Xera OpenScience · Living systematic review

Does the emulation match the trial?

Every published target trial emulation that reports a randomized benchmark, paired with the trial it emulates. 474 comparisons from 129 studies, of which 279 cleared the benchmark gate and enter the pooled models.

Studies
129
2019–2026
Comparisons
474
279 in the pooled analysis
Target trials
94
named randomized benchmarks
Interval coverage
53.4%
46.1%–61.3% (95% CI)
The finding

No systematic bias on average. Wide disagreement case by case.

Pooled ratio of estimates · HR
1.02
95% CrI 0.96 to 1.07

Emulation divided by trial hazard ratio, across 228 comparisons. An interval straddling 1 means no detectable average bias in either direction.

Prediction interval
0.681.52
τ = 0.12 on the log scale

Where the next comparison would be expected to land. The average is reassuring; any single emulation is not.

Interval coverage
53.4%
149 of 279 comparisons

How often the emulation’s 95% interval contains the trial estimate. A well-calibrated method would sit near 95%.

Where they diverge

The widest disagreements

All comparisons
StudyOutcomeTrialEmulationApartIntervals
Baptiste 2024
Cardiology
Cough
vs ONTARGET
0.26 RR1.2912.6 SEno overlap
Garry 2022
Critical care
28-day mortality
vs RECOVERY
0.82 HR1.325.8 SEno overlap
Hu 2024
Endocrinology
Anemia event occurrence or anemia treatment initiation
vs DAPA-CKD
0.39 HR0.805.7 SEno overlap
Suissa 2024
Pulmonology
Exacerbations
vs IMPACT
0.85 HR1.135.3 SEno overlap
Wang 2023
All areas
Moderate or severe COPD exacerbation
vs IMPACT
0.85 HR1.135.3 SEno overlap

Clinical areas

129 studies
Oncology32
Cardiology31
Endocrinology13
Rheumatology7
Orthopedics5
Critical care4
Neurology4
Nephrology3

Effect measures

474 comparisons
HR304 · 228 pooled
RR110 · 34 pooled
MD30 · 15 pooled
RD14 · 2 pooled
OR13 · 0 pooled
SMD3 · 0 pooled

Reported practice

of 129
COI declaration93.0%
Funding disclosure93.0%
Protocol registration16.3%
Data availability10.1%
Code availability17.1%
DAG reported14.7%
QBA performed14.7%

A disclosure count, not a quality score: a data link does not prove the data are reusable.

Most recent

Latest emulations

All 129 studies