(T-066) A Novel Concentration-Based Time-Imputation Algorithm for Large Cohort Studies Missing Time-After-Dose Data for Stored Samples

Andrew Edmonds, PhD – Associate Professor, Department of Epidemiology, Gillings School of Global Public Health, The University of North Carolina at Chapel Hill, Chapel Hill, NC, USA; Bonnie Shook-Sa, DrPH – Assistant Professor, Department of Biostatistics, Gillings School of Global Public Health, The University of North Carolina at Chapel Hill, Chapel Hill, NC, USA; Yen Chang, MS – Graduate Research Assistant, Department of Biostatistics, Gillings School of Global Public Health, The University of North Carolina at Chapel Hill, Chapel Hill, NC, USA; Stephen Cole, PhD – Professor, Department of Epidemiology, Gillings School of Global Public Health, The University of North Carolina at Chapel Hill, Chapel Hill, NC, USA; Cecile Lahiri, MD – Associate Professor, Division of Infectious Diseases, Department of Medicine, Emory University School of Medicine, Atlanta, GA, USA; Phyllis Tien, MD – Professor, Department of Medicine, University of California, San Francisco, San Francisco, CA, USA; Igho Ofotokun, MD – Professor, Division of Infectious Diseases, Department of Medicine, Emory University School of Medicine, Atlanta, GA, USA; Mirjam-Colette Kempf, PhD – Professor, Schools of Nursing, Public Health, and Medicine, University of Alabama at Birmingham, Birmingham, AL, USA; Maria Alcaide, MD – Professor, Division of Infectious Diseases, Department of Medicine, University of Miami Miller School of Medicine, Miami, FL, USA; Audrey French, MD – Professor, Department of Internal Medicine, Division of Infectious Disease, Stroger Hospital of Cook County, Chicago, IL, USA; Joseph Margolick, MD – Professor, Johns Hopkins University Bloomberg School of Public Health, Baltimore, MD, USA; Todd Brown, MD, PhD – Professor, Johns Hopkins University School of Medicine, Baltimore, MD, USA; Kathryn Anastos, MD – Professor, Department of Medicine, Albert Einstein College of Medicine and Montefiore Medical Center, Bronx, NY, USA; Ken Ho, MD – Associate Professor, Department of Medicine, University of Pittsburgh, Pittsburgh, PA, USA; Jack DeHovitz, MD – Professor, Downstate Health Sciences University, State University of New York, Brooklyn, NY, USA; Stephen Gange, PhD – Professor, Johns Hopkins University Bloomberg School of Public Health, Baltimore, MD, USA; Daniel Merenstein, MD – Professor, Department of Medicine, Georgetown University Medical Center, Washington, DC, USA; Bani Tamraz, PharmD, PhD – Associate Professor, School of Pharmacy, University of California, San Francisco, CA, USA; Bradley Aouizerat, PhD – Professor, Department of Physiological Nursing, University of California, San Francisco, CA, USA; Anandi Sheth, MD – Professor, Division of Infectious Diseases, Department of Medicine, Emory University School of Medicine, Atlanta, GA, USA; Steven Wolinsky, MD – Professor, Division of Infectious Diseases, Department of Medicine, The Feinberg School of Medicine, Northwestern University, Chicago, IL, USA; Julie Dumond, PharmD, MS – Associate Professor, Division of Pharmacotherapy and Experimental Therapeutics, UNC Eshelman School of Pharmacy, Chapel Hill, NC, USA

Objectives: Longitudinal studies with biospecimen collection are rich for pharmacokinetics (PK)-adverse effects analyses but may lack time-after-dose (TAD). Using dolutegravir (DTG) PK and TAD in women (Women’s Interagency HIV Study; WIHS), we developed a novel method to infer TAD in men (Multicenter AIDS Cohort Study; MACS), which never collected TAD.

Methods: We observed a trimodal TAD distribution in WIHS women on DTG, with each of the 3 components characterized by mean, scale, and shape parameters of a skew-normal distribution. We assumed a similar distribution in MACS and verified it by comparing TAD in WIHS to TAD in MACS/WIHS Combined Cohort Study (MWCCS), which has collected TAD since 2020, with most participants originally from WIHS or MACS. We developed an algorithm inspired by the k-nearest neighbors algorithm and multiple imputation methods to impute TAD in MACS using WIHS (R v.4.3.2). Our algorithm assigned each MACS timeless DTG concentration to one of the WIHS TAD components by sampling from the predicted population-level distribution of WIHS DTG concentrations at each component’s mode. It then picked the component where the sampled concentration was closest in absolute distance to the timeless concentration, using a correction factor inversely proportional to the number of observations per component. A time was then sampled from the chosen component’s TAD distribution and paired with the timeless concentration. This process repeated for all concentrations over 1000 iterations, creating 1000 imputed datasets, which were fit in NONMEM using a 1-compartment PK model. Population-level PK parameters of 1000 estimates were summarized as median and median absolute deviation (MAD).

Results: Like WIHS, we observed a trimodal TAD distribution in MWCCS participants regardless of their HIV regimen, with 3 TAD modes reflecting 3 trends: 2 hrs (AM dose/AM visit), 13.5 hrs (PM dose/AM visit), and 24.5 hrs (AM dose on the day prior). For internal validation of algorithm performance, we used WIHS data for training and testing. The WIHS population estimate and 1000 estimates’ medians (MAD; % change) for clearance, volume of distribution, and absorption rate constant were 1.27 vs. 1.27 L/h (0.01; 0%), 36 vs. 30 L (3.3; -16%), and 1.23 vs. 1 h-1 (0.32; -23%), respectively. For external validation, we used WIHS for training and a Phase 1 DTG PK study in men [1] for testing. The corresponding values for the external dataset were 0.94 vs. 1.14 L/h (0.04; +20%), 20.8 vs. 20.8 L (2.3; 0%), and 1.91 vs. 2 h-1 (1.4; +4.5%).

Conclusions: We observed a trimodal TAD distribution in WIHS and MWCCS that may apply to similar studies with daily dosing. Our method leverages known PK to predict TAD and allows the use of data with missing TAD. Future work includes adding a drug taken with DTG to the algorithm and exploring alternative statistical methods. Our goal is to estimate AUCs in MACS to analyze the causal effects of DTG exposure on cardiometabolic outcomes.

Citations: [1] Greener et al., J Acquir Immune Defic Syndr. 2013;64(1):39-44. PMID: 23945251