Diagnostic and Prognostic Role of alpha-Fetoprotein in Hepatocellular Carcinoma: Both or Neither?

The clinical usefulness of alpha-fetoprotein (AFP) in hepatocellular carcinoma (HCC) management is debatable.To assess, in a large multi-centric survey, diagnostic and prognostic reliability of AFP, predictive factors, and any correlation with the tumor immunophenotype.A total of 1,158 patients with HCC were analyzed with reference to serum AFP levels at diagnosis. We evaluated: HCC grading, histotype, and size; Okuda, tumor-nodes-metastases (TNM), and Child-Pugh scores; liver function, symptoms, presence of metastases or portal thrombosis, etiology, survival, and treatment. In 66 patients with histological diagnosis, the pathologists evaluated p53 overexpression, MIB 1 labeling index, BCL-2 positive cells (index of apoptosis), and CD44 (adhesion molecule) positivity.Patients were divided into three AFP groups: normal (20 ng/mL) [46%], elevated (21-400 ng/mL) [36%], and diagnostic (400 ng/mL) [18%]. Statistical correlations were significant for: weight loss (p= 0.0056), pain (p= 0.0025), Child-Pugh score (p= 0.001), tumor size, Okuda's and TNM stages, metastases, thrombosis, type of treatment (all p0.0001), and female sex (p0.004). AFP correlated with survival overall, in patients untreated, transplanted, or undergoing locoregional treatments; but not in those surgically treated. In the discriminant analysis, the related variables were size, female sex, Child-Pugh score, TNM staging (steps 1-4). When using the receiver operating characteristic curve, the prognostic reliability of AFP was limited with area under the curve of 0.59. Finally, patients with low expression of BCL2 had high AFP levels (p0.05). AFP positively correlated with Edmonson score (p0.0001).The evaluation of this large series of HCC patients allowed us to: confirm the low sensitivity (54%) of AFP in the diagnosis of HCC and its prognostic value, albeit limited, being tumor size, female sex (intriguingly enough), Child-Pugh score, and TNM staging independent predictors.