Use of Observa,onal Data to Make Causal Inferences About Treatment Decisions in Mul,ple Sclerosis. Brian Healy, PhD

Transcription

1 Use of Observa,onal Data to Make Causal Inferences About Treatment Decisions in Mul,ple Sclerosis Brian Healy, PhD

2 Disclosures n I receive research support from Merck Serono and Novar,s

3 Outline n Background Treatment op,ons in MS Remaining clinical ques,ons n Approaches used for causal inference with examples Propensity score matching Inverse probability weigh,ng

4 Disease course In the initial stage of the disease, patients experience relapses. In the second phase of the disease, patients experience progressive worsening of the disease. Both clinical and MRI measures of each phase are available Meier DS, Weiner HL, GuNmann CR. Time- series modeling of mul,ple sclerosis disease ac,vity: a promising window on disease progression and repair poten,al? Neurotherapeu,cs. 2007;4(3):485-98

5 Treatment op,ons n Eleven FDA approved treatments are available n These treatments have generally been approved because of a reduc,on in the relapse rate n Several addi,onal treatments are in phase III trials Interferon beta- 1b (Betaseron) 1993 Interferon beta- 1b (Extavia) 2009 Interferon beta- 1a (Avonex) 1996 Fingolimod (Gilenya) 2010 Gla,ramer acetate (Copaxone) 1996 Teriflunomide (Aubagio) 2012 Mitoxantrone (Novantone) 2000 Dimethyl fumarate (Tecfidera) 2013 Interferon beta- 1a (Rebif) 2002 Peginterferon beta- 1a (Plegridy) 2014 Natalizumab (Tysabri) 2006

6 RCT results n Despite the large number of subjects who have par,cipated in trials, RCTs provide direct informa,on about a rela,vely small number of the possible treatment comparisons or treatment decision points Most clinical trials have compared a treatment to placebo so ac,ve treatment comparisons are limited Many possible treatment decision points are not evaluated in clinical trials Trials also provide only indirect informa,on about long- term treatment effects

7 What we know n Treatment with the agents from the previous slide reduce the relapse rate/,me to next relapse All treatments also have an impact on GD+ lesions The treatments are reducing the inflammatory component of the disease n In some case, treatment reduces the,me to sustained disability accumula,on n Some head- to- head treatment comparisons have been completed Several studies have shown limited differences between GA and forms of IFN- β

8 Unanswered ques,ons n Several important ques,ons remain unanswered Treatment comparisons How does the efficacy of the available treatments compare in the absence of direct RCT comparison? What is the impact of the treatment on long- term disability accumula,on? Treatment decision points Should treatment be changed ajer a relapse? n None of these ques,ons will be addressed in a RCT so alterna,ve methods are required

9 Comparison of treatments: Combina,on of trial results n One approach to address the treatment comparisons is to combine the informa,on from the available clinical trials n Two recent papers have combined informa,on across mul,ple trials to assess treatment comparisons not directly available in trials Indirect meta- analysis: Roskell et al. Comparison of fingolimod and first line treatments Network analysis: Zintzaras et al. Comparison of all poten,al treatment comparisons Requires many indirect comparisons n Despite these results, the comparison of treatments outside of a clinical trial selng is desirable

10 Comparison of treatments: Observa,onal data n One key ques,on at present is whether subjects should choose fingolimod or natalizumab Each of these treatments are considered second line treatments No trial has compared these treatments Based on the trial data, natalizumab had a greater difference rela,ve to placebo Natalizumab leads to PML in a small number of subjects who are JC virus posi,ve n Several recent papers have inves,gated this ques,on using causal inference approaches

11 Considera,ons n When using observa,onal data to make inferences about treatment comparisons, several considera,ons are needed Pa,ent selec,on The choice of the subject selec,on can impact the validity and generalizability of the findings All available vs. Specific inclusion/exclusion criteria Sta,s,cal methods Since the two treatment groups are not balanced, adjustment for group differences are required Propensity scores Inverse probability weigh,ng

12 Two studies n Two recent studies have inves,gated this ques,on Braude et al Subject selec,on: All available pa,ents Sta,s,cal approach: Propensity score stra,fica,on Carruthers et al Subject selec,on: Subjects who used JC virus serology for the choice of the treatment Sta,s,cal approach: Mul,variate regression and inverse probability weigh,ng n Each of these studies provides important informa,on about the treatment comparison

13 Treatment decision point: Changing treatment ajer a relapse n In addi,on to the head- to- head treatment comparisons, causal inference approaches applied to observa,onal datasets provide the opportunity to assess treatment decision points MS physicians are faced with many poten,al treatment decision points over the course of the disease Example: Should a pa,ent who is being treated change to a new treatment ajer a single relapse?

14 Framework n Using the causal inference framework, observa,onal data from the CLIMB was selected to mimic the equivalent clinical trial Pa,ent selec,on: Pa,ents who had a relapse ajer ini,a,on of gla,ramer acetate Treatment regimen: Subjects who changed to a new treatment within 180 days were compared to subjects who remained on GA Sta,s,cal approach: Mul,variate Cox model and inverse probability weigh,ng to adjust for differences between groups n Results showed no significant benefit of changing to a new treatment ajer a single relapse (Healy et al, 2010)

15 Comparison of treatments: Long- term follow- up n Several studies have anempted to address the long- term benefit of disease modifying treatment n These studies have two poten,al issues with regards to es,ma,ng the causal effect of a treatment Choice of treatment at the beginning of the study can be confounded Similar to previous examples Handling of changes in treatment More complex because we have,me dependent confounding

16 Approaches n Es,mate difference in treatment groups without accoun,ng for changes ajer treatment ini,a,on Long- term follow- up studies of trials use this approach Simplifies the analysis, but only addresses the causal effect of early vs. late ini,a,on of treatment n Censor subjects at the,me of treatment change and inverse probability weight subjects who remain on treatment This approach has been used less frequently, but it is beginning to be used Es,mates the parameter of most interest, but can suffer because a very small number of subjects remain on a treatment for many years if worsening May work bener if we switch to comparison of treatment regime

17 Conclusions n Observa,onal data provide the only source of informa,on regarding important clinical ques,ons in MS The number of clinical ques,ons is too large to have an RCT to address all poten,al comparisons Many ques,ons cannot be addressed due to challenges in comple,ng the studies n Research for applying these approaches and for combining indirect treatment comparisons from RCTs and observa,onal data analysis are on- going

18 CLIMB Medical Director Tanuja Chitnis, MD Director, Partners MS Center Howard Weiner, MD Director, Clinical Research Operations Bonnie Glanz, PhD Head of Biostatistics Brian Healy, PhD CLIMB Study Staff Sandra Cook, RN Mira Weiner Study Coordinators Grace Little Emily Greeke Immunology/Biomarkers Head of MS Biomarker Group Head of Biorepository & Assay Development Biomarker Development Translational Genomics & Integrative Analyses Roopali Gandhi, PhD Pia Kivisäkk, MD PhD Francisco Quintana, PhD Philip De Jager, MD PhD IT/Database Database Manager Software Architect System Administrator Data Analyst Mariann Polgar-Turcsanyi, MS Adam Polgar Mark Anderson Alicia Chua, MS Neuroimaging Research Rohit Bakshi, MD Charles Guttmann, MD MRI Analyst Svetlana Egorova, MD PhD THANK YOU to our CLIMB patients!

19 References n n n n n Braune S, Lang M, Bergmann A; NTC Study Group. Second line use of Fingolimod is as effec,ve as Natalizumab in a German out- pa,ent RRMS- cohort. J Neurol Dec;260(12): Carruthers RL, Rotstein DL, Healy BC, Chitnis T, Weiner HL, Buckle GJ. An observa,onal comparison of natalizumab vs. fingolimod using JCV serology to determine therapy. Mult Scler Sep;20(10): Healy BC, Glanz BI, Stankiewicz J, Buckle G, Weiner H, Chitnis T. A method for evalua,ng treatment switching criteria in mul,ple sclerosis. Mult Scler Dec;16(12): Roskell NS, Zimovetz EA, Rycroj CE, Eckert BJ, Tyas DA. Annualized relapse rate of first- line treatments for mul,ple sclerosis: a meta- analysis, including indirect comparisons versus fingolimod. Curr Med Res Opin May;28(5): Zintzaras E, Doxani C, Mprotsis T, Schmid CH, Hadjigeorgiou GM. Network analysis of randomized controlled trials in mul,ple sclerosis. Clin Ther Apr;34(4):