Vedolizumab Exposure Levels and Clinical Outcomes in Ulcerative Colitis

Determining the Potential for Dose Optimisation

Mark T. Osterman; Maria Rosario; Karen Lasch; Morris Barocas; Jayson D. Wilbur; Nathanael L. Dirks; Marc R. Gastonguay


Aliment Pharmacol Ther. 2019;49(4):408-418. 

In This Article


Patient Population

In the induction phase of GEMINI 1, patients received either double-blind or open-label vedolizumab 300 mg at weeks 0 and 2. Patients with a clinical response at week 6 were re-randomised 1:1:1 to receive vedolizumab 300 mg every 8 weeks, vedolizumab 300 mg every 4 weeks or placebo, and followed to week 52 (Figure S1).[13] Patients without a response at week 6 received vedolizumab 300 mg every 4 weeks throughout the maintenance phase. Patients treated with placebo during induction continued to receive placebo during maintenance. Ethical guidelines have been previously published for GEMINI 1 ( number, NCT00783718).[12]

Supporting Figure 1.

GEMINI 1 study design.
*Vedolizumab and placebo were administered on Days 1 and 15 of the induction phase. Patients who discontinued vedolizumab or placebo during the induction phase were assigned to the respective non-intent-to-treat groups for the maintenance phase. Randomisation was stratified by 1) concomitant use of corticosteroids, and 2) concomitant use of immunosuppressants or prior history of anti-TNFα therapy. §A reduction in Mayo Score of ≥3 points and a ≥30% decrease from baseline, with a decrease of ≥1 point on the rectal bleeding sub-score or an absolute rectal bleeding score ≤1. Randomisation was stratified by: 1) cohort, 2) concomitant use of corticosteroids, and 3) concomitant use of immunosuppressive agents or prior anti-TNFα therapy.
PBO, placebo; Q4W, dosing every 4 weeks; Q8W, dosing every 8 weeks; TNFα, tumour necrosis factor alpha; VDZ, vedolizumab

Outcome Measures

In GEMINI 1, clinical assessments were performed using the complete Mayo Score at weeks 0, 6 and 52 and the 9-point partial Mayo Score (eg, complete Mayo Score without the endoscopy sub-score) at all other time points. To examine the vedolizumab exposure-response relationship in the present analysis, two outcomes were used: (a) clinical response, defined as a reduction in complete or partial Mayo Score of ≥3 points and ≥30% from baseline, as well as a decrease of ≥1 point on the rectal bleeding sub-score or an absolute rectal bleeding sub-score ≤1 and (b) clinical remission, defined as a complete or partial Mayo Score of ≤2 points with no individual sub-score >1. To examine the association between early vedolizumab concentrations at weeks 2, 4 and 6 and clinical outcomes at weeks 14 and 52, clinical remission was selected as the sole outcome measure, given that it is a more rigorous endpoint, and was defined for both time points as a partial Mayo Score of ≤2 points with no individual sub-score >1.

Vedolizumab Concentration Measurement

In GEMINI 1, vedolizumab concentrations in serum samples were measured using a direct capture, pharmacokinetic, enzyme-linked immunosorbent assay (ELISA; sandwich), with a lower limit of detection of 0.125 μg/mL. The time points at which these concentrations were drawn were just prior to vedolizumab infusions at weeks 2 and 6 and at the week 4 study visit.

Covariates and Estimation of Individual Vedolizumab Exposures

A number of variables that can potentially affect the clearance of vedolizumab have been identified in a prior population pharmacokinetic analysis and included patient age, weight, history of anti-TNFα treatment, serum albumin concentration and faecal calprotectin concentration.[19] In that analysis, which included approximately 20 000 serum samples from 2000 patients, vedolizumab clearance and serum concentrations were estimated based on multiple vedolizumab studies, including a phase 1 healthy volunteer study, a phase 2 study in UC and the large phase 3 GEMINI 1 and 2 randomised controlled trials (GEMINI 2 showed that vedolizumab was effective as induction and maintenance therapy for active CD).[12,19,25,26] These data led to the development of a full covariate model capable of characterising vedolizumab pharmacokinetics and pharmacodynamics more completely.[19] In the current analysis, the relationship between patient-specific covariates and estimated vedolizumab clearance and vedolizumab concentrations were expressed as P values for the monotonic trend analysis. P values for trends were also explored using the exact Cochran-Armitage trend test.

Propensity-score Case-matching Adjustment of the Exposure-response Relationship

The GEMINI 1 trial lacked a prospectively randomised dose-ranging design, thus creating the potential for an imbalance in observed and unobserved patient-specific covariates that could confound causal inference in the vedolizumab exposure-response relationship. Candidate confounding covariates in the observed data were identified based on the prior multi-study population pharmacokinetic analysis mentioned above.[19] Imbalance in the observed covariates across exposure quartiles was evaluated graphically and with quartile-specific data summaries.

To characterise the vedolizumab exposure-response relationship, a propensity-score-based case-matching analysis was performed, adjusting for potential imbalance in observed covariates (patient age, weight, history of anti-TNFα treatment, serum albumin concentration and faecal calprotectin concentration) across the exposure range. For each vedolizumab exposure quartile at both week 6 and steady state, a logistic propensity-score model was fitted to data from the vedolizumab-treated patients and all control subjects (ie, those receiving placebo), using all measured covariates as predictors. The propensity score was used to match treated patients with controls to establish a reference point for response at zero drug exposure (eg, absence of vedolizumab). A robust estimate of the standard deviation (SD) of the propensity-score distribution was then obtained based on the median absolute deviation of the fitted propensity scores. For each patient in the exposure quartile, a match was randomly selected with replacement from the subjects in the control arm with propensity scores within a calliper of 0.2 times the robust estimate of the SD obtained in the previous step. Treated patients without a matched control were excluded from the outcome analysis. The matching step was repeated 1000 times for each candidate match, and the absolute standardised difference in means (ASDM) was calculated for all covariate main effects and two-way interactions. The optimal subset of matched controls was identified as the candidate match with the lowest maximum ASDM among interaction effects that satisfied an ASDM <0.2 for all main effects. This method resulted in an exposure-response data subset that was balanced across observed covariates. The extent of remaining imbalance across unobserved factors was not possible to assess.

The exposure-response analysis was conducted using the full data set and the case-matched subset. Specifically, the rates of clinical response and remission and the distribution of the odds ratios of clinical response and remission were calculated for each quartile of estimated vedolizumab clearance at week 6 and estimated vedolizumab concentrations at week 6 and steady state (trough, during maintenance). Trends in the distribution of the odds ratios of clinical response and remission with increasing estimated vedolizumab clearance or estimated concentration quartiles were examined to determine the robustness of the exposure-response relationship.

For this analysis, a clinically meaningful target response was defined as a reduction in partial Mayo Score of at least 3 units. The quartiles of clearance and exposure associated with this response magnitude were identified. Given the quartile-based boundary on clearance, the FDA label-specified dosing regimen with vedolizumab maintenance at 300 mg every 8 weeks, and the prior population pharmacokinetic model, potential target vedolizumab concentrations associated with a clinically meaningful response were proposed at therapeutically important time points: week 6 (during induction), week 14 (end of induction period) and steady state (representing trough, during maintenance after day 128 [ie, after five 25.5-day linear elimination half-lives of vedolizumab]).

To determine whether early vedolizumab concentrations at weeks 2, 4 and 6 were associated with clinical remission at week 14, the analysis cohort was restricted to GEMINI 1 patients in the intention-to-treat population who were randomised to receive maintenance vedolizumab 300 mg every 8 weeks, as this is the FDA-approved dose and reflects current clinical practice. Two separate analyses were conducted. First, vedolizumab concentrations at weeks 2, 4 and 6 were stratified by clinical remission status at week 14. Second, clinical remission status at week 14 was stratified by vedolizumab concentration quartiles at weeks 2, 4 and 6. Patients were not case-matched to maximise the number of samples for analysis. The nonparametric Wilcoxon rank-sum test was used to compare vedolizumab concentrations at weeks 2, 4 and 6 stratified by remitters vs nonremitters at week 14. Trends in clinical remission status at week 14 or week 52 stratified by vedolizumab serum concentration quartiles at weeks 2, 4 and 6 were explored using the exact Cochran-Armitage trend test.

Compared with week 6, vedolizumab concentrations at weeks 2 and 4 did not have as consistent an association with rates of clinical remission at week 14 (see Section 2.1). The association between clinical remission at week 52 and early vedolizumab concentrations was therefore restricted to week 6 concentrations and was calculated by vedolizumab concentration quartile. To increase the sample size for this analysis, the entire GEMINI 1 cohort12 was used, including subjects who received placebo and open-label induction vedolizumab, and patients were not case-matched.

Of note, clinical outcomes with respect to the presence of anti-vedolizumab antibodies were not assessed in this analysis, as only 3.7% of GEMINI 1 patients were antibody positive at any time and only 1.0% were persistently positive (ie, at ≥2 consecutive visits).12 Thus, the sample size of the antibody-positive population was too small to perform any meaningful analysis.