Study selection and characteristics
A total of 222 papers were initially retrieved from PubMed, Web of Science, EMBASE and Cochrane library database. As shown in Fig. 1, the literature search process was summarized in the flow diagram according to Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) [23]. Ultimately, 20 studies [8, 15-18, 24-38] including 6512 patients were enrolled in this meta-analysis after excluding ineligible studies.
Basic characteristics and main outcomes of the included 20 studies were shown in Table 1. These studies were published from 2011 to 2020, with a research duration ranging from 2000 and 2017. The sample size of these studies ranged from 47 to 1347, with a total number was 6512. As for the therapeutic methods, patients in 14 studies received chemotherapy, and pancreatectomy was chosen in another 6 studies. The NOS score of all included studies were equal or greater than 5, which meant relatively high quality. 16 studies were conducted in Asia (including China, Japan, and South Korea), and 4 studies were performed in western countries (including the United Kingdom, Italy and Australia). Most studies were assessed with multivariate analysis except that 4 studies were only assessed with univariate analysis.
Prognostic value of mGPS in pancreatic cancer
As shown in Table 2 and Fig. 2, a total of 20 studies evaluated the association between the level of the mGPS and OS for pancreatic cancer patients. The mGPS ranged from 0 to 2 based on the CRP and albumin levels as discussed above. Since 12 cohorts divided the participants into 2 groups (high vs. low, mGPS=1 as the cutoff value), and other participants in another 8 cohorts were grouped into 3 groups (mGPS=0, 1 and 2), we separately performed a meta-analysis for different groupings. When divided into two groups, we defined an mGPS of 0 as the low group and an mGPS of 1 or 2 as the high group. There was evidence for moderate heterogeneity among studies (I2=61.7% and P=0.003), so random-model was applied. The results indicated the statistically significant relationship between the mGPS and prognosis of patients with pancreatic cancer, and the OS may be better for patients with lower mGPS compared with patients with higher mGPS (HR=1.50, 95% CI=1.20–1.89, P<0.0001) (Fig. 2A). When divided into three groups, the random-effect model was applied due to significant heterogeneity (I2=86.6% and I2=80.4% for mGPS=1 vs. mGPS=0 and mGPS=2 vs. mGPS=0, respectively). The results also demonstrated a statistically significant difference between the high mGPS and poor survival for pancreatic cancer patients (mGPS=1 vs. mGPS=0: HR=1.68, 95% CI=1.25-2.27, P=0.001; mGPS=2 vs. mGPS=0: HR=1.90, 95% CI=1.36–2.67, P<0.0001, Fig. 2B to C).
Subgroup analyses of the association between mGPS and OS
In view of moderate heterogeneity among studies, we conducted subgroup analyses for OS by factors of the therapeutic method and study region, and the heterogeneity partly decreased in several subgroups. Detailed results of subgroup analyses are summarized in Table 2 and Fig. 2. In 12 studies that chose mGPS=1 as the cutoff value, the subgroup analyses based on the region showed that patients in eastern areas with higher mGPS had a decline of OS (HR=1.60, 95% CI=1.24-2.06, P<0.0001), which was not observed in patients in western areas (HR=1.34, 95% CI=0.80-2.26, P=0.268; Fig. 2D). Patients with higher mGPS in another 4 studies also demonstrated poor OS in eastern areas (mGPS=1 vs. mGPS=0, HR=1.71, 95% CI=1.25-2.35, P=0.001 and mGPS=2 vs. mGPS=0, HR=1.83; 95% CI=1.29-2.62, P=0.001; Fig. 2E to F). Further subgroup analyses based on treatment regimen showed that higher level of mGPS was significantly associated with worse OS in patients receiving chemotherapy (HR=1.45, 95% CI=1.05–2.02, P=0.025) as well as pancreatectomy (HR=1.64, 95% CI=1.25–2.15, P<0.0001) in 12 studies that chose mGPS=1 as the cutoff value (Fig. 2G). However, as for the 8 studies in which patients were divided into 2 groups (mGPS=1 vs. mGPS=0 and mGPS=2 vs. mGPS=0; Fig. 2H to I), the stratified analysis by the factor of therapeutic methods indicated that higher mGPS was linked to the poor OS in patients receiving chemotherapy (HR=1.78, 95% CI=1.30-2.44, P<0.0001 and HR=1.87, 95% CI=1.17-2.98, P=0.008, respectively), but not in patients undergoing pancreatectomy (HR=1.52, 95% CI=0.60-3.84, P=0.375 and HR=2.01, 95% CI=0.94-4.30, P=0.078, respectively).
Publication bias and Influence analyses
The Begg’s funnel plots seemed to be symmetrical, suggesting the absence of significant publication bias in all overall meta-analyses (Fig. 3A to C). The Begg’s test linear regression test also proved that there was no significant publication bias (each P>0.05). Using trim and fill analysis, we only found that 3 studies evaluating the prognostic role of mGPS in OS in pancreatic cancer remained unpublished when participants were divided into 2 groups (high vs. low, mGPS=1 as the cutoff value, Fig. 3D to F). The filled meta-analytic results for OS (pooled HR = 1.31, 95% CI [1.01-1.69], p < 0.001) also supported our original results. To examine the stability of the pooled HRs in OS, influence analysis was carried out with the successive omission of each study. The leaving-one-out study revealed that there was no individual cohort influencing the results greatly (Fig. 4A to C).