-to-interpret Geenius results, a few of those specimens would still remain and could potentially trigger additional testing or blood draws.

“This alternative algorithm has the potential to reduce the total number of tests performed, avoid the potentially lengthy turnaround time related to obtaining Aptima Qual results, especially in laboratories that don’t do this in-house. It can also potentially expedite the availability of viral load results to improve patient care,” Pitasi said.