Longitudinal Concurrent and Predictive Biomarkers of Clinical Disease Severity in Systemic Lupus Erythematosus in BLISS-52/BLISS-76

Proposal
1691

Title of Proposed Research

Lead Researcher

Austin Huang

Affiliation

Pfizer

Funding Source

No external funding other than FTE effort.

Potential Conflicts of Interest

Has company shares as part of Pfizer's standard compensation package.

Data Sharing Agreement Date

22 May 2017

Lay Summary

The proposed study looks at the relationship between laboratory observations (e.g. blood measurements) and clinical symptoms of disease severity (e.g. joint pain) in Lupus patients.
Previously, studies have focused on comparing these relationships at the beginning and the end of the study. By comparison, we will examine how these measurements change over time in the study as the disease gets better or worse. The goal of this research is to use these findings to help with designing future clinical trials in Lupus.

Study Data Provided

[{ "PostingID": 1416, "Title": "GSK-HGS1006-C1056", "Description": "A Phase 3, Multi-Center, Randomized, Double-Blind, Placebo-Controlled, 76-Week Study to Evaluate the Efficacy and Safety of Belimumab (HGS1006, LymphoStat-B™), a Fully Human Monoclonal Anti-BLyS Antibody, in Subjects with Systemic Lupus Erythematosus (SLE)" },{ "PostingID": 1417, "Title": "GSK-HGS1006-C1057", "Description": "A Phase 3, Multi-Center, Randomized, Double-Blind, Placebo-Controlled, 52-Wk Study to Evaluate the Efficacy and Safety of Belimumab (HGS1006, LymphoStat-B™), a Fully Human Monoclonal Anti-BLyS Antibody, in Subjects With Systemic Lupus Erythematosus (SLE)" }]

Statistical Analysis Plan

The questions to be addressed and subsequent planned analyses are:1. What are the concomitant markers of disease severity?Marginal associations will be assessed using Bayes shrinkage estimates of marginal correlation between individual molecular markers and components and combined disease score. For this purpose, longitudinal measurements will be treated as repeated measurements for a subject.The disease score will be modelled as:disease_score[t,s] ~ marker[t,s,m] + covariates + errorHere t is the time point, s is the subject ID, m is the index for a biomarker identifier (for Anti-dsDNA). Covariates are based on available demographic information. Individual and combined components of the SLEDAI and BILAG score will be used as a measure of disease severity.Covariates include the demographic and clinical information specified in A.5.Beyond this basic cross-sectional, univariate data perspective, canonical correlation analyses (CCA) will be used to obtain a joint model of the multivariate relationship between molecular markers and disease components.A canonical correlation analysis relates two multivariate descriptors of subjects:X = (x1, x2, x3,…) : biomarker featuresY = (y1, y2, y3, …) : clinical featuresand infers a low-dimensional set of linear projections of features aX, bY such that the correlation between low-dimensional representations of biomarkers and clinical features are maximized.Although treated subjects are included in the analysis in order to sample a wider variety of disease states, treatment efficacy and response prediction will not be assessed in the analysis.Since we are interested in markers that are not specific to a particular treatment, biomarker discovery will focus on placebo data with treatment used to verify whether markers are robust to Belumimab treatment. Identification of markers specific to the context of Belumimab treatment are not the focus of the analysis.2. What are the prognostic markers of disease severity?Prognostic markers are markers in which there is a time lag between the molecular marker and disease severity. This analysis is similar to aim 1 except that there is a time lag between the biomarker measurement and the disease severity measurement allowing assessment of predictive value of the markers. Evaluating prognostic markers will take place in two ways: disease_score[t,s] = marker[t-i,s,m] + covariates + errorAs in analysis (1), canonical correlation analysis will be used to obtain a joint model of the multivariate relationship between molecular markers and disease components, but with molecular biomarkers assessed at an earlier time point.Finally, links between prognostic markers and subsequent disease severity will be explored via longitudinal models linking the predictors and disease severity identified in part 1 and quantifying the relationships and variability over time. Models will be developed within R using available Bayesian modelling packages (e.g. R2OpenBUGS, Rstan). Post processing will also be conducted within R using core packages and ggplot2, dplyr and other R packages as needed.

Publication Citation

Research Proposal 1691 Summary of Results
Pre-competitive Analysis of Markers of Systemic Lupus Erythematosus (SLE) to aid design of Phase II studies using patient level Phase III Clinical Study Data
Satyaprakash Nayak, Austin Huang, Mark Peterson, Maria Dilleen
Affiliation: Pfizer Inc.,
Objectives: Systemic Lupus Erythematosus (SLE) is a complex, auto-immune disease with an unmet medical need and only one approved therapy (Belimumab) in the past half-century. The objective of this analysis was to analyze Phase III patient level data to enhance our understanding of biomarker response and its predictive relationship to treatment efficacy in an effort to provide ways of gaining inference and leading to design of smaller, shorter and more efficient Phase 2 studies. The main objective was to characterize the relationship between biomarkers from the BLISS Phase 3 studies and co-occurring and subsequent changes in SLE clinical disease severity. The second objective was to identify prognostic markers of disease severity within patient disease trajectories. While other studies have examined these data for baseline predictors of response, this analysis examines markers at multiple intermediate time points as potential predictors of disease progression.
Methods: Patient-level data from two Belimumab Phase III studies (BLISS-52 and BLISS-76) were requested through the GSK SAS Clinical Study Data Request portal which allows a precompetitive data access for analysis. There were 1,122 patients starting on Belimumab and 562 patients starting on Placebo for up to 76 weeks. We examined relationships using univariate marginal correlations as well as multivariate analyses using R software, which modelled biomarker predictors of clinical response in a longitudinal model. Clinical response was assessed using the SLE Response Index [SRI(4)]. The univariate and multivariate relationships between lab (biomarker) and clinical endpoints on clinical response, were quantified and graphically depicted for comparisons. Biomarkers examined included anti-double stranded DNA (Anti-dsDNA), antinuclear antibody (ANA), primary compliments C3 and C4, lymphocytes, and CPX (cardiopulmonary stress test). Clinical predictors were examined using the systematic lupus erythematosus disease activity index SLEDAI-2K.
Results: Consistent with prior findings and our initial hypotheses, SLE response to treatment as measured by SLE Response Index [SRI(4)] (both active and placebo) was seen to be correlated most strongly with baseline disease severity as measured by baseline SLEDAI-2K. The population distribution of disease severity equilibrates between 160-200 days. Multivariate clustering using individual discrete time assessments over a timespan up to 200 days of SLE disease activity index (SLEDAI-2K), indicated that the future disease status is highly dependent on current disease status. With respect to biomarkers - anti-dsDNA and complement levels (C3 and C4) were found to be highly correlated with the disease activity SLEDAI-2K, however even for these biomarkers, the SLEDAI-2K changes were found to be most highly correlated with baseline levels of these markers, and thus under-score the limitations of the activity index to study disease progression and treatment efficacy.
Conclusions: We present a mechanism to access and analyze rich, subject-level data for understanding SLE and designing clinical studies. Our analysis demonstrates an explanatory hypothesis that the common inclusion criteria in the clinical trials tend to enrich for transiently-flaring subjects, and thus a need for longer than 6 months study as the disease equilibrates at around 200 days. However, it should also be noted that patients in a clinical trial receive better care and monitoring when they enroll in a study. Caution should be applied in extrapolating short term trial outcomes to long term disease activity. We also observed a positive correlation between baseline SLEDAI-2K and placebo response rate as measured by SRI.