Skip navigation

Tumour markers and cancer treatment

It is often the case that the more important a paper, the harder it is to précis. That is certainly true of a fundamental exposition of the use of predictive and prognostic tests in cancer, and how they may be used to guide patient and doctor in choosing the correct, or any, adjuvant systemic therapy [1]. This paper is of massive importance to those using or producing tumour marker test results because it is likely to change the way you think.

The problem

I have a woman who has breast cancer. How do I know what her chances are of surviving the next 10 years? How do I know whether it is appropriate to treat her with systemic adjuvant therapy which might help her, but which has the certainty of some toxic effects that will harm her?

Are there any tests I can use that will tell me something about the likelihood of metastasis or growth rate of the cancer? If such a test existed, it would be a prognostic test.

Are there any tests associated with sensitivity and/or resistance to particular therapeutic agents? If such a test existed it would be a predictive test.


Not the easiest acronym to tip off the tongue, but the 'Tumour Marker Utility Grading System' is one which has been defined by an expert panel of the American Society of Clinical Oncology. It uses a grading system from 0 to 3+, with 0 implying that sufficient data exist to say that a test is of no utility, while 2+ or 3+ implies that a marker should be considered or absolutely should be used.

The basis of these gradings is on levels of evidence, with the highest level being that of prospective, highly powered studies specifically addressing the issue of tumour marker utility. The lowest level of evidence is that where specimens happen to have been collected for a variety of reasons.

It won't surprise thoughtful readers that most tests fall into a category with low levels of evidence which mean that their utility is uncertain. This means that most work done on most tumour markers is almost worthless. We may have highly valuable tools to assist decision-making in some tumour marker tests, but we just don't know it.


This paper suggests a further extension of TMUGS into TMUGS-plus, in which the statistical associations found in reports on tumour markers is overturned in favour of clinical utility. This latter categorisation of strong, moderate or weak prognostic ability depends on how far a patient might be moved across prognostic borders - for instance if a positive or negative test moves a patient from a low risk of dying over 10 years to a high risk of dying.

Clearly, this is of immense clinical importance. It would mean, for instance, that while adjuvant treatment may not be sensible if the risk of death is low, it would be an absolute imperative if the risk of death was high and the chance of success moderate.

The clever bit

What makes this paper so important is that it shows us how to combine prognostic and predictive factors in ways that will help doctors and patients make decisions about therapy. They focus on pulling them together to produce a figure for the absolute reduction in mortality due to systemic therapy in patients for whom a marker is positive compared to those in whom a maker is negative.

The authors use a pre-determined set of recommendations for treatment. Thus if 10% or more of a group of patients being treated benefited (in this case lived for at least 10 years), then treatment would be absolutely recommended. If it was between 6% and 9% treatment would be probable, between 4% and 5% considered but not strongly recommended, and between 0% and 3% not recommended. These figures would be equivalent to NNTs of 10 or less, 11-17, 18-33 and more than 33.

A new paradigm

This paper may be one of the most important ever written on diagnostic tests, but it will never be an easy read, especially the first time. By the third of fourth time you will start to see that it opens up new horizons in the way that diagnostic tests are looked at, and the way in which tests and treatments can be combined together to produce real benefits for patients.

For cancer, it will help define what is acceptable in the evaluation of diagnostic tests. For cancer therapy, it will help define how clinical trials may be designed to demonstrate effectiveness. For new genetic tests which might have predictive or prognostic relevance it provides a framework which will allow their evaluation to be faster and more certain.

Any paper which says that a p value, even if very low, does not necessarily imply that a factor (test, result, whatever) is useful tells you that it has some thought behind it.

Taken together, this paper, and that on diagnosis of prosthetic joint infection on page 5, form a primer on how to think about diagnostic testing. They are a "must read" for anyone working in laboratories and producing tests. They should be a "must teach" for those responsible for educating doctors and other health professionals. Bandolier is delighted to see them, and will continue to seek out other examples of excellence in diagnostic testing.


  1. DF Hayes, B Trock, AL Harris. Assessing the clinical impact of prognostic factors: When is 'statistically significant' clinically useful? Breast Cancer Research and Treatment 1998 52: 305-319.

previous or next story in this issue