This document summarizes a seminar on population-adjusted treatment comparisons using the matching-adjusted indirect comparison (MAIC) and simulated treatment comparison (STC) methods. It discusses how these methods can be used to adjust for imbalances in effect modifiers between trials when indirectly comparing treatments that have not been directly compared head-to-head. The seminar outlines the assumptions of these methods, provides recommendations for their appropriate use, and emphasizes the importance of justifying choices, specifying the target population, and adhering to reporting guidelines.
Similar to Population-adjusted treatment comparisons: estimates based on MAIC (Matching-Adjusted Indirect Comparisons) and STC (Simulated Treatment Comparisons)
Similar to Population-adjusted treatment comparisons: estimates based on MAIC (Matching-Adjusted Indirect Comparisons) and STC (Simulated Treatment Comparisons) (20)
VIP Service Call Girls Sindhi Colony 📳 7877925207 For 18+ VIP Call Girl At Th...
Population-adjusted treatment comparisons: estimates based on MAIC (Matching-Adjusted Indirect Comparisons) and STC (Simulated Treatment Comparisons)
1. York CHE TEEHTA seminar
16th March 2017
1
Population-adjusted treatment comparisons
Estimates based on MAIC and STC
David M. Phillippo,1 A. E. Ades,1 Sofia Dias,1 Stephen Palmer2
Keith R. Abrams,3 Nicky J. Welton1
1 School of Social and Community Medicine, University of Bristol
2 Centre for Health Economics, University of York
3 Department of Health Sciences, University of Leicester
3. York CHE TEEHTA seminar
16th March 2017
3
Outline
• Background
• Standard indirect comparisons
• Population adjustment
• MAIC and STC
• Assumptions and properties
• Recommendations
4. York CHE TEEHTA seminar
16th March 2017
4
Background: Indirect Comparisons
Wish to compare two treatments B and C
• Not studied in the same trial
• Instead, each compared with a common comparator
A through AB and AC trials.
B C
A
5. York CHE TEEHTA seminar
16th March 2017
5
Background: Indirect Comparisons
Standard indirect comparisons:
• 𝑑 𝐵𝐶 = 𝑑 𝐴𝐶 − 𝑑 𝐴𝐵
• Biased if there are imbalances in effect modifiers
(EMs) between AB and AC; 𝑑 𝐴𝐵 𝐴𝐵 ≠ 𝑑 𝐴𝐵 𝐴𝐶
B C
A
𝑑 𝐴𝐵 𝑑 𝐴𝐶
6. York CHE TEEHTA seminar
16th March 2017
6
Background: Effect Modifiers
• Effect modifiers alter the effect of treatment
relative to control
• Prognostic variables affect absolute outcomes
but not the relative treatment effects
7. York CHE TEEHTA seminar
16th March 2017
7
Prognostic Variable
Treatment A Treatment B Treatment A Treatment B
Effect Modifier
8. York CHE TEEHTA seminar
16th March 2017
8
Background: Effect Modifiers
• Effect modifiers may also be prognostic
• We are only concerned with individual level
effect modifiers
• Not possible to adjust for study characteristics
• Effect modifiers must be prespecified
9. York CHE TEEHTA seminar
16th March 2017
9
Background: Population Adjustment
• Standard indirect comparisons assume
constancy of relative effects
• Population adjustment methods seek to adjust
for imbalance in EMs
• Relaxed constancy assumption
• Create a fair comparison in a specific target
population
10. York CHE TEEHTA seminar
16th March 2017
10
Background: Target Population
• A decision has a pre-specified target population
• e.g. UK population for NICE TA
• Methods for population adjustment produce
estimates in a specific target population
• This may not match the target population for the
decision!
11. York CHE TEEHTA seminar
16th March 2017
11
Background
Ideal scenario: full individual patient data (IPD)
• “Gold standard” – IPD meta-regression
B C
A
𝒀𝒊 𝑻𝒊 𝑿 𝟏𝒊 𝑿 𝟐𝒊 ⋯
AB trial: IPD
𝒀𝒊 𝑻𝒊 𝑿 𝟏𝒊 𝑿 𝟐𝒊 ⋯
AC trial: IPD
12. York CHE TEEHTA seminar
16th March 2017
12
Background
Common scenario: limited IPD
• Several recent methods make use of mixed data
B C
A
𝒀𝒊 𝑻𝒊 𝑿 𝟏𝒊 𝑿 𝟐𝒊 ⋯
AB trial: IPD AC trial: aggregate data
ത𝑌A, ഥY 𝐶, ത𝑋1, ത𝑋2, …
𝜎𝐴, 𝜎 𝐶, 𝑓𝑿 ⋅
13. York CHE TEEHTA seminar
16th March 2017
13
Population adjustment: MAIC
Matching-Adjusted Indirect Comparison
• Weight individuals in the AB trial to balance covariate
distributions with the AC trial
• Take a weighted mean to estimate mean outcomes
on A and B in the AC trial
• Similar idea to Propensity Score reweighting
Signorovitch et al. (2010)
14. York CHE TEEHTA seminar
16th March 2017
15
Population adjustment: MAIC
• AB and AC population must have sufficient
overlap
• Compare covariate distributions, inclusion/exclusion
criteria
• Check distribution of weights
• Compute Effective Sample Size (ESS) ≈
σ 𝑤 2
σ 𝑤2
• Traditional Propensity Score “balance checking”
not necessary/possible
15. York CHE TEEHTA seminar
16th March 2017
16
Population adjustment: STC
Simulated Treatment Comparison (STC)
• Create an outcome regression model in the AB trial
• Use this to predict mean outcomes on treatments A
and B in the AC trial
Ishak et al. (2015)
16. York CHE TEEHTA seminar
16th March 2017
18
Population adjustment: STC
• AB and AC population must have sufficient
overlap
• Avoid extrapolation
• Compare covariate distributions, inclusion/exclusion
criteria
• Use standard tools for model checking
• AIC/DIC, examine residuals, …
17. York CHE TEEHTA seminar
16th March 2017
19
Population adjustment: MAIC and STC
• Extensive computation is not required
• Can be implemented in standard statistical
software and routines (SAS, R, STATA, …)
• A worked example in R is available accompanying the
TSD on the NICE DSU website
18. York CHE TEEHTA seminar
16th March 2017
20
Population adjustment
Two possible forms of indirect comparison
B C
A
B C
Anchored Unanchored
19. York CHE TEEHTA seminar
16th March 2017
22
Population adjustment
Two possible forms of indirect comparison
Anchored
B vs. C = (C – A) – (B – A)
Unanchored
B vs. C = C – B
• Comparison is on a given transformed scale
• The latter requires much stronger assumptions, but
doesn’t need a common comparator arm
20. York CHE TEEHTA seminar
16th March 2017
23
Assumptions and properties
Some form of constancy assumption
• Constancy of relative effects
• Relative A vs. B effect constant across studies
• No EMs in imbalance
• Used by standard indirect comparisons
• Conditional constancy of relative effects
• Conditional constancy of absolute effects
21. York CHE TEEHTA seminar
16th March 2017
24
Assumptions and properties
Conditional constancy of relative effects
• Used by anchored population-adjusted indirect
comparisons
• Requires all effect modifiers to be known
• Respects randomisation, prognostic variables are
cancelled out
• A reasonable relaxation of constancy of relative
effects
22. York CHE TEEHTA seminar
16th March 2017
25
Assumptions and properties
Conditional constancy of absolute effects
• Used by unanchored population-adjusted indirect
comparisons
• Requires all effect modifiers and prognostic variables
to be known
• Ignores randomisation
• Widely regarded as infeasible
23. York CHE TEEHTA seminar
16th March 2017
26
Assumptions and properties
Other assumptions:
• Studies are internally valid
• Lack of joint distribution leads to additional
assumptions about correlations between covariates
24. York CHE TEEHTA seminar
16th March 2017
27
Assumptions and properties
Both MAIC and STC produce estimates of relative
treatment effect that are specific to the AC
population
• This is unlikely to be representative of the decision
target population
• If so, population-adjusted estimates are irrelevant for
the decision…
25. York CHE TEEHTA seminar
16th March 2017
28
Shared Effect Modifier Assumption
• Satisfied by active treatments which:
• Have the same set of effect modifiers
• Change in treatment effect for each EM is the same
for all treatments
• Likely to be valid for treatments in the same class
• If valid for treatments B and C then an estimate
of B vs. C is valid for any population
26. York CHE TEEHTA seminar
16th March 2017
29
Recommendations
Motivation of the recommendations
• Reproducibility, consistency, transparency
• Minimising bias and maximising precision
27. York CHE TEEHTA seminar
16th March 2017
30
Recommendations
1. Anchored vs. unanchored
2. Justifying anchored comparisons
3. Justifying unanchored comparisons
4. Variables to adjust for
5. Scale of comparison
6. Target population
7. Reporting guidelines
28. York CHE TEEHTA seminar
16th March 2017
31
Recommendation 1
• Unanchored comparisons require much stronger
assumptions, so anchored comparisons are always
preferred
When connected evidence with a common comparator is available, a population-
adjusted anchored indirect comparison may be considered. Unanchored indirect
comparisons may only be considered in the absence of a connected network of
randomised evidence, or where there are single-arm studies involved.
29. York CHE TEEHTA seminar
16th March 2017
32
Recommendation 2
• Applies to anchored comparisons
• Justification is necessary for moving away from
standard methods
• Effect modification alters the decision scenario
Submissions using population-adjusted analyses in a connected network need to
provide evidence that they are likely to produce less biased estimates of treatment
differences than could be achieved through standard methods.
30. York CHE TEEHTA seminar
16th March 2017
33
NICE Methods Guide
Treatment effect modifiers
5.2.7 Many factors can affect the overall estimate of relative treatment
effects obtained from a systematic review. Some differences between studies
occur by chance, others from differences in the characteristics of patients
(such as age, sex, severity of disease, choice and measurement of outcomes),
care setting, additional routine care and the year of the study. Such potential
treatment effect modifiers should be identified before data analysis, either by
a thorough review of the subject area or discussion with experts in the clinical
discipline.
NICE (2013)
31. York CHE TEEHTA seminar
16th March 2017
34
Recommendation 2 (continued)
a) Evidence must be presented that there are grounds for considering one or
more variables as effect modifiers on the appropriate transformed scale. This
can be empirical evidence, or an argument based on biological plausibility.
b) Quantitative evidence must be presented that population adjustment would
have a material impact on relative effect estimates due to the removal of
substantial bias.
• Combine between-trial difference in EMs with
knowledge of likely strength of interaction
• Judge possible bias in relation to relative treatment
effect, clinical importance
32. York CHE TEEHTA seminar
16th March 2017
35
Recommendation 3
Submissions using population-adjusted analyses in an unconnected network need
to provide evidence that absolute outcomes can be predicted with sufficient
accuracy in relation to the relative treatment effects, and present an estimate of
the likely range of residual systematic error in the “adjusted” unanchored
comparison.
• Applies to unanchored comparisons
• Need to justify that we are doing any better than a
naïve comparison of arms
• Otherwise amount of bias is unknown, likely
substantial, and could exceed size of treatment effect
33. York CHE TEEHTA seminar
16th March 2017
36
Recommendation 4
a) For an anchored indirect comparison, propensity score weighting methods
should adjust for all effect modifiers (in imbalance or not), but no prognostic
variables. Outcome regression methods should adjust for all effect modifiers in
imbalance, and any other prognostic variables and effect modifiers that
improve model fit.
• For anchored comparisons, only adjustment for EMs
is necessary to minimise bias
• Adjusting for other variables may unnecessarily
reduce precision
34. York CHE TEEHTA seminar
16th March 2017
37
Recommendation 4
b) For an unanchored indirect comparison, both propensity score weighting and
outcome regression methods should adjust for all effect modifiers and
prognostic variables, in order to reliably predict absolute outcomes.
• For unanchored comparisons all covariates must be
adjusted for, as predictions of absolute outcomes are
required
35. York CHE TEEHTA seminar
16th March 2017
38
Recommendation 5
Indirect comparisons should be carried out on the linear predictor scale, with the
same link functions that are usually employed for those outcomes.
• Better statistical properties
• Effect modification defined with respect to this scale
• Interpretability
• Biologically and clinically, as well as statistically
• Consistency between appraisals
36. York CHE TEEHTA seminar
16th March 2017
39
Recommendation 6
The target population for any treatment comparison must be explicitly stated, and
population-adjusted estimates of the relative treatment effects must be generated
for this target population.
• If there are effect modifiers, then the target
population is crucial
• An “unbiased” comparison is not good enough for
decision making, must also be in the correct
population
• Can use the shared EM assumption, if justified
37. York CHE TEEHTA seminar
16th March 2017
40
Recommendation 7
• Reporting guidelines available in the TSD
• Largely correspond to recommendations 1-6,
reporting the evidence and justification at each
step
38. York CHE TEEHTA seminar
16th March 2017
41
Thank you
Phillippo, D.M., Ades, A.E., Dias, S., Palmer, S., Abrams, K.R., Welton, N.J. NICE DSU Technical
Support Document 18: Methods for population-adjusted indirect comparisons in submission
to NICE. 2016. Available from http://www.nicedsu.org.uk
The TSD was commissioned and funded by the Decision
Support Unit at the National Institute for Health and Care
Excellence.
Ongoing work is supported by the Medical Research
Council grant number MR/P015298/1.