Network meta-analysis of biological response modifiers in rheumatoid - - PowerPoint PPT Presentation

▶
network meta analysis of biological response modifiers in
SMART_READER_LITE
LIVE PREVIEW

Network meta-analysis of biological response modifiers in rheumatoid - - PowerPoint PPT Presentation

Dec 02, 2022 603 likes •834 views

Network meta-analysis of biological response modifiers in rheumatoid arthritis including multiple outcomes at multiple time points David Jenkins, Reynaldo Martina, Sylwia Bujkiewicz, Pascale Dequen & Keith Abrams Department of Health

slide-1
SLIDE 1

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Network meta-analysis of biological response modifiers in rheumatoid arthritis including multiple outcomes at multiple time points

David Jenkins, Reynaldo Martina, Sylwia Bujkiewicz, Pascale Dequen & Keith Abrams Department of Health Sciences, University of Leicester, U.K.

slide-2
SLIDE 2

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Background

  • GetReal is a three-year project of the Innovative Medicines Initiative

(IMI), a EU public-private consortium consisting of pharmaceutical companies, academia, HTA agencies and regulators patient

  • rganisations
  • GetReal aims to investigate how robust new methods of Real World

Evidence (RWE) collection and synthesis could be adopted earlier in pharmaceutical R&D and the healthcare decision making process

  • A case study in Rheumatoid Arthritis (RA) looking at how to utilise ALL

available evidence in order to produce a framework for maximising the evidence base from multiple sources

slide-3
SLIDE 3

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Methods

  • Systematic review & Network Meta-Analysis (NMA) undertaken

for biologics as monotherapy or in combination with methotrexate (MTX)

  • Binary outcome of interests were ACR50 and DAS28 remission
  • NMA of licenced dose at 6 months
  • All dose NMA at 6 months
  • Bivariate NMA (1) at 6 months and multivariate NMA for each
  • utcome across multiple time points
  • Modelling profile of treatment effect over time using linear and

polynomial models

(1) Achana, F. A., Cooper, N. J., Bujkiewicz, S., Hubbard, S. J., Kendrick, D., Jones, D. R., & Sutton, A. J. (2014). Network meta-analysis of multiple outcome measures accounting for borrowing of information across outcomes. BMC medical research methodology, 14(1), 92.

slide-4
SLIDE 4

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

1st line RCT licenced dose network (ACR50 at 6 months)

Placebo Placebo+DMARDs ADA CTZ ANA RIT ETA GOL ABA TOC INF

slide-5
SLIDE 5

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

All dose network (ACR50 at 6 months)

Placebo Placebo+MTX Abatacept10+MTX Abatacept2+MTX Anakinra+MTX Infliximab3+MTX Infliximab10+MTX Adalimumab20+MTX Adalimumab40+MTX Adalimumab Adalimumab80+MTX Certolizumab200+MTX Certolizumab400+MTX Etanercept25+MTX Etanercept25 Golimumab Golimumab100+MTX Golimumab50+MTX Tocilizumab8+MTX Tocilizumab4+MTX Tocilizumab8 Rituximab500+MTX Rituximab1000+MTX Rituximab+MTX Rituximab

slide-6
SLIDE 6

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu 1 2 3 7 8 10 11 12 13 14 16 17 20 21 22 23 24 25 26 29 30 31 33 37 38 39 41 43 44 45 46 47 48 49

Networks at 3, 6 and 12 months (ACR50)

slide-7
SLIDE 7

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

NMA – Standard

  • Let 𝜀𝑗(𝑐𝑙) represent the study specific log-odds ratio (LOR) of the

treatment in arm k of study i

  • Assuming the treatment effect is normally distributed,

𝑧𝑗𝑙~𝑂𝑝𝑠𝑛𝑏𝑚(𝜄𝑗𝑙, 𝑇2𝑗𝑙) 𝜄𝑗𝑙 = 𝜈𝑗𝑐 𝑗𝑔 𝑙 = 𝑐 𝜈𝑗𝑐 + 𝜀𝑗(𝑐𝑙) 𝑗𝑔 𝑙 ≠ 𝑐 𝑐 = 𝐵, 𝐶, 𝐷 𝜀𝑗(𝑐𝑙)~𝑜𝑝𝑠𝑛𝑏𝑚(𝑒𝑐𝑙 = 𝑒𝐵𝑙 − 𝑒𝐵𝑐, 𝜐2𝑐𝑙)

  • 𝑧𝑗𝑙 is the log odds of remission in arm k of study i
  • 𝜈𝑗𝑐 is the study specific baseline effect
  • 𝜀𝑗(𝑐𝑙) is the study specific log odds ratio for treatment k relative to

treatment b

  • Hence, 𝑒𝑐𝑙is the pooled effect of treatment k relative to treatment b and

𝜐2𝑐𝑙is the between study variance (heterogeneity parameter)

slide-8
SLIDE 8

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Results of ACR50 at 6 months

Treatment Licenced dose NMA All dose NMA LOR LCI UCI LOR LCI UCI Abatacept + MTX 1.09

  • 0.15

2.31 1.08

  • 0.22

2.41 Adalimumab 1.04

  • 1.68

3.83 Adalimumab + MTX 1.24 0.01 2.51 1.24

  • 0.08

2.58 CTZ + MTX 2.27 0.99 3.62 2.30 0.91 3.70

Etanercept 1.70

  • 1.45

4.90

Infliximab + MTX 0.91

  • 0.73

2.54 0.89

  • 0.85

2.64 Placebo

  • 18.68
  • 149.00

71.16

  • 0.82
  • 3.16

1.55 Abatacept

  • 16.80
  • 147.00

72.95 1.05

  • 2.06

4.18 Rituximab + MTX 1.58 0.30 2.86 1.57 0.19 2.93 Tocilizumab 1.22 0.14 2.40 1.75 0.31 3.252

Tocilizumab + MTX 1.61 0.28 2.98 1.72 0.71 2.72

slide-9
SLIDE 9

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Multivariate model

  • Data at many time points are often collected in clinical trials

and more than one outcome is usually reported

  • On average an outcome is reported at two time points in RA
  • Real world evidence can provide longer term follow up
  • This extra evidence that is not normally utilised but may provide

valuable information to decision makers

  • One method to utilise this extended evidence base is to use a

multivariate approach by modelling separate outcomes simultaneously using the correlation to borrow information across;

– Multiple outcomes – Multiple time points

slide-10
SLIDE 10

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Multivariate model (within study)

  • For each arm k of study i let 𝑍

𝑗𝑙𝑛 be the observed log-odds of an event for outcome m

(m=1,….,M) jointly following a multivariate normal distribution, then, 𝑍

𝑗𝑙1

⋮ 𝑍

𝑗𝑙𝑁

~ 𝑂𝑝𝑠𝑛𝑏𝑚 𝜄𝑗𝑙1 ⋮ 𝜄𝑗𝑙𝑁 , 𝑇2𝑗𝑙1 ⋯ 𝑠1𝑁𝑗𝑙𝑇𝑗𝑙1𝑇𝑗𝑙𝑁 ⋮ ⋱ ⋮ ⋮ … 𝑇2𝑗𝑙𝑁

  • The 𝑇2𝑗𝑙 matrix is the associated within-study covariance matrix
  • If 𝑠1𝑁𝑗𝑙 = 0 then the problem reduces to M independent outcomes/NMAs

𝜄𝑗𝑙1 ⋮ 𝜄𝑗𝑙𝑁 = 𝜈𝑗𝑐1 ⋮ 𝜈𝑗𝑐𝑁 𝜈𝑗𝑐1 + 𝜀𝑗 𝑐𝑙 1 ⋮ 𝜈𝑗𝑐𝑁 + 𝜀𝑗 𝑐𝑙 𝑁 𝑗𝑔 𝑙 = 𝑐 𝑗𝑔 𝑙 ≠ 𝑐 𝑔𝑝𝑠 𝑐 = 𝐵, 𝐶, 𝐷, …

slide-11
SLIDE 11

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Multivariate model (between study)

  • Then,

𝜀𝑗 𝑐𝑙 1 ⋮ 𝜀𝑗 𝑐𝑙 𝑁 ~ 𝑂𝑝𝑠𝑛𝑏𝑚 𝑒 𝑐𝑙 1 = 𝑒 𝐵𝑙 1 − 𝑒 𝐵𝑐 1 ⋮ 𝑒 𝑐𝑙 𝑁 = 𝑒 𝐵𝑙 𝑁 − 𝑒 𝐵𝑐 𝑁 , 𝜐2 𝑐𝑙 1 … 𝜍1𝑁

𝑐𝑙𝜐 𝑐𝑙 1𝜐 𝑐𝑙 𝑁

⋮ ⋱ ⋮ ⋮ … 𝜐2 𝑐𝑙 𝑁

  • Where 𝜐2(𝑐𝑙) is the covariance matrix containing terms for the between

study variances (𝜐2 𝑐𝑙 𝑛) with 𝜍𝑛𝑜

𝑐𝑙 being the between-study correlations

between effects measured by outcome m and n (𝑛 ≠ 𝑜) specific to each k versus b comparison

  • Multiple arms and treatments were adjusted for in all models
  • Bivariate case for ACR50 and DAS28
  • Trivariate case using ACR50 at 3,6 and 12 months
slide-12
SLIDE 12

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Multivariate results for ACR50 at 6 months

Treatment Full NMA at 6 months Bivariate ACR50 + DAS28 Trivariate case ACR50 LOR LCI UCI LOR LCI UCI LOR LCI UCI Abatacept + MTX 1.09

  • 0.22

2.41 1.10

  • 0.25

2.45 1.00

  • 0.28

2.20 Adalimumab 1.04

  • 1.69

3.84 0.91

  • 1.94

3.70 1.04

  • 1.66

3.69 Adalimumab + MTX 1.25

  • 0.09

2.59 1.25

  • 0.11

2.61 1.22

  • 0.04

2.42 CTZ + MTX 2.31 0.91 3.71 2.29 0.87 3.70 2.26 0.81 3.63

Etanercept 1.71 -1.45 4.90 1.54 -1.82 4.74 1.81 -1.00 4.73

Infliximab + MTX 0.89

  • 0.85

2.65 0.90

  • 0.88

2.68 0.92

  • 0.72

2.47 Placebo

  • 0.82
  • 3.17

1.56

  • 0.95
  • 3.38

1.42

  • 0.76
  • 2.98

1.54 Abatacept 1.05

  • 2.06

4.19 0.90

  • 2.32

4.03 1.11

  • 1.71

4.08 Rituximab + MTX 1.57 0.20 2.93 1.73 0.26 3.21 1.56 0.36 2.82 Tocilizumab 1.76 0.31 3.25 1.73 0.22 3.22 1.78 0.41 3.22

Tocilizumab + MTX 1.72 0.72 2.72 1.71 0.65 2.75 1.72 0.72 2.64

slide-13
SLIDE 13

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Multivariate results

  • Broadly similar results to the standard NMA models
  • Less extreme results are to be found in the multivariate analysis
  • Reduction in uncertainty around effectiveness estimates

– Larger reduction with stronger correlation between outcomes

  • Borrows information to ‘strengthen’ results
  • Predicts missing values based on correlation

– If there is a missing treatment effect for an outcome, it can be predicted

slide-14
SLIDE 14

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Polynomial models for Outcomes over time

  • Further to the previous models, polynomial models including

multivariate normal distribution allowing for borrowing of strength across time points can be applied (2)

  • 1st (linear) and 2nd order polynomial models were applied for

ACR50 at 3, 6 & 12 months

  • Correlation between ACR50 at multiple time points from the

same study was incorporated using a multivariate approach

  • Deviance Information Criterion (DIC) was used to assess the

‘goodness of fit’ of the models and to choose the final model

(2) Jansen, J. P. (2011). Network meta-analysis of survival data with fractional polynomials. BMC medical research methodology, 11(1), 61.

slide-15
SLIDE 15

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Polynomial model

  • The polynomial model for the log odds at time t for treatment k of study i, is as

follows, 𝑧𝑗𝑙𝑢~𝑂𝑝𝑠𝑛𝑏𝑚 𝜄𝑗𝑙𝑢, 𝑇2𝑗𝑙𝑢 𝜄𝑗𝑙𝑢 = 𝛾0𝑗𝑙 +

𝑛=1 𝑁

𝛾𝑛𝑗𝑙𝑢𝑞𝑛 𝛾0𝑗𝑙 ⋮ 𝛾𝑁𝑗𝑙 = 𝜈0𝑗𝑐 ⋮ 𝜈𝑁𝑗𝑐 𝑗𝑔 𝑙 = b 𝜈0𝑗𝑙 ⋮ 𝜈𝑁𝑗𝑙 + 𝜀0𝑗𝑐𝑙 ⋮ 𝜀𝑁𝑗𝑐𝑙 𝑗𝑔 𝑙 ≠ 𝑐

  • Where 𝜄𝑗𝑙𝑢 reflect the log odds of treatment k at time t for study i
slide-16
SLIDE 16

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Polynomial model

  • The vectors

𝜈0𝑗𝑐 ⋮ 𝜈𝑁𝑗𝑐 and 𝜀0𝑗𝑐𝑙 ⋮ 𝜀𝑁𝑗𝑐𝑙 are trial specific and represent the parameters 𝛾0, 𝛾1, … , 𝛾𝑁 for the ‘baseline’ treatment b and the difference in 𝛾0, 𝛾1, … , 𝛾𝑁 for treatment k relative to b, respectively

  • As in the previous multivariate model, δ then follows a multivariate

normal distribution to account for study correlation

𝜀0𝑗𝑐𝑙 ⋮ 𝜀𝑁𝑗𝑐𝑙 ~ 𝑂𝑝𝑠𝑛𝑏𝑚 𝑒0 𝑐𝑙 = 𝑒0 𝐵𝑙 − 𝑒0 𝐵𝑐 ⋮ 𝑒𝑁 𝑐𝑙 = 𝑒𝑁 𝐵𝑙 − 𝑒𝑁 𝐵𝑐 , 𝜐2

  • Where τ2 is the between study covariance matrix
slide-17
SLIDE 17

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Polynomial results

  • On the left are the results from the 1st order (linear) polynomial model
  • On the right are the results from the 2nd order polynomial model and the

lowest DIC

slide-18
SLIDE 18

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

1st vs 2nd order polynomial of ACR50

  • The blue and red lines represent the credible intervals (dashed)

and mean log odds obtained from the 1st and 2nd order model for tocilizumab + MTX vs MTX, respectively

  • 4
  • 3
  • 2
  • 1

1 2 3 4 5 3 4 5 6 7 8 9 10 11 12 Log odds ratio Time (months)

slide-19
SLIDE 19

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Conclusions

  • Licenced and full NMA

– Routinely used in decision making – Full NMA provides more information and comparisons for decision makers with potentially reduced uncertainty

  • Multivariate model

– Allows more information to be utilised – Can reduce uncertainty by borrowing strength across outcomes – Predicts outcomes when information is missing for various treatments

  • Polynomial models

– Estimates treatment effect profiles over time – Allows for more information to be utilised

slide-20
SLIDE 20

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

Further Work

  • Including Real World Evidence (RWE)

– RWE could be included directly in NMA with appropriate bias adjustment – RWE could be used to inform the correlation structure, i.e. between

  • utcomes, as a prior distribution for correlation parameters in

multivariate NMA models – RWE could be used to extend and support estimation of the treatment profile over time, i.e. having longer follow-up, but appropriate bias adjustment is required.

slide-21
SLIDE 21

The research leading to these results has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement no [115303], resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution. www.imi.europa.eu

THANK YOU FOR LISTENING!