A multivariate morphometric analysis of the glomeruli in the normal and pathologically changed human kidney

Summary Morphometric studies have shown several significant differences in certain features of the kidney of normal individuals, those with minimal changes disease (MC), mesangioproliferative glomerulonephritis (MPGN) or diabetic glomerulosclerosis (DGS). However, there is a considerable overlap. As this could prevent the application of morphometry in diagnostic kidney pathology, we have applied multivariate analysis.In total, material from 89 different patients was studied (13 normals, 30 MC's, 13 MPGN's and 33 DGS patients).A two-step approach has heen used because of the pattern of deviations between the different groups. First, the normals and MC's as one group were distinguished from the MPGN's and DGS's as another. With 6 features 90.5% of all the patients were correctly classified (sensitivity 95.6%, specificity 84.6%).For the distinction between the normals and MC's, three features (mesangial cell percentage, total glomerular cells and endothelial cell percentage), was the best discriminating combination. Using 0.75 as a numerical classification probability threshold (for doubtful or inconclusive) none of the minimal changes were misclassified, and only two of the normal patients (16%). Four of the normals were inconclusive (33%) as were four of the minimal changes (14%). This result should be considered with the initial selection criteria in mind (no observable histological changes after careful subjective evaluation, in the presence of a clinical nephrotic syndrome in the minimal change patients). This emphasizes the possibility of morphometry todetect differences, which escape qualitative observations.An even better discrimination can be obtained between the MPGN's and DGS's. Only one of the MPGN's was misclassified, but in contrast to all the other cases, the numerical classification probability of this patient was low (0.65 in comparison with 0.79 to 1.0).It is concluded that in kidney pathology, multivariate analysis of morphometric data gives a better discrimination between different groups than single variate analysis.