Translate this page into:
Prognostic factors in gastric cancer using log-normal censored regression model
Reprint requests: Dr Bijan Moghimi-Dehkordi, Research Center for Gastroenterology & Liver Disease, 7th floor of Taleghani Hospital Tabnak Street, Evin, Tehran, Iran e-mail: b_moghimi_de@yahoo.com; Aminphg@gmail.com
-
Received: ,
Abstract
Background & objectives:
Gastric cancer is one of the most common cancers in the world. It is rarely detected early, and the prognosis remains poor. Cox proportional hazard model is used to examine the relationship between survival and covariates. Parametric survival models such as log normal regression model can also be used for this analysis. We used log normal regression model in this study to evaluate prognostic factors in gastric cancer and compared with Cox model.
Methods:
We retrospectively studied the 746 patients diagnosed with gastric cancer admitted in a referral hospital in Tehran, Iran, from February 2003 through January 2007. Age at diagnosis, sex, extent of wall penetration, histology type, tumour grade, tumour size, pathologic stage, lymph node metastasis and presence of metastasis were entered into a log normal model. Hazard rate (HR) was employed to interpret the risk of death and the results were compared with Cox regression. The AIC (Akaike Information Criterion) was employed to compare the efficiency of models.
Results:
Univariate analysis indicated that with increasing age the risk of death increased significantly in both log normal and Cox models. Patients with grater tumour size were also in higher risk of death followed by those with poorly differentiated and moderately differentiated in tumour grade and advanced pathologic stage. The presence of metastasis was significant prognostic factor only in log normal analysis. In final multivariate model, age was still a significant prognostic factor in Cox regression but it was not significant in log normal model. Presence of metastasis followed by histology type were other prognostic features found significant in log normal results. Based on AIC, log normal model performed better than Cox.
Interpretation & conclusion:
Our results suggest that early detection of patients in younger age and in primary stages and grade of tumour could be important to decrease the risk of death in patients with gastric cancer. Comparison between Cox and log normal models indicated that log normal regression model can be a useful statistical model to find prognostic factors instead of Cox.
