Every year approx. 4.500 people are diagnosed with colorectal cancer (CRC) in Denmark. The 5-year survival rate has been observed to be approx. 50% as CRC is often diagnosed at a late stage. In 2014 a screening program will be established in Denmark. However, current screening methods are being criticized, making chemical biomarkers the center of attention. At present no useful biomarkers for detecting CRC have been identified.
Data mining in The North Denmark Region’s data warehouse, that contains laboratory and patient administrative data, should allow for detection of new, possible chemical biomarkers for CRC.
Using pattern recognition, 15 biochemical tests were investigated with linear regression, mean and standard deviation, in the period before patients were diagnosed with CRC. Linear and quadratic discriminant analyses were used to separate 125 patients with CRC from 6,149 patients without cancer. Data were normalized to obtain independence of the laboratories’ normal values.
The simplest classifier in relation to the highest sensitivity and specificity was a 3-dimensional quadratic clas- sifier based on the linear regression of P-Albumin including the mean value of B-Thrombocytes and P-Bilirubin. Sensitivity was 92% and specificity was 79.5%. Completeness and distribution of the data in the data warehouse are major limitations in this project, due to the retrospective nature, as it reduces the number of samples and may skew the results.
Despite the limitations, the classifier has a higher sensitivity but a lower specificity compared to fecal occult blood test, meaning that the classifier could detect a greater percentage of patients with CRC compared to the fecal occult blood tests. Thus, possible chemical biomarkers for CRC have been identified with the data warehouse through pattern recognition.