Ten RCTs (n=3,499) were included. Two further trials (n approximately 200) were excluded as IPD were not available.
PVI versus no further treatment.
The reduction in the annual odds of death was 13.6% (standard deviation, SD=4.9; P=0.006), indicating that cytotoxic PVI was more effective than no further treatment. This corresponded to an absolute improvement in survival of 4.7% (SD=1.2; P=0.006). When stratified for analysis by Dukes' disease stage, the reduction in odds of death with PVI increased to 18.0% (SD=5.1; P=0.0004). The reduction in the annual odds of death was 27% (SD=14) for participants with Dukes' stage A, 18% (SD=8) for participants with Dukes' stage B, and 17% (SD=8) for participants with Dukes' stage C. Cause-specific mortality showed that post-operative and non-colorectal cancer death rates were not significantly different between the two treatment groups, but deaths from colorectal cancer were significantly fewer in the PVI group than in the control group (OR 17%, SD=5%; P=0.002). An analysis of site of first recurrence showed that 36% involved the liver; however, heterogeneity significantly affected the result (P=0.0003).
PVI (5-FU plus heparin) versus PVI with heparin alone or with urokinase (4 RCTs).
The annual odds of death were significantly lower with cytotoxic PVI than non-cytotoxic PVI (OR 20, SD=10; P=0.04). This treatment effect was identical when the analysis was stratified by stage, or included participants with Dukes' stages A, B or C. There were significantly fewer liver recurrences (35% reduction, SD=17; P=0.04) and deaths from colorectal cancer (24% reduction, SD=12; P=0.04) with cytotoxic PVI than non-cytotoxic PVI.