Population pharmacokinetics and exposure-response analysis of a single dose of sotrovimab in the early treatment of patients with mild to moderate COVID-19

Abstract

Sotrovimab is a recombinant human monoclonal antibody that has been shown to prevent progression to hospitalization or death in non-hospitalized high-risk patients with mild to moderate coronavirus disease 2019 following either intravenous (i.v.) or intramuscular (i.m.) administration. Population pharmacokinetic (PopPK) and exposure-response (ER) analyses were performed to characterize single dose sotrovimab pharmacokinetics (PK) and the relationship between exposure and response (probability of progression), as well as covariates that may contribute to between-participant variability in sotrovimab PK and efficacy following i.v. or i.m. administration. Sotrovimab PK was described by a two-compartment model with linear elimination; i.m. absorption was characterized by a sigmoid absorption model. PopPK covariate analysis led to the addition of the effect of body weight on systemic clearance and peripheral volume of distribution, sex on i.m. bioavailability and first-order absorption rate (KA), and body mass index on KA. However, the magnitude of covariate effect was not pronounced and was therefore not expected to be clinically relevant based on available data to date. For ER analysis, sotrovimab exposure measures were predicted using the final PopPK model. An ER model was developed using the exposure measure of sotrovimab concentration at 168 h that described the relationship between exposure and probability of progression within the ER dataset for COMET-TAIL. The number of risk factors (≤1 vs. >1) was incorporated as an additive shift on the model-estimated placebo response but had no impact on overall drug response. Limitations in the ER model may prevent generalization of these results to describe the sotrovimab exposure-progression relationship across severe acute respiratory syndrome-coronavirus 2 variants.

---

Study Highlights

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

The efficacy of sotrovimab in preventing the progression of coronavirus disease 2019 (COVID-19) in non-hospitalized patients with mild to moderate COVID-19 at high risk for disease progression has been evaluated in two pivotal clinical trials.

• WHAT QUESTION DID THIS STUDY ADDRESS?

What are the sources of variability in sotrovimab exposure and exposure-response (ER)? What is the relationship between sotrovimab exposure and prevention of progression of COVID-19?

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

Body weight, sex, and body mass index were covariates of sotrovimab exposure but are not anticipated to be clinically relevant. Sotrovimab concentrations at 96 and 168 h are significant predictors of COVID-19 progression. Number of risk factors is a covariate of the ER model–predicted placebo progression rate, with no impact on drug effect.

• HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

The population pharmacokinetic model supports dosing recommendations for sotrovimab in early treatment and can be used to inform any future applications. The ER model informs on exposure measures found to be predictive of progression of mild to moderate COVID-19 in early treatment of COVID-19.

---

INTRODUCTION

Preventing infection and reducing the risk of progression of coronavirus disease 2019 (COVID-19) remains an urgent public health priority. Although currently available vaccines have been shown to be efficacious in preventing severe COVID-19,^1,2 certain subgroups of individuals remain at higher risk of severe COVID-19, resulting in hospitalization and increased risk of mortality. These subgroups include people 55 years of age or older; those with unvaccinated status; and those with comorbidities, including cardiovascular disease, diabetes, renal disease, neurologic conditions, and immune suppression.^3–6 Care of patients with severe COVID-19 is resource intensive and poses a substantial burden on hospital reserves; thus, preventing progression to severe disease among high-risk patients with mild or moderate COVID-19 is an important goal of treatment.^7,8

Sotrovimab was developed to treat mild to moderate COVID-19 in non-hospitalized patients at high risk of disease progression.^9,10 Sotrovimab is a recombinant human IgG1 monoclonal antibody (mAb) that binds to a conserved epitope within the virus spike protein receptor binding domain of severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2).^11,12 The fragment crystallizable domain of sotrovimab includes the two amino acid “LS” modification that has been shown to extend antibody half-life in studies of other mAbs and may enhance distribution of sotrovimab to the respiratory mucosa.^13–15

The efficacy of early treatment with sotrovimab in preventing the progression of COVID-19 in those who are at risk for hospitalization or death has been demonstrated in two pivotal clinical trials of non-hospitalized patients with mild to moderate COVID-19 at high risk for disease progression.^9,10,16 In the COMET-ICE study, a single 500 mg dose of sotrovimab administered intravenously (i.v.) demonstrated a relative risk reduction of 79% in hospitalization greater than 24 h for acute management of any illness or death due to any cause through day 29 compared with placebo.¹⁰ In the COMET-TAIL clinical trial, efficacy of a single 500 mg dose of intramuscular (i.m.) sotrovimab demonstrated noninferiority* compared with i.v. sotrovimab given at the same dose, with a low incidence of COVID-19 progression observed for both routes of administration.¹⁶ Sotrovimab was well tolerated in these studies, specifically with a low frequency of infusion-related reactions and only mild and transient injection-site reactions.

The objective of this work was to characterize factors that influence systemic exposure and response of sotrovimab in order to inform and optimize future clinical usage. Population pharmacokinetic (PopPK) and exposure-response (ER) analyses were conducted to (1) characterize the PopPK of sotrovimab following i.v. and i.m. administration, (2) identify and quantify the effects of intrinsic and extrinsic factors influencing the pharmacokinetics (PK) of sotrovimab using systematic covariate analysis, (3) describe the relationship between sotrovimab exposure and probability of progression of COVID-19 based on i.v. and i.m. data from the COMET-TAIL study, and (4) identify clinical covariates that influence variability in efficacy response.

METHODS Clinical studies

Data for PopPK analysis were derived from four clinical studies in non-hospitalized patients with COVID-19: COMET-ICE (ClinicalTrials.gov identifier, NCT04545060), COMET-TAIL (NCT04913675), COMET-PEAK (NCT04779879), and BLAZE-4 (NCT04634409), together with one study in healthy volunteers of Japanese or Caucasian descent (NCT04988152). Data for the sotrovimab ER analysis were derived from COMET-TAIL. Descriptions of the clinical study designs are provided in Table S1; additional details have been published elsewhere.^9,10,16–19

All participants who received sotrovimab and had at least one measurable concentration of drug were included in the PopPK analysis dataset. Study participants in the analysis dataset received sotrovimab as single i.v. (500 mg) or i.m. (250 mg or 500 mg) doses. In BLAZE-4, sotrovimab (500 mg i.v.) was administered in combination with bamlanivimab. All studies were conducted in accordance with the ethical principles derived from the Declaration of Helsinki and Council for International Organizations of Medical Sciences International Ethical Guidelines, applicable International Council for Harmonisation Good Clinical Practice guidelines, and applicable laws and regulations. Ethics approval was obtained from institutional review boards and ethics committees. Written informed consent was provided prior to study entry; participants less than 18 years of age signed an assent form, and a parent/guardian provided written consent.

PK sampling and assay

Venous blood samples were obtained at different times prior to and after treatment administration (Table S1). Serum sotrovimab concentrations were determined using a validated electrochemiluminescent method validated on the Meso Scale Discovery, with a lower limit of quantification of 0.1 μg/mL.

PopPK modeling software and dataset

PopPK modeling of sotrovimab concentration-time data was performed using NONMEM (version 7.3). A pooled NONMEM-ready dataset was constructed using SAS (version 9.4 or higher). The dataset contained dosing history, infusion rate, sotrovimab plasma concentration data, relevant laboratory baseline values, and demographic and covariate information.

PopPK model development

Model development was performed in a two-stage approach. An adequate model for sotrovimab PK was first developed with data following i.v. administration, after which data collected following i.m. administration were added to the analysis and the model was extended to include absorption processes specific to the i.m. route of administration. Initial development of the joint i.v. and i.m. PopPK model included only participants with dense PK data. At the end of base model development, parameter estimation was performed on the entire dataset including all participants with dense or sparse data.

Exploratory analysis and prior knowledge of typical PK of therapeutic antibodies informed the selection of the functional form of the base structural model. Initially, a two-compartment model was assessed for appropriateness in describing the PK of sotrovimab. The variability model included random effect terms on the PK parameters elimination clearance (CL), central and peripheral volume of distribution (V2 and V3, respectively), absorption rate (KA), and bioavailability after i.m. injection (F_IM) to describe the interindividual variability (IIV), and a combined additive and proportional error model to describe the residual variability (RV). Model evaluation was based on model diagnostics, goodness-of-fit (GOF) plots, and simulation-based visual predictive checks (VPCs). Various alternative models were applied to the data and assessed for their capacity to sufficiently characterize the PopPK of sotrovimab, as needed.

Absolute estimates of disposition parameters following i.v. administration of sotrovimab were available from the clinical study data. Additionally, the rate and extent (i.e., absolute bioavailability) of sotrovimab absorption after i.m. injections were estimated. A first-order model for sotrovimab absorption after i.m. administration was initially tested, followed by more complex absorption models.

Following the development of an appropriate base structural model, the influence of covariates on selected parameters was evaluated using a systematic forward inclusion and backward elimination approach. Covariates were added sequentially to the base model starting with the covariate contributing the most significant change in the minimum objective function value (OFV; smallest p < 0.01) and a reduction in IIV in the parameter of interest of at least 5%. This process was repeated until there were no further covariates that produced significant changes in the OFV. Each covariate's significance was tested individually with backward elimination until all remaining covariates were significant (change in OFV of at least 10.83; p < 0.001). Covariates investigated included age, sex, self-reported racial classification,²⁰ disease state (healthy vs. COVID-19), body weight, body mass index (BMI), hepatic function category (National Cancer Institute [NCI] classification), renal function category (based on estimated glomerular filtration rate; Modification of Diet in Renal Disease equation), serum albumin concentration, concomitant use of dexamethasone and/or remdesivir, baseline viral load, and sotrovimab clinical trial material (Gen 1 pool vs. Gen 2 clonal cell line; Table S2). In the healthy volunteer study, no serum albumin data were available, so covariate analysis for this factor was restricted to subjects from the remaining studies. For continuous variables, missing values were replaced by the respective study- and sex-specific median values in the dataset. For categorical variables, missing values were grouped with the respective unspecified category (e.g., “unknown” or “other”).

The reduced multivariable model, including all significant covariates, was evaluated for any remaining biases in the IIV and RV error models. Adequacy of the final model was evaluated using a simulation-based VPC method. Utilizing NONMEM, the final model was used to simulate 500 replicates of the analysis dataset sufficient to achieve at least 100,000 patients per stratum of the VPC. Statistics of interest were calculated from simulated and observed data for comparison.

To characterize the impact of intrinsic and extrinsic factors on sotrovimab PK, simulations were performed based on the final PopPK model and individual Bayesian estimates of PK parameters following actual treatment received. Numerical integration was performed to predict sotrovimab maximum concentration (C_max), concentration at 96 h (C_96h), and concentration at 168 h (C_168h) for each patient. Summary statistics of the simulated exposures were calculated and stratified by covariate group (discrete covariates) or quantile (continuous covariates). Clinically meaningful impact on sotrovimab PK was determined based on geometric mean ratios (GMRs) and 90% confidence intervals (CIs) when compared to clinical bounds of 0.5 and 2.0 relative to the reference group (additional details are provided in the Supplementary Material).

ER modeling

For the COMET-TAIL ER analysis, a NONMEM-ready dataset was constructed, which included sotrovimab dosing (dose and timing), treatment assignment, efficacy responses, demographic data, and clinical covariates. Patients excluded from PopPK analysis were also excluded from the ER analysis. The endpoint used for COMET-TAIL ER modeling included the primary efficacy endpoint from COMET-TAIL, which was the progression of COVID-19 through day 29 as defined by hospitalization greater than 24 h for acute management of illness due to any cause or death. Individual measures of sotrovimab exposure for each subject in the COMET-TAIL PK dataset were generated via integration of the predicted concentration-time profile for each patient based on the final PopPK model and individual empiric Bayesian PK parameter estimates. These measures included predicted serum concentrations 24, 48, 72, 96, and 168 h postdose, average concentrations, and area under the curve (AUC) from time zero to 24, 48, 72, 96, and 168 h after the dose, as well as AUC from time zero to day 28 postdose (AUC_0-day28).

Separate logistic regression models were developed for each exposure measure to determine if sotrovimab exposure was a statistically significant (α = 0.05) predictor of the probability of the progression endpoint. The exposure measure(s) selected for inclusion in the base logistic regression models was chosen based on statistical assessment in addition to clinical considerations.

The influence of covariates on the probability of progression of COVID-19 through day 29 was evaluated using forward selection with α = 0.01. Covariates evaluated included age, sex, BMI, number of risk factors (inclusive of age and BMI), number of other risk factors (exclusive of age and BMI), symptom duration (continuous and categorical), baseline viral load, and route of administration (i.v. or i.m.).

Following forward selection, the full logistic regression model was used to predict the probability of the efficacy endpoint for various levels of categorical variables, or over the observed range of each continuous covariate that was statistically significant. This full model was evaluated for any remaining biases using a simulation-based VPC method.

RESULTS Final PopPK analysis dataset

A total of 1984 participants contributed 14,269 sotrovimab concentration measurements to the PopPK model. The final PopPK dataset included 11,772 samples. A total of 2497 samples were excluded from the analysis due to missing sample date and/or time information, duplicate sample date and/or time, predose and postdose below the lower limit of quantification samples, predose measurable concentration, or analyst-identified outliers (conditional weighted residuals [CWRES] < −5 or CWRES >5; Table S3). PK samples were also excluded from the analysis if they were deemed nonphysiologic or anomalous measured concentrations (end of infusion concentration <50 μg/mL or >500 μg/mL, assuming typical plasma volume of 3 L [range 1 to 10 L]). These records were retained but flagged in the data file and excluded during the analysis. The numbers of participants and concentrations available by study and treatment group are presented in Table S4.

PopPK demographics and baseline characteristics

Median age was 49 years, 44.9% were male, and 88.4% were White; median body weight was 83.6 kg, and median BMI was 30.4 kg/m² (Table S5). The analysis included 1891 patients with COVID-19 and 38 healthy volunteers. In total, 1415 participants had normal renal function, whereas 401, 65, and five participants had mild, moderate, and severe renal impairment, respectively. Moreover, 1393 participants had normal hepatic function, and 487 and five participants had mild and moderate hepatic impairment (NCI criteria), respectively.

PopPK analysis Base model

The PK of sotrovimab was best described by a two-compartment base model with first-order elimination. The absorption of the i.m. data was best described by a sigmoid absorption model, which was implemented using a zero-order input process into a depot compartment followed by first-order absorption into the central compartment (Figure 1).

FIGURE 1. Base population PK model for i.v. and i.m. sotrovimab. CL, elimination clearance; FIM, bioavailability after i.m. injection; i.m., intramuscular; i.v., intravenous; KA, first-order rate of absorption of the i.m. absorption compartment; PK, pharmacokinetic; Q, distribution clearance; R1, zero-order input rate for first i.m. absorption compartment; V2, central volume of distribution; V3, peripheral volume of distribution.

Various models were tested to characterize the IIV of the PK of sotrovimab, including IIV terms on V2, V3, CL, KA, and F_IM. The constant coefficient of variation (CCV) and combined CCV and additive error models were tested to characterize the RV. The IIV terms were added to CL, V2, V3, KA, and F_IM using the full covariance matrix. RV was described by a combined additive and CCV RV model. All parameters were estimated with good precision (percent relative standard error [%RSE] <8%), except the logit of F_IM (124% RSE). Parameter shrinkage for the IIV parameters was 11.2% to 33.0%. The magnitude of the residual errors was moderate (%CV <14% for concentrations >10 μg/mL).

Final model

Covariate analysis via a stepwise forward selection and backward elimination approach led to the addition of the effect of body weight on systemic CL and V3, sex on F_IM and KA, and BMI on KA (Table 1). Variability in sotrovimab PK was additionally described by IIV on CL, V2, V3, F_IM, and KA with a full covariance matrix and an additive plus CCV RV model. The final PopPK parameter estimates, standard errors, and covariate effects are shown in Table 1. All fixed effect parameters were estimated with good precision (≤20% RSE). The magnitudes of the IIV were 29–57% CV for CL, V2, V3, F_IM, and KA. With three exceptions concerning non-diagonal elements in the covariance matrix, the random effect parameters were estimated with good precision (<28% RSE).

TABLE 1 Final population PK parameter estimates and covariate effects.

Parameter	Final parameter estimate		Magnitude of variability
	Population mean	%RSE	Final estimate	%RSE
CL
Elimination clearance in participants of 83.6 kg (L/day)	0.0960	1.33	38.2 %CV	3.33
Power of body weight effect	0.494	7.18
V2
Central volume of distribution (L)	3.33	2.20	57.2 %CV	2.54
Q
Distribution clearance (L/day)	0.667	1.49	NE	NA
V3
Peripheral volume of distribution in participants of 83.6 kg (L)	4.51	1.32	29.4 %CV	5.17
Power of body weight effect	0.757	6.08
KA
Absorption rate in male participants with BMI of 30.41 kg/m2 (L/h)	0.00643	4.67	55.4 %CV	11.9
Power of BMI effect	−0.711	20.0
Proportional shift in female participants	−0.323	11.9
FIM
i.m. bioavailability in male participants	0.582	2.64	42.9 %CVa	8.53
Shift in female participants, on logit scaleb	−0.449	16.4	42.9 %CVc
R1
Rate of input (mg/h)	130	6.69	NE	NA
cov(IIV in V2, IIV in CL)	0.140d	3.57	NA	NA
cov(IIV in FIM, IIV in CL)	0.180e	11.8	NA	NA
cov(IIV in FIM, IIV in V2)	0.317f	11.7	NA	NA
cov(IIV in KA, IIV in CL)	−0.0233g	51.5	NA	NA
cov(IIV in KA, IIV in V2)	−0.0874h	27.6	NA	NA
cov(IIV in KA, IIV in FIM)	0.303i	12.0	NA	NA
cov(IIV in V3, IIV in CL)	0.0672j	6.23	NA	NA
cov(IIV in V3, IIV in V2)	0.0348k	20.8	NA	NA
cov(IIV in V3, IIV in FIM)	0.00744l	308	NA	NA
cov(IIV in V3, IIV in KA)	−0.0171m	74.7	NA	NA
CCV residual variability component	0.0175	0.693	177–13.2 %CV F [0.1-4000]n	NA
Additive residual variability component	0.0312	10.3		NA
Minimum value of the objective function = 48,830.476

Abbreviations: %CV, coefficient of variation expressed as a percent; %RSE, relative standard error expressed as a percent; BMI, body mass index; CCV, constant coefficient of variation; CL, clearance; F, model prediction; IIV, interindividual variability; i.m., intramuscular; KA, absorption rate; LF_IM, logit of bioavailability after i.m. injections; NA, not applicable; NE, not estimated; PK, pharmacokinetic; SQRT, square root.

^aThe magnitude of IIV (%CV) of i.m. bioavailability in male participants was calculated using the following equation: 100 × (1 − 0.582) × 1.03.

^bThe covariate effect of sex on F_IM was incorporated using the following transformation: F_IM = exp(LF_IM)/(1 − exp(LF_IM)), LF_IM = log(F_IM/(1 − F_IM)) + sexf × (−0.449), where sexf = 1 if sex is female and 0 else.

^cThe magnitude of IIV (%CV) of additive shift in female participants was calculated using the following equation: 100 × (1–0.582) × 1.03.

^dThe calculated correlation coefficient (r) associated with cov(IIV in V2, IIV in CL) was 0.714 with r² = 0.509.

^eThe r associated with cov(IIV in F_IM, IIV in CL) was 0.480 with r² = 0.231.

^fThe r associated with cov(IIV in F_IM, IIV in V2) was 0.588 with r² = 0.345.

^gThe r associated with cov(IIV in KA, IIV in CL) was −0.122 with r² = 0.0148.

^hThe r associated with cov(IIV in KA, IIV in V2) was −0.318 with r² = 0.101.

ⁱThe r associated with cov(IIV in KA, IIV in F_IM) was 0.577 with r² = 0.333.

^jThe r associated with cov(IIV in V3, IIV in CL) was 0.631 with r² = 0.399.

^kThe r associated with cov(IIV in V3, IIV in V2) was 0.227 with r² = 0.0514.

^lThe r associated with cov(IIV in V3, IIV in F_IM) was 0.0255 with r² = 6.49E−04.

^mThe r associated with cov(IIV in V3, IIV in KA) was −0.115 with r² = 0.0131.

ⁿThe magnitude of residual variability (%CV) was calculated using the following equation: (SQRT(0.0175 × F² + 0.0312)/F) × 100. Shrinkage estimates: 14.6% in IIV for CL, 21.0% for IIV in V2, 24.2% for IIV in F_IM, 6.0% for IIV in KA, and 24.9% for IIV in V3.

Based on the final model and a reference male participant of 83.6 kg and BMI of 30.41 kg/m², the typical value for systemic CL was 0.096 L/day, V2 was 3.33 L, V3 was 4.51 L, and the rate of absorption following i.m. dosing was 0.00643 L/h. The typical value of bioavailability in male participants was 0.582 and in female participants was 0.471. The model-estimated median half-life was 61.2 days (Table S6). GOF plots for the final PopPK model, stratified by route of administration, show the model adequately described the observed data for both routes of administration (Figure S1). VPC plots stratified by route and dose show that observed concentrations were mostly contained within the range of the 5th and 95th percentiles of the model-predicted concentration values (Figure 2). Overall, the final model captured the central tendency (median) and the extent of variability (5th and 95th percentiles) of observed PK data following i.v. and i.m. administration well.

FIGURE 2. Visual predictive check plots for the final population PK model of sotrovimab. CI, confidence interval; PK, pharmacokinetic.

The magnitude of covariate effects on key exposure measures (C_max, C_96h, and C_168h) were summarized using forest plots (Figure 3; Figures S2 and S3). All of the statistically significant covariates (body weight, BMI, and sex) were contained within the prespecified 0.5 to 2.0 bounds. For i.v. sotrovimab PK, body weight was associated with GMRs (90% CI) fully contained within standard bioequivalence bounds (0.8, 1.25). For all significant covariates of i.m. sotrovimab (body weight, BMI, and sex), the GMRs (90% CI) ranged between 0.59 and 1.21 (0.53–1.38); thus, were fully contained within prespecified relevance bounds of 0.5 to 2.0 (additional details are provided in the Supplementary Material).

FIGURE 3. Forest plots of GMRs (90% CI) of model-estimated dose-normalized C0-168h, after (a) i.v. and (b) i.m. dosing. n is the number of patients in each group, [or] indicates respective endpoint is included in the interval, and (or) indicates respective endpoint is not included in the interval. C168h, concentration at 168 h; CI, confidence interval; GMR, geometric mean ratio; i.m., intramuscular; i.v., intravenous; SARS-CoV-2, severe acute respiratory syndrome coronavirus 2; VL, viral load.

ER modeling

The source data for the ER analyses included 2877 records of 959 patients from COMET-TAIL. After excluding patients who were not in the intent-to-treat population and removing patients lacking PK data, a total of 2706 observations from 902 patients were included in the COMET-TAIL ER efficacy dataset for these analyses. Table S7 summarizes the numbers of patients included in the efficacy analyses by treatment group.

Summaries of demographic characteristics for the ER efficacy analysis population are provided in Table S8. Among the 902 patients, 493 (45.3%) were male, median age was 50 years, and median baseline BMI was 31.0 kg/m². The median duration of symptoms was 4 days, and median log₁₀ baseline SARS-CoV-2 viral load was 6.09 (range 3.2–10.2). An estimated 69.8% of patients had less than or equal to one risk factor and 30.2% had greater than one risk factor. Overall, 2.2% of patients had progression of COVID-19 through day 29 (Table 2), with the highest percentage in the sotrovimab 250 mg i.m. arm (4.0%).

TABLE 2 Summary of progression occurrence rates for the primary endpoint in the COMET-TAIL study, by dose (ER population).

End point
Progression of COVID-19, n (%), [probability]	Sotrovimab (250 mg i.m.)	Sotrovimab (500 mg i.m.)	Sotrovimab (500 mg i.v.)	Overall
No	167 (96.0) [0.960]	353 (97.8) [0.978]	362 (98.6) [0.986]	882 (97.8) [0.978]
Yes	7 (4.0) [0.040]	8 (2.2) [0.020]	5 (1.4) [0.014]	20 (2.2) [0.022]

Abbreviations: COVID-19, coronavirus disease 2019; ER, exposure-response; i.m., intramuscular; i.v., intravenous.

Of the exposure measures evaluated, only sotrovimab AUC_0-day28, C_96h, and C_168h were significant predictors of the probability of progression (p < 0.05). Whereas AUC_0-day28 was a significant predictor, only sotrovimab C_96h and C_168h were selected for further model development, as both were also considered clinically meaningful exposure parameters based on timing of progression (median time from randomization to progression event was 5.5 days; Table S8). Parameter estimates and standard errors from the base ER progression of COVID-19 through day 29 models of sotrovimab C_168h and C_96h are shown in Table S9.

For each treatment arm, the base model was used to predict the progression rates associated with the respective range of exposures. Table S10 shows that for the C_168h model, within each treatment arm the observed progression rate was encompassed by the range of predicted progression rates. Conversely, for the C_96h model, whereas the observed progression rate in the ER dataset fell within the predicted range for the 500 mg i.v. and 500 mg i.m. treatment arms, predicted progression rates fell below the observed point estimate in the 250 mg i.m. treatment arm (Table S11). Therefore, the C_168h model will be the focus of the remainder of the ER results summary.

Covariate analysis led to the addition of the number of risk factors (≤1 vs. >1) as an additive shift on the model intercept (model-estimated placebo response). The impact of the number of risk factors was only on the intercept (placebo progression rate), with no impact on drug response.

Parameter estimates and standard errors from the final ER model (progression of COVID-19 through day 29 model vs. sotrovimab C_168h) are shown in Table 3. All parameters were estimated with good precision (<55% RSE).

TABLE 3 Parameter estimates and standard errors from the final ER model for the occurrence of progression of COVID-19 through day 29 (primary endpoint)—sotrovimab concentrations at 168 h in the COMET-TAIL study.

Parameter	Final parameter estimate
168 h	Population mean	%RSE
Intercept
Overall response (logit) (−)	−4.169	12.03
Additive shift in INT for RISKCATN = 1	1.887	27.34
Slope
Slope for concentration at 168 h (1/[μg/mL])	−0.02037	54.43
Minimum value of the objective function = 170.913

Abbreviations: COVID-19, coronavirus disease 2019; ER, exposure-response; RISKCATN, number of risk factors; %RSE, relative standard error expressed as a percentage. Intercept = model-estimated placebo rate.

Using the ER efficacy analysis dataset in conjunction with the parameter estimates from the final C_168h model, 500 replicates of the analysis dataset were simulated. Figure 4 illustrates the predicted probability of progression of COVID-19 through day 29 (primary endpoint) and 95% CI from the simulated datasets (blue line) overlaid on the observed proportion of patients with progression (red line) versus sotrovimab C_168h. The observed proportion of data fell within the 95% CI of predicted proportions across the range of sotrovimab concentrations, indicating an adequate model fit.

FIGURE 4. Visual predictive check plots for the final ER model for the occurrence of COVID-19 progression through day 29 (primary endpoint) versus sotrovimab concentrations at 168 h. CI, confidence interval; COVID-19, coronavirus disease 2019; ER, exposure-response.

DISCUSSION

Understanding the factors influencing both exposure and response of therapeutic agents can inform dosing recommendations when considered alongside cumulative safety and efficacy data. A key objective of this analysis was to establish a PopPK model for sotrovimab to identify sources of variability in exposures in order to inform on clinical usage and any necessary optimization in special populations. Additionally, individual exposure measures for patients in COMET-TAIL were predicted using the PopPK model and included in the ER analysis dataset to establish an ER relationship for the primary efficacy endpoint (probability of progression of COVID-19) and identify any potential sources of variability in drug response.

Sotrovimab PK was described by a linear, two-compartment model with first-order elimination, and i.m. absorption was characterized by a sigmoid absorption model. A systematic covariate analysis found that body weight was a statistically significant descriptor of the variability in sotrovimab PK and influenced the CL and V3. Other covariate effects retained in the final model were sex on KA, and F_IM and BMI on KA. However, based on the magnitudes of the covariate effects, optimization of dosing on the basis of these factors is not expected. Additional safety and efficacy data could inform the clinical relevance of covariate effect on exposure.

The ER analyses aimed to assess the relationship between sotrovimab serum exposure and clinical response of progression of COVID-19, as defined as hospitalization greater than 24 h or death. ER models for the probability of progression of COVID-19 through day 29 (primary endpoint) using only data from COMET-TAIL were developed. To our knowledge, this is the first published ER model developed to characterize the relationship between the exposure of a therapeutic monoclonal antibody and probability of progression of COVID-19 in high-risk patients with mild to moderate COVID-19. VPC plots showed that the observed proportion of data fell within the 95% CI of predicted proportions across the range of sotrovimab concentrations at 168 h, indicating an adequate model fit. Furthermore, the number of risk factors (≤1 vs. >1) as an additive shift on intercept was the only significant covariate effect. This covariate impacted only the model-estimated placebo response (intercept) but had no impact on overall drug response, suggesting that dose optimization would not be anticipated on the basis of number of risk factors.

Despite the COMET-TAIL model performance, the ER analysis has a number of limitations that may prevent the generalization of these results to describe the overall exposure-progression relationship for sotrovimab in early treatment across SARS-CoV-2 variants, highlighting many challenges currently faced in the development of COVID-19 therapeutics.

1. In the early treatment setting, rates of progression to hospitalization are low. COMET-TAIL included a limited number of progressors (20 progressors and 882 nonprogressors). Additionally, upon hospitalization, PK samples are not routinely collected, further limiting the dataset feeding into ER analysis. Notably, PK data were not collected in several progressors (3/10 in the 250 mg i.m. arm and 2/10 in the 500 mg i.m. arm); therefore, the ER dataset is smaller than the clinical efficacy dataset, resulting in differences in progression rates/treatment arm between the clinical efficacy and ER datasets.
2. Pivotal trials evaluate efficacy only in the context of contemporaneous variants. The COMET-TAIL dataset allowed for assessment of ER for only one predominant variant of concern (VOC). COMET-TAIL recruitment was from June to September 2021, when the predominant circulating strain of SARS-CoV-2 in the United States was the delta variant.²¹ Consistent with variant circulation at the time of study enrollment, the predominant VOC/variant of interest detected in COMET-TAIL participants with available sequencing data was the delta (B.1.617.2) variant (88.2%, 674/764 participants).²² The translatability of ER analyses from delta variant to other variants is unknown.
3. Conduct of placebo-controlled trials is challenging when standard of care exists, introducing challenges for ER modeling. COMET-TAIL evaluated efficacy of a single 500 mg dose of i.m. sotrovimab against a 500 mg i.v. comparator arm and did not include a placebo. Although the model fit was adequate on the basis of VPCs between observed and predicted data in the ER dataset, the model-predicted placebo progression rate was also lower than expected considering the largely unvaccinated high-risk COMET-TAIL study population who were predominantly recruited in the state of Florida, USA (85%), from June to September 2021, when the predominant circulating strain of SARS-CoV-2 in the United States was the delta variant.²¹ Real-world evidence (RWE) data point to an estimated placebo progression rate of 9.1% in unvaccinated individuals in Florida during the delta period (data on file).

Therefore, whereas the ER analysis provided some initial insights into exposure measures influencing response and factors expected to influence placebo progression rates in COMET-TAIL, the dataset may be too limited to directly inform the exposure-progression relationship for sotrovimab in early treatment and dosing recommendations for current or future VOCs. Not only does the limited efficacy dataset limit the model's utility, but the translatability of ER analysis conducted in the context of a single variant to future VOCs requires further investigation across the field. Therefore, the consideration of cumulative evidence from in vitro neutralization, clinical safety, RWE of effectiveness, and PopPK data may be needed to guide future dosing recommendations.

AUTHOR CONTRIBUTIONS

All authors wrote the manuscript. J.E.S., A.E.-Z., J.P., S.R., M.A., A.N., A.S., E.L.A., W.W.Y., E.M., C.G., A.P., A.E.S., and M.R. designed the research. J.E.S., A.E.-Z., J.P., and S.R. performed the research. J.E.S., A.E.-Z., J.P., S.R., and X.L. analyzed the data.

ACKNOWLEDGMENTS

The authors thank Michelle Preston, MSc, and Jeanne McKeon, PhD, of Lumanity Scientific Inc., for medical writing support, which was funded by Vir Biotechnology, Inc. and GSK.

FUNDING INFORMATION

This work was supported by Vir Biotechnology, Inc. and GSK.

CONFLICT OF INTEREST STATEMENT

Jennifer E. Sager, Asma El-Zailik, Melissa Aldinger, Elizabeth L. Alexander, Wendy W. Yeh, Erik Mogalian, Chad Garner, and Maribel Reyes are employees of Vir Biotechnology and report stock ownership in Vir Biotechnology; third-party funding from GSK to Vir Biotechnology for the submitted work; and patent applications planned, pending, or issued on or related to the content of this manuscript. Julie Passarell and Stefan Roepcke are employees of Cognigen Division, Simulations Plus, Inc. and report third-party funding from GSK to Vir Biotechnology for the submitted work; consulting fees were paid to Cognigen by Vir Biotechnology, Inc. for the submitted work. Xiaobin Li, Ahmed Nader, Andrew Skingsley, and Amanda Peppercorn are employees of GSK and report stock ownership in GSK and third-party funding from GSK to Vir Biotechnology for the submitted work. Adrienne E. Shapiro reports acting as a trial investigator for Vir Biotechnology and receiving nonfinancial support from Vir Biotechnology during the conduct of the study.

DATA AVAILABILITY STATEMENT

Individual participant data will not be made publicly available. The authors confirm that the data supporting the findings of this study are available within the article and its Supplementary Materials.

Endnote:

^*A post hoc change was made to the multiple imputation algorithm from daily to weekly imputation due to the bias that was observed in the imputed progression rates from the daily imputation algorithm.