Artikel
Bayesian survival analysis of colorectal cancer registry data
Suche in Medline nach
Autoren
Veröffentlicht: | 29. August 2017 |
---|
Gliederung
Text
Introduction: Survival analysis provides several modeling approaches to analyze cancer registry data. Model choice depends on the underlying data.
In colorectal cancer, prognosis strongly depends on radicality of the surgical therapy. The extent of lymph node metastasis can be described by the pN-classification, whereas the quality of the regional radical surgery is substantially indicated by the number of detected lymph nodes.
Methods: Explorative data analysis was followed by a variable and model selection. Finally, we applied a bayesian accelerated failure time model assuming a weibull-distributed survival function. Survival time in months was modeled depending on seven main effects and one interaction. Data analysis was done in R and WinBUGS.
Results: During 01/2010–12/2014 a total of 470 patients were registered in a hospital colorectal cancer center providing a 100%-follow-up. The median age was 72 years (range 24 to 95). Overall, 148 (31.5%) patients died, 15.1% died due to cancer and 16.4% due to other/unknown reasons. A median number of 23 lymph nodes were detected by histology. Survival time of the colorectal cancer patients strongly depended on lymph node ratio (number of positive lymph nodes divided by the total number of removed lymph nodes), UICC-stage, age and chemotherapy (yes/no). An acceleration factor of 0.21 (95% CI: 0.08–0.47) was observed for age group >80 years (baseline age group <60 years). The estimated acceleration factor strongly decreased with increasing lymph node ratio. An acceleration factor of 6.23 (95% CI: 3.31–12.0) was observed for those receiving chemotherapy compared to those who did not.
Discussion: Our results show the relevance of the lymph node ratio as a predictor of colorectal cancer survival. This result highlights the importance of the request of removing and pathologically detecting a large number of lymph nodes.
Die Autoren geben an, dass kein Interessenkonflikt besteht.
Die Autoren geben an, dass kein Ethikvotum erforderlich ist.
Der Beitrag wurde bereits vorgestellt: DGVS 2016