We evaluated the application of three machine learning algorithms, including logistic regression, support vector machine and back-propagation neural network, for diagnosing congenital heart disease and colorectal cancer. By inspecting related serum tumor marker levels in colorectal cancer patients and healthy subjects, early diagnosis models for colorectal cancer were built using three machine learning algorithms to assess their corresponding diagnostic values. Except for serum alpha-fetoprotein, the levels of 11 other serum markers of patients in the colorectal cancer group were higher than those in the benign colorectal cancer group (P < 0.05). The results of logistic regression analysis indicted that individual detection of serum carcinoembryonic antigens, CA199, CA242, CA125, and CA153 and their combined detection was effective for diagnosing colorectal cancer. Combined detection had a better diagnostic effect with a sensitivity of 94.2% an specificity of 97.7%; combining serum carcinoembryonic antigens, CA199, CA242, CA125, and CA153, with the support vector machine diagnosis model and back-propagation, a neural network diagnosis model was built with diagnostic accuracies of 82 and 75%, sensitivities of 85 and 80%, and specificities of 80 and 70%, respectively. Colorectal cancer diagnosis models based on the three machine learning algorithms showed high diagnostic value and can help obtain evidence for the early diagnosis of colorectal cancer.
False positive results
- Presence of inflammatory processes
- Benign liver diseases and consequential disturbancies in metabolism and excretion (AFP, CA 15.3, CA 19.9, CEA, TPA)
- Disturbancies of renal function (beta-2-microglobulin, calcitonin, CA 15.3, CA 19.9, CEA, PSA)
- Extensive tumor necrosis
- As a consequence of diagnostic and therapeutic procedures (digitorectal examination - PSA, mamography, surgery, radio and chemotherapy)
- As a consequence of different physiological conditions (pregnancy - hCG, CA 125, CA 15.3, MCA, AFP, menstrual cycle - CA 125)
- Complete absence of production (e.g. CA 19.9 in Le(a-b-) persons)
- Insufficient expression of a certain antigenic determinant (or production in only some of tumor cells)
- Insufficient blood circulation in the tumor
- Production of immune complexes with autoantibodies
Cancer may be diagnosed in advanced stages, when the patient has already developed metastasis, with symptoms that can be also observed in benign diseases. The objective of this study was to evaluate tumor marker sensitivity and specificity in the differential diagnosis of patients with suspected signs of cancer. We studied 2.711 consecutive patients admitted to the Internal Medicine Department of our hospital with suspected cancer; 1.240 patients had non-malignant processes and 1.471 had malignant disease. Determinations were considered positive for suspected malignancy when serum levels were carcinoembryonic antigen >15 ng/ml (>20 in patients with renal failure or liver disease), alpha fetoprotein >40 ng/ml (>80 ng/ml in patients with liver diseases), carbohydrate antigen (CA) 19.9 > 200 U/ml (>500 U/ml in patients with liver diseases or gamma glutamyl transpeptidase (GGT) 1.000 U/ml in patients with jaundice or GGT > 150 UI/L), neuron-specific enolase >45 ng/ml (renal failure >50 ng/ml; samples with hemolysis were excluded), prostate-specific antigen > 30 ng/ml (excluding acute prostatitis), tumor-associated glycoprotein-72 >80 U/ml, cytokeratin 19 fragment 21-1 > 7.5 ng/ml (>19 ng/ml in patients with renal failure; >11 ng/ml in patients with liver cirrhosis or jaundice), >3.5 ng/ml for squamous cell carcinoma (excluding patients with renal failure or skin disorders), CA 15.3 >100 U/ml, and CA 125 >350 U/ml (>600 U/ml in patients with pleural effusion and >900 U/ml in those with ascites). There was a specificity of 97.6% in patients without malignancy, 67.4% of sensitivity in patients with malignancy, and 75.4% of sensitivity in the 1,280 patients with epithelial tumors (53.7% in patients with locally advanced tumors and 79.4% in patients with metastases). Sensitivity was 81.4% in patients with cancer of unknown primary site. Tumor markers were useful in the differential diagnosis between epithelial and non-epithelial tumors, brain masses (metastases vs. primary tumors), and between benign or malignant origin of different clinical situations such as wasting syndrome, effusions, liver or bone lesions, and effusions with a positive predictive value higher than 95%. Tumor markers are useful as an aid in the evaluation of the risk of cancer of these patients with suspected cancer and may be useful to reduce the hospitalization time, morbidity, and the number of diagnostic tests required for diagnosis.