Predictive System Incorporates Multiple Prognostic Markers

October 1, 2006

Researchers are using large patient datasets and computer programs to develop an expanded cancer staging system, moving beyond the conventional three markers—tumor size, nodal involvement, and metastasis—used in TNM staging. The new system, presented at the International Union Against Cancer's World Cancer Congress, uses TNM stage and other factors, such as histology and tumor grade, to fine tune and personalize prognosis

WASHINGTON—Researchers are using large patient datasets and computer programs to develop an expanded cancer staging system, moving beyond the conventional three markers—tumor size, nodal involvement, and metastasis—used in TNM staging. The new system, presented at the International Union Against Cancer's World Cancer Congress, uses TNM stage and other factors, such as histology and tumor grade, to fine tune and personalize prognosis (abstract 9-57).

Group Testing Concept

Based on the concept of group testing, the new predictive system uses large databases of cancer patients, such as the National Cancer Institute's SEER database. Group testing evaluates the association between survival and all potential prognostic factors, individually and in various combinations.

"Perhaps the main point is that outcome prediction depends on the selection of the prognostic factors that are used to make the prediction," said principal investigator Donald E. Henson, MD, adjunct professor and co-director of the Office of Cancer Prevention and Control at George Washington Cancer Institute, part of The George Washington University Medical Center.

Evalulating the System

To evaluate their system, the researchers used SEER records of 70,045 lung cancer patients from 17 different parts of the country, who were diagnosed from 1998 through 2003. In addition to TNM stage, the researchers examined histologic type (adenocarcinoma or squamous cell carcinoma), grade, age, race, and sex.

As expected, different combinations of factors had different degrees of association with survival. For instance, the combination of race, stage I, grade 1-2, and adenocarcinoma had prognostic significance. Whites with this combination had significantly longer survival times than blacks (P < .05).

On the other hand, the combination of race, stage I, grade 3-4, and squamous cell carcinoma had no significance; in this scenario, blacks and whites had the same survival curves.

Similarly, female sex combined with stage III, grade 1-2, and adenocarcinoma was associated with significantly better survival than male sex combined with the same three factors. But with the combination of stage III, grade 1-2, and squamous cell carcinoma, the patient's sex made no difference.

"The general statement that women do better than men, or that blacks do worse than whites, does not hold true across the board, but does depend on the prognostic factors that have been selected," Dr. Henson said.

A Sequence of Steps

According to the authors, group testing involves a sequence of steps. The first step uses a statistical test to partition patients into groups by using one factor, such as stage. For any subsequent step, partition is done by using another factor, not used in a previous step, and the groups from the step immediately preceding the current step. The process is continued until all the factors have been used. Prognostic factors are evaluated not only by their relation to survival but also by the hazard function, which can, for instance, be used to identify time-limited prognostic factors.

The system could be employed, the authors said, to add any number of prognostic factors, including treatment, clinical variables such as comorbidities, and molecular markers, while preserving the TNM staging system.

"We feel that in the future, with expanding computerization in medicine, physicians will analyze patient care the way we did it here," Dr. Henson commented.