ESTIMAND DRIFT IN ONCOLOGY TARGET TRIAL EMULATIONS: LESSONS LEARNED FROM FOUR REAL-WORLD CASE STUDIES
Author(s)
Kai Ringwald, Dr, Carolin Lennartz, Dr, Sebastian Zavala Hoffmann, MSc., Nina Haug, Dr, Melanie Frank, Dipl. Math. Oec..
iOMEDICO, Freiburg im Breisgau, Germany.
iOMEDICO, Freiburg im Breisgau, Germany.
OBJECTIVES: The estimand framework, in combination with target trial emulation (TTE), is a widely recognized standard for reporting analyses that aim to estimate causal effects from non-randomized trials. In practice, the intended and the one that is actually feasibly quantifiable rarely coincide. We aim to draw on our experience from our former TTEs and propose reporting strategies to address this gap.
METHODS: We conducted a cross-case review of four oncology TTEs from prospective German registries, spanning static regimen comparisons (first-line CDK4/6 inhibitors in metastatic breast cancer; immunotherapy combinations vs tyrosine kinase inhibitor monotherapy in renal cell carcinoma), a dose-based dynamic strategy (reduced vs full CDK4/6 inhibitor starting dose), and a sequential strategy (first- to second-line therapy in pancreatic cancer). For each case we compared the estimand implied by the original target-trial protocol with the estimand that the data and the registry design ultimately permitted.
RESULTS: Across all four cases, operationalization shifted the population, endpoint, or intercurrent-event handling relative to the original intention, yet none of the publications displayed the initially intended estimand. The degree of deviation, however, varied markedly: static treatment comparisons required only minor adjustments, whereas the dose-based strategy underwent substantial revision. No established convention exists for documenting this revision, but the gap between the intended and the realized estimand is itself informative and should be reported. We therefore propose to report both 'before' and 'after' estimand tables, with the extent of attribute-level change serving as a qualitative signal of data fitness and emulation robustness.
CONCLUSIONS: TTE and the estimand framework are increasingly adopted. However, estimand drift remains rarely documented, although it is crucial for identifying potential barriers in data fitness and trial design. Reporting intended and realized estimands side by side should become standard practice, turning a hidden limitation into a transparent measure of evidence quality.
METHODS: We conducted a cross-case review of four oncology TTEs from prospective German registries, spanning static regimen comparisons (first-line CDK4/6 inhibitors in metastatic breast cancer; immunotherapy combinations vs tyrosine kinase inhibitor monotherapy in renal cell carcinoma), a dose-based dynamic strategy (reduced vs full CDK4/6 inhibitor starting dose), and a sequential strategy (first- to second-line therapy in pancreatic cancer). For each case we compared the estimand implied by the original target-trial protocol with the estimand that the data and the registry design ultimately permitted.
RESULTS: Across all four cases, operationalization shifted the population, endpoint, or intercurrent-event handling relative to the original intention, yet none of the publications displayed the initially intended estimand. The degree of deviation, however, varied markedly: static treatment comparisons required only minor adjustments, whereas the dose-based strategy underwent substantial revision. No established convention exists for documenting this revision, but the gap between the intended and the realized estimand is itself informative and should be reported. We therefore propose to report both 'before' and 'after' estimand tables, with the extent of attribute-level change serving as a qualitative signal of data fitness and emulation robustness.
CONCLUSIONS: TTE and the estimand framework are increasingly adopted. However, estimand drift remains rarely documented, although it is crucial for identifying potential barriers in data fitness and trial design. Reporting intended and realized estimands side by side should become standard practice, turning a hidden limitation into a transparent measure of evidence quality.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR87
Topic
Methodological & Statistical Research, Organizational Practices, Study Approaches
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
Oncology