Application of Bayesian Meta-Analysis in Synthesizing Health State Utility Values for QALY Estimation in Health Economic Evaluations: A Case Study on Advanced Neuroendocrine Tumors
Author(s)
Mohd Kashif Siddiqui, MBA, MPH, PharmD, Shawn Fernandes, PharmD, Raja Rajeeswari C, MSc Biostats, Jatin Gupta, MPharm, MBA, Vignesh Babu S, MSc Biostats;
EBM Health Consultants, New Delhi, India
EBM Health Consultants, New Delhi, India
OBJECTIVES: Health state utility values (HSUVs) are crucial inputs in economic models. Multiple HSUVs from different studies often exist and a single study may not represent the best available evidence to inform policy decisions. We utilized a Bayesian meta-analysis (BMA) framework in synthesizing HSUVs aiming to provide unified estimates for both pre-progression (PFS) and post-progression (PPS) in patients with neuroendocrine tumors (NET).
METHODS: An existing systematic literature review was utilized to identify HSUVs and its measure of variance for PFS and PPS. A BMA was conducted using ‘brms’ package. Missing variances were imputed using ‘mice’ package. We used flat priors on the mean HSUVs and a half-Cauchy prior on between-study-variance. Adequacy of priors was evaluated using ‘pp-check’ function. Model convergence was assessed through ‘rhat’, a value greater than 1.1 indicating non-convergence. Sensitivity analyses were conducted using complete case and the frequentist framework.
RESULTS: Of the identified studies, 11 reported HSUVs for PFS and six for PPS. One study did not report the measure of variance. The EuroQol-5D (EQ-5D) was the most common instrument, though dimensions (3L or 5L) were often unspecified. Pooled HSUVs from BMA were 0.74 (95% credible interval [CrI]: 0.68-0.78) for PFS and 0.70 (95%CrI: 0.60-0.79) for PPS, with no notable heterogeneity (tau <0.5 for both health states). Sensitivity analyses showed comparable results for the complete case analysis (PFS: 0.77; PPS: 0.73) and the frequentist analysis (PFS: 0.78; PPS: 0.72).
CONCLUSIONS: This study provided a methodological framework to generate unified HSUV estimates for health economic evaluations. BMA accounts for confounding in the estimates and offers flexibility in adding external information from real-world studies, reducing the need of multiple sensitivity analyses in estimating the quality-adjusted life years. This approach can be valuable for policy makers by providing consistent and robust inputs for decision-making.
METHODS: An existing systematic literature review was utilized to identify HSUVs and its measure of variance for PFS and PPS. A BMA was conducted using ‘brms’ package. Missing variances were imputed using ‘mice’ package. We used flat priors on the mean HSUVs and a half-Cauchy prior on between-study-variance. Adequacy of priors was evaluated using ‘pp-check’ function. Model convergence was assessed through ‘rhat’, a value greater than 1.1 indicating non-convergence. Sensitivity analyses were conducted using complete case and the frequentist framework.
RESULTS: Of the identified studies, 11 reported HSUVs for PFS and six for PPS. One study did not report the measure of variance. The EuroQol-5D (EQ-5D) was the most common instrument, though dimensions (3L or 5L) were often unspecified. Pooled HSUVs from BMA were 0.74 (95% credible interval [CrI]: 0.68-0.78) for PFS and 0.70 (95%CrI: 0.60-0.79) for PPS, with no notable heterogeneity (tau <0.5 for both health states). Sensitivity analyses showed comparable results for the complete case analysis (PFS: 0.77; PPS: 0.73) and the frequentist analysis (PFS: 0.78; PPS: 0.72).
CONCLUSIONS: This study provided a methodological framework to generate unified HSUV estimates for health economic evaluations. BMA accounts for confounding in the estimates and offers flexibility in adding external information from real-world studies, reducing the need of multiple sensitivity analyses in estimating the quality-adjusted life years. This approach can be valuable for policy makers by providing consistent and robust inputs for decision-making.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
P59
Topic
Methodological & Statistical Research
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Oncology