Abstract
To study the potential of multivariate classification methods in order to obtain more insight into abnormal laboratory data from patients with sickle cell disease, we investigated standard haematological and clinical chemical variables of 18 controls and 37 apparently healthy persons with heterozygous sickle cell disease (Hb AS), all women, using both univariate and multivariate classification methods. In the univariate method, those with Hb AS showed decreased serum log aspartate aminotransferase (log AST) activity, mean corpuscular volume and mean corpuscular haemoglobin (MCH) and increased sodium concentration. The multivariate method identified sodium, potassium, urea, uric acid, log AST, alanine aminotransferase and MCH as the variables that produced maximal separation between persons with Hb AS and controls. It increased the ‘non-error rate’ for classification of persons with Hb AS by 16 · 4% compared with classification based on the variable, MCH, that produced maximal separation by the univariate method.
The frequency distribution of percentage Hb S in the Hb AS group proved bimodal with maximal separation at 37 · 0% Hb S. The subgroup with 37 · 0% or less (n = 16) was considered to have concomitant heterozygous α-thalassaemia-2. In the univariate method the subgroup characterized by > 37 · 0% Hb S (n = 21) had increased serum sodium and uric acid concentrations, perhaps related to sickle cell nephropathy, whereas the subgroup with ≤ 37% Hb S did not. The multivariate method added information to the univariate method by additionally identifying abnormalities in serum potassium and urea concentrations in the former subgroup. The differences in variables related to renal function may result from a decreasing tendency of erythrocytes to sickle with decreasing percentage of Hb S, and show that even apparently healthy persons with Hb AS are heterogeneous with respect to the development of abnormalities demonstrable by clinical chemistry.