Investigation of statistical decision rules for sequential hematologic laboratory tests.

Abstract

A statistical data processing program for identifying patients who are likely to have abnormalities on various hematologic laboratory tests is described. The prediction of abnormal levels of serum vitamin B12, serum folate, transferrin saturation, and reticulocyte counts is based on a statistical analysis of the patient's age, sex, and routine blood cell measurements. The program was developed using data from normal-value studies and data from patients who had these laboratory abnormalities. The sensitivity and specificity of the program were evaluated in a controlled prospective study of about 5,000 ambulatory adult patients. The program's predictions also were compared with the laboratory tests requested by the patients' attending physicians. The program was most sensitive for predicting low serum vitamin B12 (74%) and low transferrin saturation (78%). In the prospective evaluation, the predictive values varied from 11% for predicting low serum folate to 65% for predicting low transferrin saturation.