Statistical Methods

We performed analysis using a random-effect model in case that there was significant heterogeneity. We also performed sensitivity analysis to assess whether the summary estimates are robust to inclusion of studies. One study was removed every time, and the rest were analyzed to evaluate whether the results could have been affected significantly by a single study. Heterogeneity was assessed by value of I². Publication bias was evaluated by the use of Begg funnel plot and Egger linear regression test. A pooled HR >1 suggested that underweight, overweight, or obesity predicted an unfavorable prognosis for pancreatic cancer patients. Oppositely, a pooled HR <1 suggested a favorable prognosis for those patients. It was regarded as statistically significant if the 95% CI of HR did not overlap 1. All P values were 2-sided. P < 0.05 was regarded as statistically significant. All analyses were performed using the Stata version 12.0 software (StataCorp, College Station, TX, http://www.stata.com). Given that our study was a review of previous published studies, ethical approval or patient consent was not required.