TABLE 2.

Summary of existing methods comparison.

	Accuracy	Privacy protection?	Communication efficient?	Capable of handling rare event?	Collapsibility of effect
Pooled Analysis a	High	NO (patient‐level data are shared)	NO (large amount of patient‐level data are shared)	YES	YES
Meta Analysis b	Varying (Not accurate for rare diseases)	YES	YES	NO	YES
GLORE [44]	High	YES	NO (iterative algorithms)	YES	NO (odds ratio is obtained)
(Robust)‐ODAL, dCLR c [22, 45, 46, 47]	High	YES	YES	YES	NO (odds ratio is obtained)
ODAP‐B	High	YES	YES	YES	YES

Note: Comparisons among pooled analysis, meta‐analysis, distributed algorithms for a logistic regression model, and the proposed method. Accuracy is evaluated through mean squared error (MSE) and bias to the true value: The smaller the MSE or bias is, the better the accuracy is. Privacy is evaluated based on whether the method is an aggregated data‐based approach without sharing patient‐level information. The evaluation of communication is through the number of rounds of transferring aggregated data across sites and the number of digits to be communicated within each round.

^a Pooled analysis: Fitting the modified Poisson regression on the pooled data.

^b Meta‐analysis: Modified Poisson regression is fitted within each site and then meta‐analyses the summary statistics to obtain pooled RR.

^c ODAL: One‐shot distributed algorithm for logistic regression; dCLR: Distributed algorithm for the conditional logistic regression model [22, 45, 46, 47].