Title: "A new frequentist implementation of the Daniels and Hughes bivariate meta-analysis model for surrogate outcomes"

Abstract: Surrogate outcomes are used when the primary endpoint is difficult to measure accurately. Determining if an outcome is suitable to use as a surrogate is a challenging task and a variety of meta-analysis models have been proposed for this purpose. The Daniels and Hughes bivariate model for surrogate outcomes is gaining traction but presents difficulties for frequentist estimation and hitherto only Bayesian solutions have been available. This is because it is non-linear and the number of unknown parameters increases at the same rate as the number of studies. This second property raises immediate concerns that the maximum likelihood estimator of the between-study variance may be downwardly biased. We derive maximum likelihood estimating equations to motivate a bias-adjusted estimator of this parameter. The bias-correction terms in our proposed estimating equation are easily computed and have an intuitively appealing algebraic form. From simulation studies and empirical examples, we conclude that our new estimation method enables satisfactory maximum likelihood-based estimation of the Daniels and Hughes bivariate model.

Talk given on Tuesday 21st November as part of the BSU seminar series - https://www.mrc-bsu.cam.ac.uk/news-an...