Twenty-two studies (n=4,019) were included.
Disease stage and recurrence.
PCR positivity rates were significantly higher in patients with stage I or II disease compared with stage III disease: 95.1% versus 46.6% (p<0.001); the OR was 22.3 (95% CI: 14.7, 33.9; 20 studies, n=2,249). All studies used adequate sample sizes. Disease recurrence was significantly higher among PCR-positive patients than PCR-negative patients: 16.8% versus 8.7% (p<0.0001); the OR was 2.11, 95% CI: 1.69, 2.63; 15 studies, n=3,034). The sensitivity was 57.4% (95% CI: 52.1, 62.5), the specificity 61.1% (95% CI: 59.2, 62.9), the positive predictive value 16.8% (95% CI: 14.8, 19.0) and the negative predictive value 91.3% (95% CI: 89.9, 92.5).
Survival.
Overall survival (8 studies, n=2,140): 5 studies reported a significant correlation between overall survival and PCR status, while 3 studies reported negative results. The risk of death was significantly increased among PCR-positive patients compared with PCR-negative patients (HR 5.08, 95% CI: 1.83, 14.08, p=0.002). Significant heterogeneity was found (p<0.0001; I2=87.1%). The results were similar after including only studies reporting the Cox multivariate model. Two studies appeared to account for the heterogeneity (their characteristics were discussed). Meta-regression showed that none of the covariates were significantly associated with HRs. There was no evidence of publication bias from the funnel plot or Egger’s test (p=0.94). None of the studies estimated the sample size based on this outcome.
Disease-free survival (13 studies, n=1,292): 10 studies reported a significant correlation between disease-free survival and PCR status, while 3 studies reported negative results. The risk of disease progression was significantly increased among PCR-positive patients compared with PCR-negative patients (HR 3.41, 95% CI: 1.86, 6.24, p<0.0001). Significant heterogeneity was found (p<0.0001; I2=94.4%). The results were similar after including only studies reporting the Cox multivariate model. Four studies appeared to account for the heterogeneity. Meta-regression showed that none of the covariates were significantly associated with HRs. Eight studies were considered to have adequate sample size.