Resources
evidence-data· 2 June 2026· 12 min read
Indirect treatment comparisons without the hand-waving
When MAIC, STC, and ITC actually disagree — and how to choose the one your reviewer will accept.
Author
Chief Innovation Officer
Pick the method, then defend it
The reviewer does not punish you for picking MAIC over STC. They punish you for not knowing why.
When the methods diverge
In a recent oncology ITC we ran four methods against the same IPD. The hazard-ratio estimates ranged from 0.61 (anchored MAIC) to 0.78 (unanchored STC). The reviewer will ask why.
Hazard ratio by ITC method (lower = stronger comparator effect)
| Method | When to default to it | Reviewer red flag |
|---|---|---|
| Bucher | Single common comparator, well-balanced populations | Population imbalance > 0.1 SMD on any prognostic |
| Anchored MAIC | IPD available, anchor exists, modest imbalance | ESS drops below 50% of original |
| Unanchored MAIC | No common comparator | Treat as hypothesis-generating only |
| STC | Outcome model is well specified, anchor exists | Pre-specified model not registered before analysis |
The minimum R sanity check
# Pre-MAIC balance check
library(MAIC)
bal <- estimate_weights(
intervention_data = ipd,
matching_vars = c("age","ecog","prior_lines"),
comparator_pop = aggregate_summary
)
summary(bal$ess) # flag if < 50% of NESS attrition
If your effective sample size drops below half after weighting, the reviewer will treat the comparison as exploratory regardless of how favourable the HR looks.
Reader poll
Which ITC method do you default to?
Sources
- NICE DSU TSD 18: Methods for population-adjusted indirect comparisons· NICE DSU· 2024
- Phillippo et al. MAIC and STC primer· Medical Decision Making· 2018
ITCs with the reviewer already in the room
Knowledgeable runs the balance checks, ESS attrition, and method rationale alongside the analysis — so the defence is written before the analyst stops typing.
See the ITC workspace