One coder was used for the entire coding process, but validity
was established at the onset with two readers coding a sample of
the articles. In this stage, inter-coder reliability was established by
sampling 20 percent of the database and coding them until both
coders reached agreement on the claims 95 percent of the time. The
remaining articles were coded by a single coder unless determination
was too difficult, at which time both coders consulted each
other. Such cases were determined via coder negotiation and
consensus.