Comparing the risk of important health outcomes under various, possibly counterfactual, treatments, interventions, or exposure plans provides critical information to guide decision making, but such exposures are often mismeasured. Multiple imputation is an appealing approach to account for exposure misclassification because it may be used with almost any data analysis approach and draws on familiar missing data methods. However, existing work describing multiple imputation for exposure measurement error is limited to settings with internal validation data. Here, we describe a reparameterized form of imputation via misclassification weighting (m-weighting) that may be implemented using prior knowledge about sensitivity and specificity, external validation data, or internal validation data that is not drawn from the main study at random. We evaluated the finite sample properties (bias, coverage, and mean squared error of the hazard ratio and risk difference) of the m-weighting approach using simulations and illustrated the approach to estimate 2-year risk differences comparing time to antiretroviral therapy initiation between fisher folk and non-fisher folk in East Africa.