A predictive algorithm for evaluating elevated serum prolactin in patients with a sellar mass

Hyperprolactinemia occurs in patients with a prolactinoma and in those with a sellar mass compressing the pituitary stalk. Distinguishing these two diagnostic possibilities guides treatment with dopamine agonist therapy or surgical resection. We aimed to identify a simple, predictive algorithm to aid in the diagnosis of prolactinoma in patients with an elevated serum prolactin and a sellar mass. A case-control analysis of pathologically confirmed prolactinomas and non-endocrine secreting controls from the University of California, San Francisco was performed. From 2001 to 2011, this resulted in 177 patients with prolactinomas and 87 controls. Univariate and classification and regression tree (CART) analysis determined the significance of demographic variables, patient symptoms, laboratory values, and radiographic findings in distinguishing pathology. Additionally, a subset of patients with mildly elevated serum prolactin (25–125 ng/ml) was independently analyzed. Prolactinomas had a mean pre-operative prolactin of 858 ng/ml versus 17.57 ng/ml in controls (p < 0.01). One hundred and two (62.6%) of the prolactinomas were macroadenomas (size >10 mm) compared to 74 (92.5%) of the controls (p < 0.01). CART analysis identified preoperative prolactin (>41.5 ng/ml), age (<40.5 years), and size (<17 mm) as being predictive of prolactinoma with a misclassification rate of 7.9% (21/264). Similar analysis on the subset of patients with mildly elevated serum prolactin (<125 ng/ml) identified size (<2.5 cm) and pre-operative prolactin (>40 ng/ml) as key variables. These two factors correctly predicted 98.6% (69/70) of cases. Our model correctly classifies most patients with elevated serum prolactin and identifies those patients most amenable to surgical treatment.