summarizes the results of the rating. It shows that about 90% of the coding was at least sufficient, i.e., had all of their critical findings represented by codes. lists 5 kinds of errors identified through the review along with their frequency.
In 4 cases, the UMLS in that version did not contain an appropriate concept for the phrase (e.g., “fibrocystic change”, “squamous proliferation”, “lipofibroma”, and also “lipoma”.) Note that missing histo-pathologic concepts were the cause of less than optimal coding in a number of other cases as well but did not come to light in our review to date (since we did not rate the coding quality with more scrutiny, see discussion below.)
In 2 cases, the UMLS lexicon list did not contain a synonym or the coder did not find a synonym that could have been found with word stemming (“ethmoids” for “ethmoid bone” and “content” for “contents”.)
In 5 cases the coder could not handle exceptional situations (e.g., “unknown site, procedure not specified”) or unusual expressions and typos. E.g., “disorder proliferative endometrium” (for the histologic substrate of dysfunctional uterine bleeding) was coded simply as “disease”. A common type of problems is the coding of description of normal findings (e.g., “Benign endocervical elements”) or inconclusive findings (e.g.,”Necro-inflammatory debris and food fragments”.)
In one case “Right upper lobe of lung” (as tissue type) was coded as “Right upper zone pneumonia” because this coding had received a slightly better score than a more appropriate coding. This problem with spurious or sub-optimal codes can be seen more frequently in report-fragments otherwise satisfactorily coded.
In 1/5 of the cases the key concept was coded but the coding was rejected because of an unexpected semantic type. We failed to add “Body Substance” (bdsu) to the set of semantic types acceptable for tissue- type and therefore failed to code “Urine” and “CSF”. In one case we missed “acute cellular rejection” because we do not accept “Organ or tissue function” (ortf) as observations. The reason we rejected this semantic type was that it would have caused the very common reference “See note” to be coded as the concept of “vision”.
1/5 of the errors were due to unexpected text format that the parser was not prepared to handle. This includes interjected notes inside the diagnosis sections, and specimen headers broken into two lines of text.
Two of the defective results were due to missed negation (just 2% of 143 properly detected negations.) In these two cases the negated concept was separated from the negation keyword by a comma (e.g., “no evidence of granuloma, dysplasia or malignancy” coded as negated “granuloma” but affirmed “malignant neoplasm”.)