Using Propensity Weighting Methods With Real-World Data and Noncontrolled Extension Data From a Clinical Trial to Estimate Long-Term Treatment Effect: An Example in Alzheimer’s Disease
Author(s)
Marie-Ange Paget, PhD1, Can Mert, PhD2, Antje Tockhorn, PhD1, Ben Rothwell, MSc3, Lars Lau Raket, PhD1.
1Eli Lilly and Company, Indianapolis, IN, USA, 2HaaPACS, Schriesheim, Germany, 3Eli Lilly and Company, Basingstoke, United Kingdom.
1Eli Lilly and Company, Indianapolis, IN, USA, 2HaaPACS, Schriesheim, Germany, 3Eli Lilly and Company, Basingstoke, United Kingdom.
OBJECTIVES: Randomized controlled trials rarely provide long-term treatment effect data in time for health technology assessments. Therefore, assumptions need to be made on long-term treatment effect to inform cost-effectiveness model. To support this, we estimated a three-year treatment effect for donanemab in the United Kingdom (UK) using real-world data and non-controlled trial extension data.
METHODS: In line with the UK donanemab label, only individual patient level data from AD biomarker positive participants with cognitive impairment from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and TRAILBLAZER-ALZ 2 (TB2) extension were analysed. Propensity score weights were applied to the selected ADNI patients to ensure balance in key demographic and baseline characteristics between the ADNI cohort and the TB2 placebo group. Firstly, a mixed model for repeated measures (MMRM) analysed change from baseline (CfB) to Month 18 in Clinical Dementia Rating - Sum of Boxes (CDR-SB) score in the TB2 placebo group and weighted ADNI cohort. Subsequently, a second MMRM analysed CfB to Month 36 in CDR-SB score in the TB2 donanemab group and weighted ADNI cohort to estimate the long-term treatment effect.
RESULTS: Of the 2,430 participants in ADNI, 422 participants who align with the UK label conditions and had non-missing covariates were weighted. The first MMRM demonstrated the suitability of the weighted ADNI cohort to represent the pattern of disease progression of the TB2 placebo group beyond Month 18. The estimated treatment difference (95% confidence interval) for donanemab versus the weighted ADNI cohort was statistically significant at Month 36.
CONCLUSIONS: Our analyses indicate that donanemab-treated patients continue to progress more slowly than untreated patients over 36 months, and provide further evidence for modelling long-term treatment effect in cost-effectiveness models.
METHODS: In line with the UK donanemab label, only individual patient level data from AD biomarker positive participants with cognitive impairment from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and TRAILBLAZER-ALZ 2 (TB2) extension were analysed. Propensity score weights were applied to the selected ADNI patients to ensure balance in key demographic and baseline characteristics between the ADNI cohort and the TB2 placebo group. Firstly, a mixed model for repeated measures (MMRM) analysed change from baseline (CfB) to Month 18 in Clinical Dementia Rating - Sum of Boxes (CDR-SB) score in the TB2 placebo group and weighted ADNI cohort. Subsequently, a second MMRM analysed CfB to Month 36 in CDR-SB score in the TB2 donanemab group and weighted ADNI cohort to estimate the long-term treatment effect.
RESULTS: Of the 2,430 participants in ADNI, 422 participants who align with the UK label conditions and had non-missing covariates were weighted. The first MMRM demonstrated the suitability of the weighted ADNI cohort to represent the pattern of disease progression of the TB2 placebo group beyond Month 18. The estimated treatment difference (95% confidence interval) for donanemab versus the weighted ADNI cohort was statistically significant at Month 36.
CONCLUSIONS: Our analyses indicate that donanemab-treated patients continue to progress more slowly than untreated patients over 36 months, and provide further evidence for modelling long-term treatment effect in cost-effectiveness models.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MSR217
Topic
Methodological & Statistical Research
Disease
Neurological Disorders