Prognostic value of programmed death ligand 1 (PD-L1) in glioblastoma: a systematic review, meta-analysis and validation based on dataset

Figures & data

Table 1. Characteristics of the included studies

Ref	country	patient	index(assay)	cut off	total patients	treatment(method, No.)	PD-L1(±)	source of HR	HR(95%CI)	NOS
27	Denmark	GBM	protein(immunoflu-orescence)	percentage ≥10%	17	S + R, 2S + C, 15	6/11	K-M	1.50 (1.05–2.14)	7
30	Austria	GBM	protein(IHC)	percentage ≥5%	117	S + R + C, 102unknow,15	44/73	K-M	1.18 (0.75–1.87)	8
23	China	GBM	protein(IHC)	percentage ≥5%	62	NR	33/29	K-M	1.32 (0.96–1.82)	7
25	Korea	GBM	protein(IHC)	percentage ≥5%	54	S, 2S + R / S + C, 9S + R + C, 43	17/37	Cox regression	3.06 (1.16–8.06), UA	8
29	Korea	GBM	protein(IHC)	percentage ≥5%	115	S + R + C, 93S + R, 12S, 10	37/78	Cox regression	1.79 (1.10–2.91), UA	8
28	China	GBM	protein(IHC)	percentage ≥5%	20	S, 10S + C, 3S + R, 1S + R + C, 6	9/11	Cox regression	0.65 (0.25–1.69),MA	8
31	America	rGBM	protein(IHC)	percentage ≥5%	60	NR	22/38	Cox regression	1.96 (1.11–3.45), MA	6
21	China	GBM	protein(IHC)	score >4	47	NR	29/18	K-M	1.31 (1.00–1.72)	8
22	Denmark	GBM	protein(immunoflu-orescence)	expression leve ≥50%	163	NR	81/82	Cox regression	0.89(0.66–1.23), UA	7
30	Austria	GBM	mRNA(RNA-Seq)	expression level≥50%	446	NR	223/223	Cox regression	1.14 (0.96–1.37), UA	8
26	America	GBM	mRNA(Microarray)	expression level≥37%	152	NR	56/96	Cox regression	1.54 (1.05–2.28),UA	8
20	Japan	GBM	mRNA(RNA-Seq)	expression level≥50%	158	NR	79/79	K-M	1.07 (0.88–1.29)	7
31	America	rGBM	mRNA(RNA In Situ Hybridization)	expression leve ≥50%	60	NR	30/30	Cox regression	0.84 (0.50–1.41), MA	6
24	China	GBM	mRNA(RNA-Seq)	NR	152	NR	NR	Cox regression	1.08 (1.01–1.16), UA	6
24	China	GBM	mRNA(RNA-Seq)	NR	214	NR	NR	Cox regression	1.00 (0.94–1.07), UA	6

GBM, glioblastoma; rGBM, recurrent glioblastoma; Ref, reference; IHC, immunohistochemistry; RNA-seq, RNA sequencing data; NR, not report; MA, multivariate analysis; UA, univariate analysis; K–M, Kaplan–Meier curve; NOS, Newcastle–Ottawa Scale. S, surgery; R, radiotherapy; C, chemotherapy.

Figure 1. The flowchart of article selection

Figure 2. The primary meta-analysis. The relationship between overall survival (OS) and programmed death ligand 1 (PD-L1) expression. a) PD-L1 protein expression b) PD-L1 messenger ribonucleic acid (mRNA) expression c). The pooled hazard ratio (HR) of all included studies is 1.16 (95% CI, 1.11–1.46; P= 0.001); high heterogeneity is observed across studies (I² = 56.7%, P= 0.002). After dividing all included studies into two subgroups based on detection index (protein and mRNA); no high heterogeneity is observed within the two subgroups. The results reveal that PD-L1 expression is associated with a poor OS of glioblastoma (GBM) at protein level (HR, 1.30; 95% CI, 1.13–1.48; P< 0.001) and mRNA level (HR, 1.05; 95% CI, 1.00–1.09; P= 0.041)

Table 2. Subgroup analysis for the association between programmed death ligand 1 (PD-L1) expression and overall survival (OS)

Subgroup	Number of studies	Pooled HR(95%CI)	P value	Heterogeneity		Heterogeneity between groups (p value)
				I²	p value
Detection index						0.003
protein	9	1.30 (1.13–1.48)	<0.001	0.473	0.056
mRNA	6	1.05 (1.00–1.09)	0.041	0.4	0.139
Detection method						0.007
immunofluorescence	2	1.12(0.88–1.41)	0.36	0.786	0.031
immunohistochemistry	7	1.39(1.18–1.64)	<0.001	0.269	0.224
RNA-sequencing	4	1.04(1.00–1.09)	0.045	0.172	0.305
other	2	1.24(0.91–1.69)	0.176	0.703	0.066
Race						0.51
asian	8	1.12 (1.01–1.25)	0.032	0.602	0.014
caucasian	7	1.21 (1.00–1.46)	0.049	0.482	0.072
Sample size						0.083
<100	7	1.34 (1.07–1.68)	0.012	0.415	0.114
≥100	8	1.08 (1.00–1.18)	0.05	0.476	0.064
Source of HR						0.038
Kaplan-Meier	5	1.21(1.07–1.38)	0.003	0	0.461
Cox regression	10	1.05(1.01–1.12)	0.021	0.626	0.004
Cutoff						0.001
protein staining percentage	7	1.45 (1.22–1.74)	<0.001	0.244	0.242
mRNA expression level	4	1.13 (1.00–1.27)	0.043	0.255	0.259
other	4	1.04 (0.99–1.09)	0.094	0.523	0.098
Year of publication						0.006
<2019	9	1.22(1.10–1.35)	<0.001	0.376	0.118
≥2019	6	1.04(1.00–1.09)	0.073	0.574	0.039

Heterogeneity between the groups reveals that detection index, detection method, source of hazard ratio (HR), cutoff and year of publication are the sources of significant heterogeneity (P < 0.05); however, race and sample size are not (P > 0.05).

Figure 3. Analysis of publication bias for the protein subgroup. The results reveal that the funnel plot is symmetrical and the P-values of the Egger’s and Begg’s tests are 0.417 and 0.118, respectively. The symmetrical funnel plot, Egger’s test and Begg’s test reveal no publication bias within the protein subgroup

Table 3. Clinical features of the Chinese Glioma Genome Atlas-Glioblastoma (CGGA-GBM) dataset

Clinical characterics		number(percent)
age (median,range)		50 (11-79)
gender	male	67 (62%)
	female	41 (38%)
history ofrelapse	primary	85 (78.7%)
	recurrent	23 (21.3%)
status of IDH	wildtype	89 (82.4%)
	mutant	19 (17.6%)

IDH, isocitrate dehydrogenase.

Figure 4. Clinical significance and prognostic value of PD-L1 mRNA expression based on the Chinese Glioma Genome Atlas-Glioblastoma (CGGA-GBM) dataset. The Y-axis represents the expression of PD-L1 mRNA transformed by log2 (1 + x) function. PD-L1 mRNA expression does not correlate with age a) sex b) history of relapse c) however, it correlates with the status of isocitrate dehydrogenase (IDH), and IDH wildtype GBM has higher PD-L1 mRNA expression level d). The Kaplan–Meier curve indicates that high PD-L1 mRNA expression is associated with a poor overall survival (OS) of GBM (HR, 1.53; P= 0.039) E)

Availability of data and materials

All data are provided in the manuscript and can be obtained from the corresponding author.