Gravar-mail: Reweighting to address non-participation and missing data bias in a longitudinal electronic health record study