Semaglutide s.c. Once-Weekly in Type 2 Diabetes: A Population Pharmacokinetic Analysis

The population pharmacokinetic model for semaglutide used data generated from five of the phase III SUSTAIN trials; these were SUSTAIN 1, 2, 3, 6, and SUSTAIN-Japan [13‐15, 19, 20]. The design of these trials is summarized in Table 1.

The trials included in the model were all global trials, with the exception of the Japanese trial. Male and female subjects diagnosed with T2DM were included, with an age ≥ 18 years (Japanese patients ≥ 20 years). For all trials, subjects had a minimum HbA_1c of 7.0%, and there were no restrictions on body weight or body mass index (BMI). With the exception of SUSTAIN 6, inclusion and exclusion criteria were overall similar across all trials. Concomitant oral anti-diabetic drugs (OADs) were allowed for all trials (except in SUSTAIN 1, which was a monotherapy trial). Subjects enrolled in SUSTAIN 6 were also allowed basal or pre-mix insulin. SUSTAIN 6 was a pre-approval cardiovascular and other long-term outcomes trial with an enriched CVD population (aged ≥ 50 years with clinical evidence of CVD or aged ≥ 60 years with cardiovascular risk factors). Subjects with normal or mildly impaired renal function [defined as estimated glomerular filtration rate (eGFR) ≥ 90 ml/min/1.73 m² for normal function or eGFR of 60–89 ml/min/1.73 m² for mildly impaired function] were enrolled in all trials. SUSTAIN 1, SUSTAIN 6, and SUSTAIN-Japan also included subjects with moderate renal impairment (eGFR of 30–59 ml/min/1.73 m²). In SUSTAIN 6, the pharmacokinetic (PK) properties of semaglutide were assessed in subjects with severe impaired renal function (eGFR < 30 ml/min/1.73 m²) as well as subjects with less severe renal impairment or normal renal function. For more information on inclusion and exclusion criteria in the individual trials, see the trial publications [13‐15, 19, 20].

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments and Good Clinical Practice [21, 22]. The trial protocols were approved by independent ethics committees and/or institutional review boards. All subjects provided written informed consent before initiation of any trial-related activities.

Subjects included in the population pharmacokinetic analysis were randomized to a maintenance dose of either 0.5 mg or 1.0 mg semaglutide once weekly. SUSTAIN 3 used only the 1.0 mg maintenance dose for the semaglutide treatment arm (Table 1). In all trials, subjects taking semaglutide followed a dose-escalation regimen with the aim of minimizing gastrointestinal adverse events common to the class of GLP-1RAs [23]. Subjects started with 4 weeks of treatment with 0.25 mg semaglutide once weekly before escalating to 0.5 mg semaglutide once weekly. Subjects randomized to the 1.0 mg semaglutide treatment arm would escalate to the 1.0 mg dose after 4 weeks of 0.5 mg semaglutide, remaining on this dose for the duration of the trial. The subcutaneous injections should be administered on the same day of each week, and subjects providing blood samples to be used in the population pharmacokinetic model were encouraged to inject in the same area of the body throughout the trial.

Blood samples for the PK measurement were drawn at approximately 4, 8, 16, and 30 weeks after the first dose in all trials, and all except SUSTAIN 1 included a sample during week 56. SUSTAIN 6 had additional samples taken during week 2 and at end of treatment (minimum 104 weeks after first dose). There were no restrictions on the timing of the blood sample relative to dosing.

Subjects recorded the date, time, and injection site for the initial dose and the two doses prior to each blood sampling in a diary. This information was transferred, along with the date and time of blood sampling to an electronic case report form at each sampling visit.

Subjects exposed to at least one dose of semaglutide and with at least one valid PK measurement were included in the data set, which included dosing information for all recorded injections, semaglutide concentrations, and covariate values. Records with concentration values missing or below the lower limit of quantification, other dosing deviations, or missing information were excluded from the analysis. Dosing deviations were defined as missing injections or if the last two injections prior to blood sampling were administered less than 120 h apart. Missing information was defined as incomplete diary information for 2 weeks before blood sampling.

The semaglutide plasma concentrations were measured following protein precipitation using a validated liquid chromatography assay followed by a tandem mass spectrometry assay (Celerion Inc. Fehraltorf, Switzerland); see [24] for more details. The lower limit of quantification of the assay used for samples included in this population PK model was 0.729 nmol/l.

A pre-specified full model approach was used for the population PK analysis, including a base model without covariates and a full model with all covariates included [25, 26].

The base model was a one-compartment model with first-order absorption and elimination. This model has been shown to provide an adequate description of the PK of semaglutide (Novo Nordisk, data on file). The model was parameterized for semaglutide in terms of k_a (absorption rate constant), CL/F (apparent clearance), and V/F (apparent volume of distribution). The semaglutide absorption rate constant (k_a) was set to 0.0286 h⁻¹ based on data from clinical pharmacology trials with richly sampled PK profiles (Novo Nordisk, data on file). The model was estimated on un-transformed concentration values, and a proportional error model was used to describe the residual variability. Models were estimated using first-order conditional estimation with interaction (FOCE + I).

The full model was used for estimating the potential effects of individual covariates on semaglutide plasma exposure in terms of clearance. The average semaglutide concentration (C_avg) during the dosing interval was

The area under the curve at steady state (AUC_ss,0–168h) was calculated bywhere ‘dose’ was the relevant maintenance dose for the subject, and CL/F was the individually estimated apparent clearance from the population pharmacokinetic model.

The covariates were included to investigate exposures in relevant sub-populations [2, 27, 28] and the dosing characteristics of semaglutide. Covariates were categorical, with the exception of body weight. Age was categorized into three groups, < 64, 65–74, or ≥ 75 years old at baseline. Renal impairment groups were categorized [by estimated glomerular filtration rate (eGFR)] as normal function (eGFR ≥ 90 ml/min/1.73 m²), mild (eGFR = 60–89 ml/min/1.73 m²), moderate (eGFR = 30–59 ml/min/1.73 m²), or severe (eGFR < 30 ml/min/1.73 m²) renal impairment. Baseline body weight was included as a continuous covariate. For semaglutide, there are two maintenance doses (0.5 and 1.0 mg), which were included as covariates to assess the dose dependency of semaglutide exposure. The injection site (abdomen, thigh or upper arm) was also included as a covariate, whereby the most frequently used injection site for an individual patient was used as the covariate value. Additionally, race and ethnicity were included as covariates.

The reference subject profile was defined as a non-Hispanic or non-Latino, white female < 65 years old, with a body weight of 85 kg (pre-specified, representing the expected approximate median body weight of the study population) with normal renal function and dosed in the abdomen with semaglutide 1.0 mg once weekly.

The model was parameterized as:where CL_typ was the typical semaglutide clearance (CL/F) for the reference subject, and θ was used for covariate effect parameters. Exponents used for categorical covariate relations are indicator variables assigned the value 1 for the actual category and else 0, e.g., 1 for males and 0 for females. Groups that contained < 20 subjects were merged with the largest covariate group. Between-subject variability was assumed to be log-normally distributed and was included as η.

Between-subject variability was estimated for CL/F and V/F in both the base and full PK models to account for the degree of variability that could be explained by inclusion of the covariates. If the dose level was missing in a subject’s dosing diary, it was assumed to be the planned dose.

The software program R (version 2.14.2, R Foundation; Revolution Analytics, Mountain View, CA, USA, version 6) was used for data file processing, explorative data analysis, and plotting. NONMEM (ICON Development Solutions, Ellicott City, MD, USA), version 7.1.2, was used for the population pharmacokinetic analysis. Both of these programs were run as validated server installations. PsN [29, 30] was used for the visual predictive check, and data processing was done with R.

A total of 1683 subjects treated with semaglutide were scheduled for inclusion in the population pharmacokinetic assessment. After data cleaning, the final data set included 1612 subjects with a total of 6781 PK measurements, a mean 4.2 semaglutide concentrations per subject. Subjects were excluded because of missing PK data (19 subjects), semaglutide concentrations below the lower level of quantification (33 subjects), or incomplete or missing dosing history (19 subjects), resulting in an exclusion of 8.3% of the PK samples.

The subjects included in the population pharmacokinetic analysis covered a broad range of baseline characteristics: age ranged from 20 to 86 years, body weight ranged from 39.7 to 198.3 kg, duration of diabetes ranged from 0 to 48.9 years, and HbA_1c ranged from 5.9% to 13.1%. Both sexes were well represented, and subjects covered several races and different ethnicities and were recruited from many countries. Subjects with normal to severely impaired renal function were also included. A summary of the baseline characteristics is presented in Table 2 [additional characteristics are provided in Table S1 in the electronic supplementary material (ESM)].

A one-compartment model with first-order absorption and elimination successfully described the pharmacokinetics of semaglutide. The parameter estimates for the base model can be seen in Table S2 in the ESM. Based on the full population pharmacokinetic model, values for CL/F and V/F in the reference subject profile were estimated to be 0.0478 l/h and 12.2 l, respectively (Table S3 in the ESM).

The full model was robust toward changes in k_a, evaluated using sensitivity analyses with two alternative models with modified k_a values (± 25% of the fixed value). Neither of these models had any relevant differences in terms of exposure compared with the presented model; moreover, the model was robust toward exclusion of data with high residuals, i.e., weighted residuals above 4 and below − 4 (Table S4 in the ESM).

The model was qualified in accordance with regulatory guidelines [31, 32]. The model fits for both the base and full model were acceptable, and there were no critical trends in the conditional weighted residuals vs. either semaglutide concentration or time. The individual clearance and volume of distribution estimates appeared to approximate log-normal distributions. Model evaluation was performed through visual predictive checks using PsN and R. One thousand simulated data sets were generated, with a stratification by dose level. The 2.5th, 50th, and 97.5th percentiles of the experimental data were calculated. Then, the 95% confidence intervals of the 2.5th, 50th, and 97.5th percentiles were computed and displayed graphically together with the observed percentiles. The visual predictive check showed that the full model was able to reproduce the median concentrations of the population and that the simulated 95% confidence intervals were in line with the observed data (see Fig. S2, S3 and S4 in the ESM).

The mean semaglutide plasma concentration (C_avg) for the reference subject profile at steady state was estimated to be 15.8 nmol/l [95% confidence interval (CI) 15.6–16.1] with 0.5 mg semaglutide s.c. once weekly at steady state and about twice that at 29.8 nmol/l (95% CI 29.4–30.2) with 1.0 mg semaglutide at steady state (Table 3).

There was a large overlap in the exposure at the two doses, and exposure was generally similar at the same dose level between trials (Fig. 1). Subjects in the Japanese trial appeared to have slightly higher exposures compared with those in the global trials. As explained below, this was mainly due to lower body weights in Japanese individuals and not the influence of Japanese descent.

The population pharmacokinetic model estimated the effects of the chosen covariates on semaglutide exposure. Covariate effects were considered not to be important for exposure if the 90% CI of the relative exposure was within the 0.8–1.25 standard equivalence range. The estimated effect of each analyzed covariate is presented in Fig. 2. There were no important changes in exposure dependent on sex, age, race, ethnicity, renal function, or injection site used. Of the analyzed covariates, only body weight was considered important for the exposure of semaglutide. Dose-normalized exposure was similar in the two dose groups, indicating dose proportionality (Fig. 2).

The between-subject variability of CL/F in the base model, in terms of the coefficient of variation (CV) %, was 26.6% and was reduced to 12.9% by inclusion of all covariates in the full model. This corresponds to 75.8% of the variability being explained by covariates.

Semaglutide exposure was inversely related to body weight. Compared with the reference subject of 85 kg, a subject weighing 55 kg had on average a 40% increased semaglutide exposure and a 127 kg subject had a 27% lower semaglutide exposure. These weights represent the 5% and 95% percentiles of the subjects in the trials.

The simulated mean semaglutide profile during three dosing intervals after s.c. dosing at steady state semaglutide 0.5 mg and 1.0 mg in subjects with T2DM is shown in Fig. 3. The baseline body weights of 70 kg and 100 kg represent the 25% and 75% weight quartiles in the studied population. The profiles were relatively flat, supporting low variability over time as well as low variability between subjects. There also appeared to be an overlap in predicted exposure between the two weight groups at the same dose.

The observed inverse relationship between exposure and body weight over the entire body weight range is shown in Fig. 4 and was similar between males and females (Fig. 4a). Body weight differences may also explain the apparent difference in exposure values in the Japanese trials mentioned above; at the same body weight, the same exposure was seen in Asian (Japanese or non-Japanese) and non-Asian subjects (Fig. 4b).

Exposure of semaglutide was constant over the 2 years data were collected and appeared to be time-independent (Fig. S1 in the ESM).

The dosing recommendations given to subjects during the semaglutide phase III program for delayed or missed doses were tested using this model. Subjects were recommended to take a missed semaglutide dose as soon as possible within 5 days of the planned dose; a dose delayed more than 5 days should be skipped before resuming the planned dosing schedule. The full population pharmacokinetic model was used to generate semaglutide concentration profiles in a reference subject to predict exposure following missed or delayed dosing. If a dose was missed, the simulated profile showed that a 48% decrease in minimum concentration was expected before the next planned dose (Fig. 5a). A 5-day delayed dose should cause minimum semaglutide concentrations to be 37% lower and maximum concentrations to be 14% higher compared with the same subject at normal weekly steady state (Fig. 5b). In both cases, semaglutide concentrations will be close to regular steady-state concentrations after 3 weeks.