Abstract 5502: Prognostic tools for pancreatic cancer survival

• Published in: Cancer research (Chicago, Ill.) Vol. 72; no. 8_Supplement; p. 5502
• Main Authors: Sherman, Simon, Gleason, Michael, Mdzinarishvili, Tengiz, Sherman, Alexander, Shats, Oleg, Are, Chandrakanth, Sasson, Aaron, Haynatzki, Gleb
• Format: Journal Article
• Language: English
• Published: 15-04-2012
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2012-5502

Pancreatic cancer (PC) is a devastatingly lethal disease with the median survival time below six months after diagnosis. To extend survival time of PC patients, the most optimal treatment strategy that accounts for patient-specific clinical and demographic characteristics must be taken. To assist clinicians and patients in choosing optimal treatment regimens, innovative prognostic models and computing tools aimed at estimating outcomes from prospective treatments are needed. In this study, the impact of distinct demographic (race, gender, age at diagnosis, and marital status) and clinical (tumor stage, grade, location within the pancreas, histologic subtype, type of surgery performed) risk factors on the survival of PC patients was analyzed. To do this, data from the SEER 9 database (1974-2008) were used. PC cases that were not first primary or lacked histopathologic confirmation were omitted. Remaining data (nearly 90,000 PC cases) were characterized by age at diagnosis, stage, grade, subsite, and histologic subtype, and stratified by gender and race. To fully use the statistical power of the available data, we utilized a recently developed procedure [Cancer Informatics. 2011; 10:31-44]. By accounting for interrelated age-period-cohort effects, this procedure allowed us to use PC data collected during 35 years in the SEER 9 registries, rather than limiting the utilized data to the commonly used 5-year long cross-sectional data. Initially, we estimated potential risk factors by using univariate Cox proportional hazards (PH) regression models. To build a multivariate Cox PH model, all variables that were univariately significant were used as initial predictors of PC survival. From these predictors, the final ones were determined by backward elimination. The estimates of the significant risk factors were used to build a prognostic statistical model for PC survival. This model allows one to numerically estimate the probability for a patient to be alive at any specified time after PC diagnosis, based on the patient's risk factors. Validation performed by a bootstrap method and by using an external dataset (namely, the subset of PC cases from the SEER 17 registries presented in the eight geographical areas that are not included in the SEER 9 database) shows that the accuracy and precision of outcomes assessed by this model exceed those provided by other analogous tools. The computerized implementations (nomograms) of the resulting model run on different types of devices (desktops, smart phones, and tablets) and have user-friendly graphical user interfaces. The real-time use of such nomograms will help clinicians and patients in choosing the most appropriate treatments. After consideration of the computer-generated assessments, clinicians will be able to assist PC patients in making a more educated decision regarding potential treatments.
Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 5502. doi:1538-7445.AM2012-5502