Development and Validation of a Prognostic Model to Predict the Length of Survival in Patients With Carcinomas of an Unknown Primary Site

Publisher Website
Google Scholar
Abstract
PURPOSE: To identify clinical and biologic variables with significant impact on survival in patients with carcinomas of an unknown primary site and to develop a simple prognostic model for the selection of patients in prospective clinical trials. PATIENTS AND METHODS: Univariate and multivariate prognostic factor analyses were conducted in a population of 150 unselected patients and led to the construction of two successive classification schemes. An external data set of 116 patients enrolled onto two prospective trials was used for validation. RESULTS: When studying clinical variables only, poor performance status (2 or 3) and presence of liver metastases were retained in the multivariate analysis. The first classification scheme consisted of three subgroups of patients with median survivals of 10.8, 6.0, and 2.4 months, according to the number of adverse prognostic factors. With the introduction of serum lactate dehydrogenase (LDH) levels in a further step, liver metastases were no longer significant. The second classification scheme therefore included poor performance status (relative risk [RR], 2.1) and elevated serum LDH level (RR, 2.1). Good-risk and poor-risk patients were identified, with median survivals of 11.7 months and 3.9 months, respectively (P < .0001). The 1-year survival rates were 45% and 11%, respectively. This second classification scheme was validated in an external data set: the median survival rates of patients assigned to the good-risk group and the poor-risk group were 12 months and 7 months, respectively (P = .0089). The 1-year survival rates were 53% and 23%, respectively. CONCLUSION: A simple prognostic model using performance status and serum LDH levels was developed and validated. It allows the assignment of patients into two subgroups with divergent outcome. Further prospective trials will be designed using this prognostic model.