Laboratory diagnosis of anemia and related diseases using multivariate analysis

To establish a simple computer program for the laboratory diagnosis of anemia and related diseases, multivariate analyses were applied to the results of routine hematological laboratory tests obtained from 48 patients and 51 healthy volunteers. The patients studied were limited to those who had not been treated hematologically by the time of their first visit to our hospital, and their first data obtained in our laboratory were analyzed. Final diagnoses were aplastic anemia (AA) in 21, myelodysplastic syndrome (MDS) in 14, iron deficiency anemia (IDA) in 3, polycytemia vera (PV)in 3, and idiopathic thrombocytopenic purpura (ITP) in 7. Eight parameters, WBC, RBC, Hb, Ht, MCV, MCH, MCHC, and PLT, were transformed to normal distribution and then applied to principal component analysis to evaluate their independence. Very close relationships were observed between Ht and Hb, and between MCV and MCH. One each of these pairs was selected by discriminant analysis and two sets, RBC, MCH, Hb, PLT, and WBC, and RBC, MCV, Ht, PLT, and WBC, were obtained. Two canonical components gave good discrimination of these five diseases and also of normal subjects. When disease prediction was made using this analysis, 37 of 48 patients (77.1%) were predicted correctly, and furthermore, when two disease predictions were allowed, all patients were diagnosed properly. Some overlaps were observed in this two-dimensional coordinate system, especially of AA and MDS, and also with normal subjects. To improve the system further, the additional parameters of age and sex were added to construct a three-dimensional analysis which resulted in much clearer discrimination. The whole procedure described is being developed with subjects who are not taking medication. Subsequently, the general application of this analytical procedure should be limited to only those not on medications. In conclusion, this is in essence a demonstration project; however, this trial of laboratory diagnosis using routine hematological laboratory results appears to be promising. Further extension of the study by increasing numbers of patients and disorders studied, including secondary anemias, will allow the design of diagnostic software for use with personal computers at the sites of primary care.