Question 1

What is target trial emulation?

Accepted Answer

Target trial emulation is a framework for analyzing observational data as though it were a randomized trial. You specify the trial you would have run — eligibility, treatment strategies, assignment, follow-up, outcome, and analysis plan — and then emulate each component in the observational data, which makes the assumptions explicit and exposes design errors such as immortal time bias.

Question 2

How closely do emulations reproduce their target trials?

Accepted Answer

Across 279 benchmarked comparisons, the pooled hazard ratio of emulation to trial estimate is 1.02 (95% credible interval 0.96 to 1.07): no systematic bias on average. Agreement in individual comparisons is much looser — the emulation confidence interval contains the trial estimate in 53% of comparisons, and the prediction interval for a new comparison spans 0.68 to 1.52.

Question 3

Why are some comparisons excluded from the pooled analysis?

Accepted Answer

Every extracted comparison is shown, but only those whose randomized benchmark could be verified against the trial’s primary report enter the pooled models. Comparisons are held back when the benchmark is unverifiable, when the emulation and trial report incompatible effect scales, or when the reference trial had not reported before the emulation was published. Each excluded comparison states its reason.

Question 4

Is XeraTTE open access?

Accepted Answer

Yes. Every study, comparison and analysis result is browsable and downloadable as CSV or JSON with no registration, and the API is public.

Outcome & target trial	Trial vs emulation	Agreement
Annualized relapse rate vs CONFIRM·RR·Efficacy	0.780.59–1.05 0.790.69–0.90	overlapClose alignment0.08 SE apart
Annualized relapse rate vs OPERA·RR·Efficacy	0.530.43–0.66 0.200.14–0.29	no overlapClose alignment-4.52 SE apart
Annualized relapse rate vs TENERE·RR·Efficacy	1.200.62–2.30 0.950.83–1.07	overlapClose alignment-0.69 SE apart
Annualized relapse rate vs TRANSFORMS·RR·Efficacy	0.480.34–0.70 0.600.53–0.69	overlapClose alignment1.14 SE apart
Annualized relapse rate vs ASSESS·RR·Efficacy	0.590.37–0.95 0.620.53–0.72	not pooledClose alignment
Annualized relapse rate vs BEYOND·RR·Efficacy	1.060.89–1.26 1.000.94–1.07	not pooledClose alignment
Annualized relapse rate vs REGARD·RR·Efficacy	1.030.85–1.25 1.000.93–1.07	not pooledClose alignment
Annualized relapse rate vs RIFUND-MS·RR·Efficacy	0.190.06–0.62 0.250.15–0.57	not pooledClose alignment

Gavoille 2026

Comparisons

Design and methods

Data and sample

Transparency

Provenance