Dataset Information

Which is a more accurate predictor in colorectal survival analysis? Nine data mining algorithms vs. the TNM staging system.

ABSTRACT:

Objective

Over the past decades, many studies have used data mining technology to predict the 5-year survival rate of colorectal cancer, but there have been few reports that compared multiple data mining algorithms to the TNM classification of malignant tumors (TNM) staging system using a dataset in which the training and testing data were from different sources. Here we compared nine data mining algorithms to the TNM staging system for colorectal survival analysis.

Methods

Two different datasets were used: 1) the National Cancer Institute's Surveillance, Epidemiology, and End Results dataset; and 2) the dataset from a single Chinese institution. An optimization and prediction system based on nine data mining algorithms as well as two variable selection methods was implemented. The TNM staging system was based on the 7(th) edition of the American Joint Committee on Cancer TNM staging system.

Results

When the training and testing data were from the same sources, all algorithms had slight advantages over the TNM staging system in predictive accuracy. When the data were from different sources, only four algorithms (logistic regression, general regression neural network, bayesian networks, and Naïve Bayes) had slight advantages over the TNM staging system. Also, there was no significant differences among all the algorithms (p>0.05).

Conclusions

The TNM staging system is simple and practical at present, and data mining methods are not accurate enough to replace the TNM staging system for colorectal cancer survival prediction. Furthermore, there were no significant differences in the predictive accuracy of all the algorithms when the data were from different sources. Building a larger dataset that includes more variables may be important for furthering predictive accuracy.

SUBMITTER: Gao P

PROVIDER: S-EPMC3404978 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:ObjectiveDespite controversy over its origin and definition, the significance of tumour deposit (TD) has been underestimated in the tumour node metastasis (TNM) staging system for colon cancer, especially in stage III patients. We aimed to further confirm the prognostic value of TD in stage III colon cancer and to establish a more accurate 'coN' staging system combining TD and lymph node metastasis (LNM).MethodsInformation on stage III colon cancer patients with a definite TD status was retrospectively collected from the Surveillance, Epidemiology and End Results (SEER) database between 2010 and 2017. The effect of TD on prognosis was estimated using Cox regression analysis. Maximally selected rank statistics were used to select the optimal cut-off value of TD counts. The predictive power of conventional N staging and the new coN staging was evaluated and compared by Akaike's information criterion (AIC), Harrell's concordance index (C-index) and time-dependent receiver operating characteristic (ROC) curves. Clinicopathological data of stage III colon cancer patients in the Xiangya database from 2014 to 2018 were collected to validate the coN staging system.ResultsA total of 39,185 patients with stage III colon cancer were included in our study: 38,446 in the SEER cohort and 739 in the Xiangya cohort. The incidence of TD in stage III colon cancer was approximately 30% (26% in SEER and 30% in the Xiangya database). TD was significantly associated with poorer overall survival (OS) (HR = 1.37, 95% CI 1.31-1.44, p < 0.001 in SEER). The optimal cut-off value of TD counts was 4, and the patients were classified into the TD0 (count = 0), TD1 (count = 1-3) and TD2 (count ≥ 4) groups accordingly. The estimated 5-year OS was significantly different among the three groups (69.4%, 95% CI 68.8%-70.0% in TD0; 60.5%, 95% CI 58.9%-62.2% in TD1 and 42.6%, 95% CI 39.2%-46.4% in TD2, respectively, p < 0.001). The coN system integrating LNM and TD was established, and patients with stage III colon cancer were reclassified into five subgroups (coN1a, coN1b, coN2a, coN2b and coN2c). Compared with conventional N staging, the coN staging Cox model had a smaller AIC (197097.581 vs. 197358.006) and a larger C-index (0.611 vs. 0.601). The AUCs of coN staging at 3, 5 and 7 years were also greater than those of conventional N staging (0.6305, 0.6326, 0.6314 vs. 0.6186, 0.6197, 0.6160). Concomitant with the SEER cohort results, the coN staging Cox model of the Xiangya cohort also had a smaller AIC (2883.856 vs. 2906.741) and a larger C-index (0.669 vs. 0.633). Greater AUCs at 3, 5 and 7 years for coN staging were also observed in the Xiangya cohort (0.6983, 0.6774, 0.6502 vs. 0.6512, 0.6368, 0.6199).ConclusionsNot only the presence but also the number of TDs is associated with poor prognosis in stage III colon cancer. A combined N staging system integrating LNM and TD provides more accurate prognostic prediction than the latest AJCC N staging in stage III colon cancer.

Project description:The aim of our study was to explore the value of the 8th edition TNM staging system on evaluating the prognosis of colorectal carcinoid. Colorectal carcinoid patients between 1988 and 2015 were selected in the Surveillance, Epidemiology, and End Results Program (SEER) database for analysis. About 4286 patients with colorectal carcinoid tumors were identified, of which were carcinoid tumor NOS (n = 1726), neuroendocrine carcinoma (NEC) (n = 1346) and other carcinoid tumor (OCT) (n = 591). Worsening 10-year CSS rates with increasing N status, M status, and SEER historic stage were demonstrated across all three above groups (all P < .05). In carcinoid tumor NOS, significant differences in CSS were found with increasing combined 8th AJCC stages (P < .001), except for that between stage II and stage III (10-year CSS rate: 82.6% vs 84.3%, P = .68). While combined 8th TNM stage in NEC and OTC exhibited greater separations in CSS despite on-going overlaps between groups. For carcinoid tumor NOS, stage II (HR = 3.37; 95% CI: 0.97-11.76), and stage III (HR = 2.09; 95% CI: 0.51-8.66) conferred no significant difference in CSS compared with stage I, while stage IV had an increasing HR of 5.09 (95% CI: 1.08-24.08). Although combined 8th AJCC stage had a good ability to distinguish 10-year CSS of patients with NEC or OCT, detailed 8th AJCC stage did not seem to be applicable. Detailed 8th AJCC categories of advanced stages in all the three groups conferred increased HRs with overlapping CIs. However, in the early and middle status, HRs did not increase with the increase of stages, or there was no difference in HRs between adjacent stages. Combined 8th TNM stage was not practical for judging the survival outcomes of colorectal carcinoid tumor NOS, especially in patients with stages II and III, but it provided useful prognostic information for NEC and OCT. However, for all carcinoid tumors, the prognostic values of detailed 8th AJCC stage were not enough accurate in the clinic. More optimized staging methods should be developed and validated in the future.