FIGURE 1.

Schematic illustration of the proposed ODAP‐B method. Each site calculates the initial estimate ${\widehat{\mathit{\theta }}}_k$ and variance ${\widehat{\sigma }}_k^2$ in the initialization step. Then, the meta‐estimate $\bar{\mathit{\theta }}$ is obtained and transferred to all sites for the gradient calculation. Each site calculates the aggregated data (i.e., first and second gradients) with the initial value and local data, and then transfers the gradients back to the lead site or local site for the construction of the surrogate likelihood function.