(W-041) Inferring HIV rebound timing from a noisy biomarker

Daniel Rosenbloom, PhD – Direrctor, Data Science, QP2 PHARMACOMETRICS, Merck & Co., Inc., Rahway, NJ, USA; Brad Evans, PhD – Principal Scientist, Pharmacokinetics, PDMB Analytical Translational Pharmacology, Merck & Co., Inc., Rahway, NJ, USA; Brian Squadroni, B.S – Assoc Prin. Scientist, Biology-Discovery, Engineering, Merck & Co., Inc., Rahway, NJ, USA; Jennifer Nguyen, PhD – Assoc Prin. Scientist, Biostatistics, Biometrics Research, Merck & Co., Inc., Rahway, NJ, USA; Livio Azzoni, PhD – Managing director, Biomedical Research Support Core, The Wistar Institute; Erin Coppola, BS – Scientist, Biology-Discovery, E, The Wistar Institute; Guoxin Wu, PhD – Prin. Scientist, Biology-Discovery, Cell Pharmacology, Merck & Co., Inc., Rahway, NJ, USA; Jill Maxwell, BS – Assoc Prin. Scientist, Biology-Discovery, HIV Discovery 2, Merck & Co., Inc., Rahway, NJ, USA; Jessicamarie Morris, AS – BSL2+ Lab manager, The Wistar Institute; Kenneth Lynn, MBA – Managing director, HIV curative research operations, University of Pennsylvania; Paul Zuck, MS – Sr. Prin. Scientist, Biology-Discovery, Antiviral & Antimicrobial Discovery, Merck & Co., Inc., Rahway, NJ, USA; Pablo Tebas, MD – Director & Principal Investigator, AIDS clinical Trial Unit (ACTU) research site, University of Pennsylvania; Karam Mounzer, MD – Chief Scientific Officer, Philadelphia FIGHT; Brian Ross, MS – Research Assistant, The Wistar Institute; Matthew Fair, BS – Research Assistant, The Wistar Institute; Emmanouil Papasavvas, PhD – Senior Scientist, The Wistar Institute; Emily Hiserodt, MPH – Research Director, Philadelphia FIGHT; Emily Welch, PhD – Head of R&D, Tasso, Inc.; Kevin Baterman, PhD – Retired, Merck & Co., Inc., Rahway, NJ, USA; Luis J. Montaner Montaner, PhD – Leader, HIV Research Program, Vaccine & Immunotherapy Center, The Wistar Institute; Bonnie Howell, PhD – VP., Research Science., Discovery, Preclinical and Translational Medicine (DPTM), Merck & Co., Inc., Rahway, NJ, USA

Objectives: The gold standard clinical endpoint for a candidate HIV cure therapy is delayed viral rebound during analytical treatment interruption (ATI). While in ATI, HIV viral load (VL) is monitored frequently to detect rebound (visits to clinic every 1-2 weeks). At-home sampling would allow the reduction on participant burden and study complexity and dry blood (db) samples can be used in this case. Here, we assessed new analytical methods to improve precision and interpretability of at-home sampling tools of db VL (dbVL) samples, in particular those samples form Tasso-M50 devices [1-3]

Methods: We analyzed dbVL and plasma VL (pVL) data from two BEAT-HIV trials (NCT03588715 and NCT03617198) where participants collected up to 8 db samples of ~40uL each from 2 Tasso-M50 devices [1]. We analyzed data from 15 participants from those trials, for a total of 324 pVL and 449 db assays (with up to 16 replicates). Our dbVL measurements were highly variable between replicates (66% of timepoints contain at least 2 replicate measurements >2-fold apart); to examine assay precision and asses dbVL noise levels, we compared the between-replicate variability of our db samples to 6 agency-approved assays, correcting for different input volumes. Moreover, dbVL measurements were often discordant with pVL samples taken at the same time (64% of samples with undetectable pVL have dbVL counts above the detection limit); to assess such discrepancies, we developed non-parametric methods to predict pVL rebound based on our repeated dbVL measures [2,3]. We then tested which endpoint would provide a prediction that is most appropriate to drive decisions (e.g. threshold vs. probability of rebound)

Results: We showed that db assay is modestly less precise than commercially approved assays (43%CV for dbVL vs 10-33%CV, for 5000-10^5 cp/mL VL range), when correcting for the smaller input volume. Our rebound tests showed good pVL rebound prediction power (e.g., ROC curve AUC 83%-90% when 1000c/mL). A rebound threshold for each test was selected by maximizing the correct prediction of low pVL, i.e. no rebound (rebound at pVL ≥1000c/mL, test thresholds at: 1000c/ml median dbVL, 1.8 radio of dbVL to background medians, p-val of 0.007 Mann Whitney U test). Our rebound tests accounting for background signal were modestly better at predicting no rebound (0.92 vs 0.87 proportion of correct predictions)

Conclusions: We showed that (a) the noise observed in the dbVL samples is measurable and comparable to that of other HIV tests, suggesting that small sample size is the primary cause of imprecision, and can be controlled with replicate measurements; (b) our rebound tests correctly predict low pVL (no rebound) up to 92% of the time, suggesting that the dbVL data can help keep a patient at home and determine if VL has rebounded. This project further advances our knowledge of at home sampling methods for viral load and may allow for improved patient centric sampling in future ATI trials

Citations: [1] Dubé K, Agarwal H, Carter WB, et al, HIV research & clinical practice vol. 23,1 (2022): 76-90.
[2] Rosenbloom, Daniel S. et. al, Poster 1091, CROI 2024; USA; March 3-6, 2024.
[3] Azzoni, Livio et. al, Poster 1106, CROI 2024; Denver, CO, USA; March 3-6, 2024.