Assessing the clinical impact of prognostic factors: When is "statistically significant" clinically useful?

Very few tumor markers have been recommended for routine clinical care of patients with breast cancer [1]. A framework to determine the clinical utility of tumor markers is required. In a previous publication, a "Tumor Marker Utility Grading System" (TMUGS) was proposed [2]. TMUGS included a semiquantitative grading scale (0-3+) which can be used to assign a score to a given tumor marker for a given outcome. Only those markers that are felt to be sufficiently strong to influence a therapeutic decision that results in improved clinical outcome for the patient are recommended. The studies from which data are used to assign a TMUGS grade can be placed into one of five Levels of Evidence (LOE). An extension of TMUGS ("TMUGS-Plus") is now proposed in which the relative strength of a prognostic or predictive factor can be estimated and expressed in terms of a risk ratio (RR) for prognostic factors or benefit ratio (BR) for predictive factors. Three categories of prognostic factors and three categories of predictive factors are proposed (strong, moderate, and weak). It is recommended that only LOE type I studies (prospective, highly powered studies of the tumor marker, or meta-analysis of LOE II or Ill datasets), be used to estimate the RR or BR of a given factor. Finally, a matrix, based on assumptions of acceptable absolute benefits relative to risks, is proposed in which any given tumor marker can be assessed for its clinical utility. TMUGS-Plus should aid in the assessment of published data regarding clinical utility of tumor markers. Perhaps more important, clinical investigators can use TMUGS-Plus to design tumor marker studies that will fulfill criteria for clinical utility, resulting in more rapid acceptance of tumor markers for routine clinical use.