Description

Lagendijk et al studied different approaches for distinguishing between primary ovarian adenocarcinomas and metastatic tumors from the colon or breast in women. One approach uses classification rules with scores based on immunohistochemical staining of the tumor. The study was done at the Free University Hospital in Amsterdam, The Netherlands.


 

Immunohistochemical staining battery:

(1) GCDFP-15 (gross cystic disease fluid protein)

(2) ER (estrogen receptor)

(3) CEA (carcinoembryonic antigen)

(4) CK7 (cytokeratin 7)

(5) CK20 (cytokeratin 20)

(6) CA-125

 

Intensity of Staining

Points

none

0

mild

1

moderate

2

marked

3

 

immunohistochemical score =

= (percent of tumor staining as a whole number from 0 to 100) * (points for staining intensity)

 

where:

• minimum immunohistochemical score: 0

• maximum immunohistochemical score: 300

 

metastatic colorectal carcinoma =

= (-0.010 * (immunohistochemical score for GCDFP-15)) + (0.046 * (immunohistochemical score for CEA)) - (0.008 * (immunohistochemical score for ER)) + (0.021 * (immunohistochemical score for CK20)) + (0.019 * (immunohistochemical score for CK7)) + (0.124 * (immunohistochemical score for CA-125)) - 6.907

 

metastatic breast carcinoma =

= (0.350 * (immunohistochemical score for GCDFP-15)) + (0.016 * (immunohistochemical score for CEA)) + (0.024 * (immunohistochemical score for ER)) + (0.001 * (immunohistochemical score for CK20)) + (0.031 * (immunohistochemical score for CK7)) - (0.013 * (immunohistochemical score for CA-125)) – 6.175

 

primary ovarian carcinoma =

= (-0.127 * (immunohistochemical score for GCDFP-15)) + (0.008 * (immunohistochemical score for CEA)) + (0.003 * (immunohistochemical score for ER)) + (0.006 * (immunohistochemical score for CK20)) + (0.021 * (immunohistochemical score for CK7)) + (0.031 * (immunohistochemical score for CA-125)) – 3.998

 

Interpretation (page 286:

• The tumor with the highest score is selected as the probable point of origin.

• Poorly differentiated carcinomas were usually correctly classified.

• In the learning set used, 89% of tumors were correctly classified (98% of colorectal, 88% of ovarian, and 80% of breast). In the test set 82% of tumors were correctly classified.

 

NOTE: The text of the articles infers that the values from the classification rules can be used to derive the posterior probability for each diagnosis.

 


To read more or access our algorithms and calculators, please log in or register.