Characterizing SARS-CoV-2 Viral Clearance Kinetics to Improve the Design of Antiviral Pharmacometric Studies

Abstract

A consensus methodology for the pharmacometric assessment of candidate SARS-CoV-2 antiviral drugs would be useful for comparing trial results and improving trial design. The time to viral clearance, assessed by serial qPCR of nasopharyngeal swab samples, has been the most widely reported measure of virological response in clinical trials, but it has not been compared formally with other metrics, notably model-based estimates of the rate of viral clearance. We analyzed prospectively gathered viral clearance profiles from 280 infection episodes in vaccinated and unvaccinated individuals. We fitted different phenomenological pharmacodynamic models (single exponential decay, bi-exponential, penalized splines) and found that the clearance rate, estimated from a mixed effects single exponential decay model, is a robust pharmacodynamic summary of viral clearance. The rate of viral clearance, estimated from viral densities during the first week following peak viral load, provides increased statistical power (reduced type 2 error) compared with time to clearance. Antiviral effects approximately equivalent to those with currently used and recommended SARS-CoV-2 antiviral treatments, notably nirmatrelvir and molnupiravir, can be detected from randomized trials with sample sizes of only 35 to 65 patients per arm. We recommend that pharmacometric antiviral assessments should be conducted in early COVID-19 illness with serial qPCR samples taken over 1 week.

Keywords: SARS-CoV-2; antiviral drugs; clearance rate; pharmacodynamics; phase 2; time to clearance.

FIG 1

Comparing model fits for three phenomenological models of SARS-CoV-2 viral clearance. a: Spaghetti plot of the data is shown with the mean predicted values from the three models (exponential: green; bi-exponential: orange; penalized splines: purple). The daily median viral loads are shown by the pink triangles. b-d: Distribution of the residuals as a function of time since peak viral load.

FIG 2

Differences in AUC for the estimated log viral loads in infection episodes in vaccinated (n = 17) and unvaccinated (n = 60) individuals. The estimated AUC is scaled by the intercept to remove the dependence on the baseline viral load. Approximate confidence intervals were calculated from a t test.

FIG 3

Times to viral clearance in vaccinated and unvaccinated individuals. a: Kaplan-Meier survival curves (with 95% confidence intervals) of the proportion still testing positive over time. b: Individual viral load profiles with daily median values shown by the triangles. Pink: vaccinated; dark blue: unvaccinated.

FIG 4

Power calculations for time to viral clearance or rate of viral clearance based on simulated clinical trial data. a: Median viral load (log₁₀ scale) over time for the simulated data (20% of patients are recruited before their peak, shown in gray; 80% of patients are recruited after their peak, shown in yellow). The thick lines show the median clearance profiles under no intervention; the dashed lines show the median clearance profiles for interventions with effect sizes of 50% (the effect size is defined as the proportional increase in rate coefficients α, β in the data generating model). b,c: Estimated power (1 – type 2 error) when comparing rates of clearance (thick lines) under a single exponential model or times to clearance (dashed lines) for four sampling schemes (once or twice daily for 5 or 7 days).