Serum proteins profile as an indicator of malignancy: multivariate logistic regression and ROC analyses.

The electrophoretic pattern of serum proteins of 85 patients carrying different types of neoplasia and 85 matched healthy adults were comparatively studied by agarose gel electrophoresis, to find out if there is a specific protein pattern common to different types of cancer. Each protein fraction was analyzed quantitatively by densitometry. Multivariate logistic regression and receiver operating characteristic (ROC) analyses were performed using SPSS software. When total protein and albumin were measured by colorimetric methods, cancer patients, compared to controls, had a decreased concentration of total protein (66.0+/-11.5 g/L vs. 76.4+/-6.8 g/L, p< or =0.0001) and of albumin (39.0+/-8.1 g/L vs. 46.0+/-4.3 g/L, p< or =0.0001). The electrophoretic data of serum proteins showed that the ratio of albumin to globulin (0.92+/-0.30 vs. 1.21+/-0.16, p< or =0.0001), percent of the fractions albumin (46.7+/-8.5% vs. 54.4+/-3.5%, p< or =0.0001) and beta-globulin (11.6+/-4.4% vs. 13.0+/-1.9%, p< or =0.001) were decreased and alpha(1)- (5.3+/-2.5% vs. 2.9+/-0.8%, p< or =0.0001), alpha(2)- (13.5+/-4.8% vs. 11.3+/-2.1%, p< or =0.0001) and gamma-globulins (23.0+/-7.7% vs. 18.3+/-3.1%, p< or =0.0001) were significantly increased in cancer patients relative to controls. Cancer patients also had higher counts of leukocytes (7.98+/-3.11, x10(9) cells/L vs. 6.33+/-1.68 x10(9) cells/L, p< or =0.0001) and erythrocyte sedimentation rate (35.9+/-23.5 mm/h vs. 14.1+/-9.5 mm/h,p< / =0.0001). On the basis of univariate analysis, a protein profile out of the normal ranges was more prevalent in cancer patients than in controls. Analysis of the data using multiple logistic regression indicated that the prevalence of cancer was strongly associated with the serum proteins' profile, and alpha(1)-globulin, erythrocyte sedimentation rate, total protein and the ratio of albumin to globulin were the best parameters to discriminate between malignant and healthy states. The area under the ROC curves were the same for most components of the serum proteins' profile at about 0.75+/-0.09, p< or =0.001. We conclude that the profile of serum proteins indicates high diagnostic values for discriminating between cancer patients and healthy individuals and may be useful as an adjunct diagnosis for detection of malignancy.