Prognostic Factors for Survival in Patients with Gastric Cancer using a Random Survival Forest

Abstract

<br /> <strong><span style="font-size: small;">Background: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">Gastric cancer is the fifth most common cancer and the third top cause of cancer related death with </span></span><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">about 1 million new cases and 700,000 deaths in 2012. The aim of this investigation was to identify important factors for outcome using a random survival forest (RSF) approach. </span></span><strong><span style="font-size: small;">Materials and Methods: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">Data were collected from 128 gastric cancer patients through a historical cohort study in Hamedan-Iran from 2007 to 2013. The event under consideration was death due to gastric cancer. The random survival forest model in R software was applied to determine the key factors affecting survival. Four split criteria were used to determine importance of the variables in the model including </span></span><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">log-rank, conversation?? of events, log-rank score, and randomization. Efficiency of the model was confirmed in terms </span></span><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">of Harrell's concordance index. </span></span><strong><span style="font-size: small;">Results: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">The mean age of diagnosis was 63 ±12.57 and mean and median survival times were 15.2 (95%CI: 13.3, 17.0) and 12.3 (95%CI: 11.0, 13.4) months, respectively. The one-year, two-year, and three-year rates for survival were 51%, 13%, and 5%, respectively. Each RSF approach showed a slightly different ranking order. Very important covariates in nearly all the 4 RSF approaches were metastatic status, age at diagnosis and tumor size. The performance of each RSF approach was in the range of 0.2966-0.3250 and the best error rate was obtained by the log-rank splitting rule; second, third, and fourth ranks were log-rank score, conservation of events, and the random splitting rule, respectively. </span></span><strong><span style="font-size: small;">Conclusion: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">Low survival rate of gastric cancer patients is an indication of absence of a screening program for early diagnosis of the disease. Timely diagnosis in early phases increases survival and decreases mortality. </span></span>