Indexing of medical diagnoses by word affinity method

Automated coding of medical diagnoses is still an unsolved problem. Our goal in recent work was to find efficient, cheap and easy to implement method to assist the work of human encoders in hospitals. The proposed method is based on a vector-space model especially adapted to deal with short expressions, like clinical diagnoses. Using a set of coded diagnoses the co-occurrence of codes and words is more or less characteristic. The method describes these characteristics mathematically, by introduction of the so-called word adhesion. Two human encoders were asked to code the same set of 92 clinical diagnoses. Their results were compared to the ranked list of codes, produced by the computer. The results were better where the two human encoders agreed, and the overall results demonstrate the feasibility of the approach.