Our method combines information across multiple surrogates using a statistically efficient seemingly unrelated regression framework. Our method provides a robust alternative to single-surrogate-based bias correction, especially in contexts lacking information on which surrogate is superior.