Use of Term Biomarker vs. Analyte

Following is a guest blog by Dr. Jules J. Berman, commenting on a post of January 5, 2006, that can be reviewed here.

Dr. Friedman defined analyte early in his blog in the following way: "The term analyte is defined as a substance that a laboratory test seeks to detect (or analyze).  I discovered many definitions for the term biomarker on the web but the one that I liked the most is the following: any molecular species found to provide correlation to a particular phenotype or perturbation of a biological system."

He went on to discuss cancer biomarkers as monitors of previously diagnosed cancers. He seemed to think that a biomarker proven to effectively monitor the recurrence of cancer would logically have value as a screening test, writing "I have no argument with the science or business logic of this approach to cancer biomarkers.  However, from a healthcare consumer's perspective, I would suggest that the underlying logic used by Aetna does not exactly make much sense. I think that consumers would argue that if a biomarker can efficiently detect recurrences of a tumor, it could also be used for wellness screening and the initial detection of cancer."

Well, without a long explanation, it just hasn't worked out that way. Tests that have been shown to have value as monitors of tumor recurrence sometimes have no value as screening tests in the general population.  If the prevalence in a population of a given tumor is very low, then screening tests need to have a very high specificity before they could be acceptable.  Otherwise, you'd have many well people with false positive diagnoses.  Basically, it's unwise to infer too much about the value of this-or-that test in this-or-that setting without lots of clinical validation.

The issue of clinically validating the specific claims for a biomarker is really what I wanted to discuss here.  My perception is that there is intense interest in developing new predictive markers for cancers.  A predictive marker alerts you whether a patient with a specific stage and type of disease is likely to respond to a certain treatment.  In this particular case, the biomarker has nothing to do with diagnosis or staging or tumor sub-typing. All of those features of the tumor are used to determine whether the patient falls into a category of patients who might benefit from the biomarker test. 

The biomarker test may be what the FDA calls a multiplex test (such as a gene expression array or a proteomics array or a battery of measurements).  Predictive biomarkers are developed to help the oncologist choose the optimal therapy for the patient.  For many oncologists, predictive biomarkers are expected to be crucial to the future of cancer treatment. New predictive biomarker tests are developed at great expense and time using clinical trials.  When they are approved, they are typically approved only for the narrowly defined set of criteria that were tested in the clinical trial.

This is a new way of doing things.  Essentially, these predictive markers define new subsets of cancers that are not morphologically distinguishable.  If they were, you wouldn't need the biomarker.  Furthermore, the subset is determined by clinical behavior (response to a certain treatment).  The old way of doing things was that pathologists would notice some morphologic feature that distinguished one tumor from another, and the pathologist would give it a name. 

Sometimes, these morphologic variants could delineate clinical subsets of tumors (e.g., tall cell papillary thyroid tumor)  Sometimes they were useful to distinguish the tumor from some entirely different tumor that might have a similar morphologic feature (e.g. spindle cell squamous carcinoma and spindle cell sarcoma), and sometimes the morphologic variants were named to signify that something might exist and shouldn't be forgotten in the differential diagnosis (e.g. amelanotic melanoma).  The point I'm trying to make is that pathologists could invent tumor variants without first obtaining clinical validation.

Because predictive subsets of tumors only have meaning in terms of clinical response to treatment, they always need validation through a clinical trial.  Here's the rub: Clinical trials need to be clinically validated by post-trial experience.  Clinical trials are a type of scientific experiment.  They have all the problems inherent in any experiment.  They can be poorly designed, misinterpreted, invalid for under-represented patient subpopulations, un-repeatable, falsified, and things can change so that the treatment-response they were designed to predict becomes modified or obsolete. 

The true validation of predictive tests will come from clinical correlation with laboratory tests performed in medical centers wherein many different types of patients (male, female, different nationalities, different ages, complicating concurrent diseases, multiple medications) will receive the predictive test and will have documented short and long-term outcomes.

Some pathologists fear that the new predictive tests bypass the traditional role of the pathologist (as inventor of diseases).  Some pathologists resign themselves to the role of QA monitor in clinical trials for predictive markers (i.e. ensuring that patients are enrolled in clinical trials based on materials reviewed by an expert panel of pathologists). 

It is my perception that a new role for the pathologist (a role much more important than doing QA for clinical trials) is to determine the kinds of predictive tests that are most appropriate and also the conditions in which they succeed/fail based on correlations between test results (which they perform) and clinical outcome (which they will collect and review) from their own hospital data.