A critical evaluation of the clinical diagnosis of anemia.

Three physicians independently assessed whether 180 medical or surgical inpatients were anemic. The assessments were based on inspection of skin, nail beds, and conjunctivae. The observers made a positive diagnosis with significantly different frequency. Overall agreements between pairs of observers were 0.76-0.87. Agreement on negative diagnoses was higher than agreement on positive diagnoses. The overall agreements were adjusted for chance agreement, and kappa values which ranged from 0.23-0.47 were obtained. The overall accuracy of the clinical diagnoses was determined with the blood hemoglobin concentration as the "true" standard. For the three observers, the overall accuracy ranged from 0.78-0.79, the predictive value of a positive diagnosis ranged from 0.51-0.61, the predictive value of a negative diagnosis ranged from 0.81-0.84, the sensitivity ranged from 0.27-0.44, and the specificity ranged from 0.88-0.95. For each observer, the probability of detecting anemia was estimated as a function of the blood hemoglobin concentration. Compared with the "true" diagnosis, the observers were reluctant to call a patient anemic, and even in patients with the lowest blood hemoglobin concentrations, the expected probability of detecting anemia did not exceed 0.75.