Refining the assessment of the sensitivity and specificity of diagnostic tests, with applications to prostate cancer screening and non-small cell lung cancer staging.

1 United States of America, National Cancer Institute, Division of Cancer Prevention, Biometry Research Group, Bethesda, Maryland, United States, and University of Maryland at Baltimore County, Baltimore County, Maryland, United States. Send correspondence to: Vance W. Berger, Biometry Research Group, National Cancer Institute, Executive Plaza North, Suite 3131, 6130 Executive Boulevard, MSC 7354, Bethesda, Maryland 20892-7354, United States of America; telephone: (301) 435-5303; fax: (301) 402-0816; e-mail: vb78c@ nih.gov 2 The Johns Hopkins University, Baltimore, Maryland, United States of America. Diagnostic tests are often used to detect or stage a disease as well as to determine a course of subsequent treatment. For example, elevated prostatespecific antigen (PSA) levels or abnormal digital rectal examination (DRE) findings are indicators of the potential for prostate cancer (1), and they are often followed by biopsy. Likewise, computed tomography (CT) and positron-emission tomography (PET) can be used for staging non-small cell lung cancer, that is, to detect mediastinal lymph node metastases (2). Conventional statistical analysis of the detection properties of any given diagnostic test is based on a 2x2 table that cross-classifies subjects by presence or absence of cancer (according to some gold standard, often biopsy) and a positive or negative test result. Yet not all tumors are the same. Size, stage, severity, or some other factors may render some cancers, but not others, potentially lethal. Also, some cancers, though not yet symptomatic, may have metastasized to the point that treatment would not result in much benefit. As such, some tumors are more in need of being treated, and therefore of being detected, than are others. It has been recognized that two tests, even with identical sensitivities and specificities, may still be distinguishable by the type of tumors they tend to detect or miss. Therefore, it has been suggested that the standard 2x2 table, which allows no provision for this consideration, be modified by splitting the truepositive group into two categories, based on the size of the tumor detected (3). However, it is preferable to split both the true-positive group and the falsenegative group. Doing this requires a measurement of amenability to treatment that is available even for subjects with negative results. It is not the purpose of this article to identify the ideal measures of amenability to treatment, but rather to illustrate how such measures would be used to refine the assessment of a diagnostic test. To this end, we use DRE status as a measure of amenability to treatment when evaluating PSA for prostate cancer, and we use CT status as a measure of amenability to treatment when evaluating PET for non-small cell lung cancer.