The accuracy of retrospective chart review in measuring nosocomial infection rates. Results of validation studies in pilot hospitals.

To measure the accuracy and consistency of a standardized method--retrospective chart review (RCR)--for estimating nosocomial infection rates (NIRs) in individual hospitals, the authors performed a series of pilot studies in four hospitals of different types. In comparison with a standard based on diagnoses made by physician-epidemiologists supervising intensive prospective data collection teams, the RCR method was found to have an average sensitivity of 0.74 (+/- 0.02 SE; range 0.69-0.78) and an average specificity of 0.964 (+/- 0.002; 0.945-0.991). These values were comparable to those of the physician-epidemiologists' diagnoses and varied less among the hospitals. Two independent teams of chart reviewers were found to have similar levels of sensitivity and specificity, and the reliability of diagnosis at the level of the individual chart reviewer averaged 0.94. In a restudy at one of the pilot hospitals at the midpoint of the actual Medical Records Survey (MRS), there was a substantial increase in sensitivityand a slight increase in specificity as a result of improvements made in the RCR method after the original pilot studies.