Immunohistochemisty has been used in a variety of contexts for determining the

amount of protein in tissue cells. The process requires a lab technician to embed

the tissue of interest (say, for example, a cancer tumor) in parafin wax, and cut

thin slices that can be examined using a microscope. Before examining these slices,

the pathologist will stain each slice with a different chemical and then incubate the

samples for a required period of time. After the incubation period, the cells of the

tissue will flouresce if a particular protein is present (the protein differs according

to the chemical treatment). The pathologist then examines the tissue sample to

determine the amount of protein available in the tumour sample.

I have worked with a group of pathologists (Inti Zlobec, Carolyn Compton,

Jeremy Jass primarily) and an applied mathematician (Nilima Nigam) on a set of

papers that have tried to improve the reliability of the immunostaining process and

to also more effectively use immunostaining results to predict the efficacy of a par-

ticular colo-rectal cancer treatment, brachytherapy. The brachytherapy treatment

allows an oncologist to focus doses of radiation within the colon and rectum, rather

than using the standard non-specific, general field radiation or chemotherapy which

do not target specifically diseased tissue. Predicting efficacy of the treatment a pri-

ori is important, because the treatment is invasive and ineffectively treating patients

can lead to unnecessary death, as colo-rectal cancer survival time depends crucially

on the timing of treatment.

We have published work that has established several basic principles for immunostaining of colo-rectal cancer tumors:

1. Showed that the scoring of the immunostaining procedures is generally repro-

ducible and reliable, but that the reliability varies from protein to protein.

2. Established the predictive ability of the APAF, VEG-F and EGFR proteins in

determining the efficacy of the new brachytherapy procedure.

3. Showed that previously utilized cutoffs for determining positive or negative

presence of protein in cancer tumor cells were not optimal and potentially

not necessary (i.e. that actually using raw score for percentage stained can

actually help to improve prediction).

4. Obtained preliminary results that more advanced statistical techniques such

as neural network models and classification and regression trees can, in some

cases, help to improve prediction models (likely due to the highly nonlinear

and interactive nature of the proteins related to the cancer).