OBSERVER VARIATION IN THE CLASSIFICATION OF INFORMATION FROM MEDICAL RECORDS

The medical records of patients are among the most detailed existing accounts of clinical experience and should provide a unique source of data for research on the clinical course of disease. To assess the suitability of the record as a source of data for such studies we have examined the reproducibility and validity of the information recorded in the patients clinical history. Reproducibility was assessed by a study of observer variation, in which four physicians independently read the same 110 records of patients with Hodgkin's disease. Two physicians used explicit criteria to identify the clinical events of first symptom, duration of symptoms, and systemic symptoms and the remaining two observers worked without criteria. Levels of agreement between the observers who used criteria to identify these events were high, and the associated Kappa scores varied from 0.75 to 0.93, whereas the Kappa scores for the observers who did not use criteria varied from 0.29 to 0.59. When the data recorded in the clinical history was used to classify patients according to previously described prognostically distinctive groups, similar prognostic gradients were created in two patient populations, suggesting that the recorded data corresponded to true biologic differences and indicating its validity. © 1979.