Performance characteristics of the Cavidi ExaVir viral load assay and the ultra-sensitive P24 assay relative to the Roche Monitor HIV-1 RNA assay

Abstract

Background: The Cavidi viral load assay and the ultra-sensitive p24 antigen assay (Up24 Ag) have been suggested as more feasible alternatives to PCR-based HIV viral load assays for use in monitoring patients infected with HIV-1 in resource-limited settings.

Objectives: To describe the performance of the Cavidi ExaVir Load™ assay (version 2.0) and two versions of the Up24 antigen assay and to characterize their agreement with the Roche Monitor HIV-1 RNA assay (version 1.5).

Study design: Observational study using a convenience sample of 342 plasma specimens from 108 patients enrolled in two ACTG clinical trials to evaluate the performance characteristics of the Up24 Ag assay using two different lysis buffers and the Cavidi ExaVir Load™ assay.

Results: In analysis of agreement with the Roche assay, the Cavidi assay demonstrated superiority to the Up24 Ag assays in accuracy and precision, as well as sensitivity, specificity, and positive and negative predictive values for HIV-1 RNA ≥ 400, ≥ 1000 and ≥ 5000 copies/mL. Logistic performance curves indicated that the Cavidi assay was superior to the Up24 assays for viral loads greater than 650 copies/mL.

Conclusions: The results suggest that the Cavidi ExaVir Load assay could be used for monitoring HIV-1 viral load in resource-limited settings.

Figure 1

Viral load measurements by Roche, Cavidi, Up24-JS and Up24-m assays for N = 342, 341, 306, and 274 plasma samples, respectively, are plotted: observed pairs of values (•), pairs with one or both values left-censored (○), pairs with one value missing (□). Missing by design and left-censored values were predicted using parameter estimates from bivariate mixed-effects models.

Figure 2

Bland-Altman plots for analysis of agreement between the Roche Monitor assay and the Cavidi, Up24-JS, and Up24-m assays are shown. For every pair of values plotted in Figure 1 (including pairs with both values observed (•), and pairs with intentionally missing (□) or left-censored (○) values) the average and difference of the two assay values was computed and plotted here along with model-based upper and lower 95% prediction limits (a.k.a. “limits of agreement”). The estimate of mean difference is indicated by a solid line segment. For Cavidi, Up24-JS, and Up24-m, both assay values were observed (•) for 244, 208, and 171 plasma samples, respectively.

Figure 3

Each logistic-linear performance curve provides estimates of the probability of detecting HIV-1 at given levels of viral load measured by the log₁₀ Roche Monitor. The relative positions of the curves favor the Cavidi RT assay for viral loads greater than 2.8.

Figure 4

Viral load change-from-baseline scores (Δ) for 234 plasma samples from 71 patients were obtained from observed assay values (•), and from predictions for assay values missing by design (□) or left-censored (○). Predicted values were computed using parameter estimates from bivariate mixed-effects models. Among [ΔCavidi, ΔRoche] pairs of scores, k = 137 were complete, c = 97 involved censored values, and m = 0 involved missing values. Among [ΔUp24-JS, ΔRoche] pairs, k = 117, c = 94, and m = 33. Among [ΔUp24-m, ΔRoche] pairs, k = 92, c = 77, and m = 65.