We derive semiparametric methods for estimating and testing treatment effects when censored recurrent event data are available over multiple periods. These methods are based on estimating functions motivated by a working “mixed-Poisson” assumption under which conditioning can eliminate...