In many applications in comparative effectiveness research and patient-centered outcomes research (PCOR), records of patient characteristics and outcomes are dispersed over multiple files. Analysis that links 2 or more separate data sources is increasingly important as researchers seek to integrate administrative and clinical datasets while adapting to privacy regulations that limit access to unique identifiers. Significant efforts have been invested in researching processes for linking 2 data sources; however, less attention has been given to analyzing these linked sources while accounting for possible false-positive errors (ie, linked records that are declared as links but are not) and false-negative errors (ie, linked records that are not declared as links and should be). In addition, no guidelines have been established to estimate causal effects from linked observational data.
