Evaluating cancer patients for metastases: an approach using Bayes' theorem

Before patients with apparently localized cancer are offered potentially curative surgery or radiation therapy, physicians often order a battery of imaging studies to exclude metastases. The yield of radionuclide liver and bone scans and computed tomography of the brain in breast, colorectal, and lung carcinomas was examined as follows: the literature was reviewed to obtain the autopsy prevalences of brain, liver and bone metastases in breast, colorectal and all histologies of lung carcinoma, to estimate the sensitivities and specificities of radionuclide liver and bone scans and computed tomography of the brain. Using Bayes' theorem, the predictive accuracies of positive and negative clinical evaluations and imaging studies were computed. For most of the examples analyzed, the imaging studies do not provide information beyond that obtained by the history, physical examination, and inexpensive laboratory tests and radiographs that would alter the physician's decision to offer curative treatment directed towards local disease. This method is presented as a model for clinical decision making where the information gained by an additional study is evaluated for its use in altering therapeutic decisions.