Informing Trial Design Decisions Through Virtual Clinical Trials

Assessing the Impact of Relaxing hs-CRP Inclusion Criteria on Patient Enrollment and Efficacy Signal in Rheumatoid Arthritis Trials

Yoni Sidi¹, Xiaomei Liao¹, Anna Fishbein¹, Zhaoling Meng¹

¹ Sanofi

Disclaimer

The views and opinions expressed by the speaker are their own and should not be attributed to their employer.

Background

• hs-CRP (high-sensitivity) C-reactive protein (CRP) is a commonly performed laboratory test used to predict the clinical course and progression of structural changes in psoriatic arthritis (PsA) and rheumatoid arthritis (RA).
• It is included in the most common efficacy outcomes for PsA and RA recommended by the FDA and EMA.
• Influence treat-to-target algorithms and third-party payer criteria for reimbursing treatment costs.

3

Motivation

• Given the importance of hs-CRP as a biomarker of disease severity, it is a commonly used screening criteria for inclusion in RA clinical trials with a wide range of cut off values, most commonly 3-8 mg/L.
• Higher cutoff thresholds can make recruitment more challenging but may also make treatment responses easier to detect.
• However, there remains a gap in knowledge regarding the optimal hs-CRP cutoff for inclusion in RA clinical trials.

4

Goal of the Analysis

Evaluate the sensitivity of the efficacy effect (American College of Rheumatology 20 (ACR20) response rate endpoint or Disease Activity Score in 28 joints using CRP level (DAS28-CRP)) with relation to the inclusion criteria of the biomarker baseline high-sensitivity C-reactive protein (hs-CRP) to support enrollment criteria.

5

Decision making with the �Clinical Modeling and Simulation Workflow

6

TransCelerate: Non-profit organization that created DataCelerate, a global data-sharing platform for reusing historical clinical data in drug development. DataCelerate includes de-identified historical placebo and standard of care data from past clinical trials in SDTM format.

​

Top Contributing Sponsors: Roche, Eli Lilly, AbbVie, AstraZeneca, GlaxoSmithKline, Boehringer Ingelheim, Pfizer, Johnson & Johnson, Novartis, Novo Nordisk, Sanofi, Amgen, UCB Pharma, Bristol-Myers Squibb, Astellas, EMD Serono, …

Data

The data that was available to generate assumptions about the endpoint and the biomarker

• Rheumatoid arthritis publications
• 15 Transcelerate studies were evaluated to characterize hs-CRP at both screening and baseline and their relationship with ACR20.
• Summary statistics of historical studies were analyzed to evaluate the relationship between ACR20, DAS28-CRP, subgroups of hs-CRP and treatment effect.

7

Modeling & Simulation Steps

8

Endpoint Modeling vs. Baseline hs-CRP

Two endpoints were evaluated to measure the effect of baseline hs-CRP on efficacy:

• Binary: ACR20
• Continuous: DAS28-CRP

Initial modeling found that there was only evidence of an interaction with baseline hs-CRP and treatment in ACR20.

Based on these finding simulations were carried out with ACR20 as the efficacy endpoint.

9

CHARACTERIZING HSCRP SCREENING POPULATIONS

Screening hs-CRP Distribution

11

Figure 2: Screening hsCRP LogNormal Distributions

HS-CRP SCREENING AND BASELINE

Characterizing the Difference from Screening and Baseline

13

Figure 3: Boxplot of difference of hsCRP Screening to Baseline.

Transformation from Screening to Baseline

• The relationship between hs-CRP screening values to baseline values is modelled.
• It was observed that the variance of the baseline measurements increases as a function of the corresponding screening measurement.
• AIC Goodness of Fit sigEmax < LinLog

14

Figure 4: Model fits of hsCRP screening to baseline

Comparison to Literature

• Generated baseline hs-CRP value ranges were confirmed by the literature.
• The corresponding interquartile ranges of these distributions align with the values found for DISCOVER1 and DISCOVER2 in Houttekiet et al. (2022).

15

Modeling ACR20 Response

16

The Event ACR20 response is modeled as a binary endpoint assuming it follows a Bernoulli distribution

ACR20 Assumptions

ACR20 response model effects were adjusted to align the overall treatment effects with hypothetical protocol assumptions

Mean Response Rate

• PBO: Placebo 35% (Protocol)
• TE1: Active 60% (Protocol)
• TE2: Active 55% (Reduced Effect)

Established ACR20 model is used to simulate ACR20 response with two interaction scenarios

1. Observed Interaction between treatment arms and baseline hs-CRP (Observed competitor publication)
2. Hypothetical 10% increase in observed interaction term.

17

Complete Workflow

18

Simulations of 32 scenarios were carried out based on the following design:

1. Simulate hs-CRP at screening (x2 SDs)
2. Apply hs-CRP inclusion criteria (x4: >3, >4, >6, >8)
3. Transform screening hs-CRP to baseline hs-CRP
4. Simulate a study (1000 per scenario)
5. Apply model assumptions (x4)

6. Simulate each patient’s endpoints based on the established model, treatment and baseline hs-CRP value
7. Run the statistical test for each simulated study. Summarize over 1000 simulation per scenario

RESULTS

ACR20 response rate by Treatment Arm

• Distributions of simulated response rates by treatment arm.
• Simulation of placebo (red) and base scenario treatment (blue) are consistent with the protocol assumptions (reference lines).

20

Figure 5: Simulation of ACR20 Response Rate

Hypothesis Testing

• Power of superiority test is consistent with the protocol assumptions
• Assuming estimated interaction level the power of the test is not sensitive to screening population
• Assuming an elevated interaction level the power of the test increases as the screening criteria increases.

21

Figure 6: Power of Significant Difference between Proportions of Treatment vs Placebo

Hypothesis Testing

• The screening population assumptions do effect the probability of success both for the mean and the standard deviation assumptions.
• A lower treatment effect than assumed in the protocol will increase the probability of not passing the Go/NoGo threshold.

22

Figure 7: Binomial confidence intervals Placebo Adjusted ACR20 response rate >= Go/NoGo Threshold

Simulations Conclusions

23

Conclusions

24

The hs-CRP modeling pipeline is indication agnostic allowing the Tukey hyper-parameter to be retuned to new indications and endpoints

The modeling and simulation workflow was applied to design, communicate and deploy analysis.

A use case was presented to provide insight on the question of protocol update on the inclusion criteria of the biomarker hs-CRP for Rheumatoid Arthritis clinical efficacy endpoint ACR20.

This workflow promotes consistent communication with stakeholders on the different stages of the simulation, while keeping the goals and impact in the direct line of sight.

Based on the simulation analysis presented it is recommended that lowering the hs-CRP inclusion criteria will not decrease the power of the planned superiority test or reduce the probability of success with relation to the Go/NoGo criteria.

Thank you

REFERENCES

Bibliography

Houttekiet, Charlotte, Kurt de Vlam, Barbara Neerinckx, and Rik Lories. 2022. “Systematic Review of the Use of CRP in Clinical Trials for Psoriatic Arthritis: A Concern for Clinical Practice?” RMD Open 8 (1): e001756.