Every organization has their own way of measuring coding accuracy (and actually some organizations don't do it at all). Rejected claims would not be a good measure because many coding errors do not cause a rejection at all, and most claim rejections have nothing to do with coding but rather are related to patient eligibility, benefits, authorizations, payer policies, etc. So looking at rejections wouldn't tell you anything about coding quality.
There's no set formula, but usually coding accuracy is determine by conducting an audit of a sample of the coder's work, either by a senior coder or quality specialist within the organization or by an outside audit company. Since no two charts are alike and some are harder than others, it's most fair to make it based on the a ratio of the number of error to the total number of codes that were assigned - this is how the organization where I work does it. So for example, if a particular record included 8 diagnosis codes and 2 CPT codes, for a total of 10 codes, and there was one wrong code, you would get credit for all 9 of the right codes and only have one error counted. Some organizations may also weight the errors, so that an error that causes an overpayment or underpayment will be considered more serious than one which is simply a reporting error. In addition, audits should always allow for a rebuttal process. No two coders ever code things exactly the same way, so there needs to be a process for deciding if the coder's choices are actually incorrect or if they were just due to a difference in interpretation.
However, as I said, organizations may do this however they see fit so you'll likely find a lot of different answers to this question. Hope this helps a little though.