Literature screening and characteristics
The literature screening process is illustrated in Figure 1. A total of 328 articles from the four databases and two articles from a manual reference search were initially selected. After removing duplicates, 224 studies remained. We excluded 189 articles after reviewing article abstracts. Next, seven articles were removed for failing to calculate the effect size; 14 studies were eliminated due to their being conference abstracts; and two studies were excluded because PD-L1 was not detected through IHC. Finally, altogether 12 articles encompassing 2,321 patients were selected for the meta-analysis.
All characteristics of the studies are displayed in Table 1. Four studies were performed in China [18-21], four in Korea [22-25], two in Japan [26, 27], and one each was in the US [28] and Norway [29], respectively. The cut-off value was determined using the form of percentage except Cho’s, which ranged from 2% to 50%. According to the cut-off values, every article described the number of patients with PD-L1 overexpression. All studies referred to each disease stage according to Ann Arbor Staging except Bi’s. In addition, all studies were retrospective and reported the association between PD-L1 and OS. Patients in the studies had a histologically confirmed NHL diagnosis and subtype.
Association between PD-L1 overexpression and OS in NHL
We calculated a pooled HR of 1.40 (95% CI: 0.90–2.19; P = 0.137) for OS. The result indicated that PD-L1 overexpression was not associated with NHL prognosis. Significant heterogeneity, however, existed among the selected studies (I2 = 70.6%, P < 0.001; Figure 2).
Association of PD-L1 overexpression with OS in DLBCL
DLBCL, accounting for 30–40% of NHL, is the most common subtype of NHL. There were 863 DLBCL patients from six articles in our study. A meta-analysis was performed that was designed to assess prognosis among DLBCL patients. The result showed that the pooled HR was 1.70 (95% CI: 1.05–2.74; P = 0.031) with I2 = 47.2% (Figure 3). This indicated that PD-L1 overexpression potentially predicted a poor prognosis in DLBCL patients.
Association between PD-L1 overexpression and clinicopathological characteristics
We also investigated the association of PD-L1 overexpression with clinicopathological characteristics. The results suggested that PD-L1 overexpression was more frequent in patients with B symptoms (OR = 1.91, 95% CI: 1.17–3.10; P = 0.09), stage III and IV (OR = 1.49, 95% CI: 1.09–2.04; P = 0.01) and international prognostic index (IPI) score of 3 to 5 points (OR = 1.79, 95% CI: 1.26–2.56; P = 0.001). However, there was no significant difference in the subgroups of gender and age (Figure 4).
Subgroup and sensitivity analysis
Subgroup analyses were conducted by tumor type, country, sample size, cut-off value, therapy, antibody source, and type. Subgroup analysis by country showed HR of 2.86 (95% CI: 1.44–5.66; P = 0.003) in China, 1.99 (95% CI: 1.29–3.08; P = 0.002) in Japan, and 0.47 (95% CI: 0.29–0.77; P = 0.002) in Korea. In addition, when cut-off value ≥ 30%, HR was 2.54 (95% CI: 1.56–4.12; P < 0.001) with I2 = 37% (Table 2). Sensitivity analyses demonstrated that our pooled results were robust even when omitting anyone of the included studies by turn in NHL and DLBCL (Figures 5 and 6).
Meta-regression analysis
Furthermore, meta-regression was performed for the source of heterogeneity in NHL. The results showed that sample size (P = 0.638), treatment (P = 0.229), location (P = 0.107), tumor type (P = 0.916), and cut-off value (P = 0.058) did not contribute to the heterogeneity.
Publication bias
Begg’s test was used to assess the publication bias, which revealed no publication bias for either NHL (P = 0.880) nor DLBCL (P = 0.920).