TY - JOUR TI - Predictive modeling for in-hospital mortality in pancreatic cancer patients DO - https://doi.org/doi:10.7282/T3474CTK PY - 2015 AB - Background: Pancreatic cancer is very aggressive with few symptoms before the cancer can diagnosed and usually the cancer is advanced when it was found. It is the most deadly type of cancer, and it is listed as the fourth leading cause of cancer-related death with a poor prognosis because of the late finding of the disease. To patients, the pancreatic cancer diagnosis is a life-changing disaster; however, to maximally extend the pancreatic cancer patients’ life is most important task after diagnosis. In this research study, we found certain statistical significant relationships between the death of pancreatic cancer patients and the comorbidities & demographics. Furthermore, patients with some comorbidities or demographics can make their life longer or shorter. Objectives: To develop a mathematic model to predict the death of pancreatic cancer patients using certain patients’ comorbidities and demographics. Also, pancreatic cancer patients with certain comorbidities or demographics can predict their rest of life will be longer or shorter. Methods: The study uses HCUP NIS year 2005-2009 data files as research source database. With retrieving pancreatic cancer patient from NIS data files, the database used in this study includes pancreatic cancer patients’ information with their comorbidities and some demographics. The algorithm used in this research study in 1) Logistic regression ROC curve calculation 2) Logistic regression Odds Ratio calculation. Results: 1) In output ROC curve, the area under the ROC curve (AUC) is 0.725 which is >0.5. It means when randomly pick one patient from the disease group (diagnosed as pancreatic cancer) and one from the no-disease group (not diagnosed as pancreatic cancer) and do the test on both. The patient with the more abnormal test result (Died) should be the one from the disease group (diagnosed as pancreatic cancer). 2) 11 comorbidities can significantly influence on life of pancreatic cancer patients, both LCL and UCL of odds ratio are either <1 or >1 in all 11 comorbidities, which means if patients were diagnosed as one of 11 comorbidities, they might either live longer or live shorter. Conclusions: The study results were cross verified by 3 types of cancer patients, which is pancreatic cancer, breast cancer and stomach cancer. Those 11 comorbidities have the same statistical significance effect on influencing patients’ life. So far no literatures can be found on such kind of study, this study can be considered as a new research field in the future. KW - Biomedical Informatics KW - Pancreas--Cancer--Treatment KW - Cancer--Patients LA - eng ER -