Data quality in computerized patient records

Publisher Website
Google Scholar
Abstract
This paper addresses the problem of data quality in electronic patient records using a computerized haematology biopsy report system as an example. Physicians extracted five parameters from a traditional free text cytology report and encoded these parameters thus producing a computer processable report. The parameters were 1) the organ biopsied, 2) quality of specimen, 3) cytological diagnosis including 4) a modifier code for the main diagnosis code (i.e. status post chemotherapy, Y-code) and 5) an additional key describing the degree of remission obtained after chemotherapy of acute leukemias. From the various steps involved in generating the electronic record we selected two critical ones: - encoding of free text terms by physician staff - entering of the coded terms into a computer by lab staff. We analyzed the rates of correct, incorrect and missing codes for each of the five parameters. - encoding of free text terms by physician staff - entering of the coded terms into a computer by lab staff. Our findings indicate that in this model of an electronic patient record there is significant inaccuracy of physicians during the process of encoding the free text report with error rates between 3.2 and 28% and omission rates up to 64%. lab staff entering these coded data into the computer introduce additional errors (0–7.8%) but rarely miss correctly encoded data (0–0.9%). introducing a revised coding system data quality improved significantly (p≤0.001) with a fivefold increase of correct and a 75% reduction of missing codes. the clinical relevance of the diagnoses encoded as perceived by clinicians is a significant factor affecting error and omission rates. a significant source of error is the machine/user interface resulting in incorrect data entry of up to 91.7%. when considering all combined errors and omissions in coding and in entering these codes into a computer only 43% of all electronic reports examined were correct and complete. We conclude that the quality of data in electronic patient records can be quite low. Amongst significant factors contributing to low electronic data quality are the type of user interacting with the system, the kind of function the user performs, relevance of codable terms as perceived by physicians, the kind of coding system used and the quality of the human/machine-interface.

This publication has 4 references indexed in Scilit: