INTRODUCTION
Cervical cancer is one of the most frequent cancers in women and the fourth primary cause of cancer-associated fatalities worldwide. In total, 604127 women are diagnosed with cervical cancer every year, which has led to 341831 deaths in 2020 [1]. With the popularization of human papillomavirus-based screening programs and the development of new diagnostic methods and therapies, the incidence and mortality of cervical cancer has declined by more than half over the last 30 years in high-income areas. However, mortality remains high in low-income countries [2]. The mortality rates differ in patients with early-stage cervical cancer, although the prognosis for patients with recurrent or metastatic disease remains poor. Considering chemotherapy resistance in cervical cancer, it is necessary to identify patients at high risk of poor responses and offer more appropriate treatments to improve OS using predictive biomarkers [3]. According to previous studies, several prognostic biomarkers have been identified. However, the lack of specificity and sensitivity in their prediction power prevents these biomarkers from being clinically suitable. Therefore, it is imperative to identify novel predictive indicators to estimate survival outcomes in patients with cervical cancer.
PD-L1 or B7-H1 is the ligand of programmed cell death protein 1 (PD-1), which is expressed on immune cells, such as activated T cells, B cells, dendritic cells, macrophages, and various tumor cells, and is involved in the immune checkpoint pathway [4, 5]. In the normal immune system, PD-L1 expression sustains the homeostasis of the immune reactions, and the PD-1/PD-L1 pathway plays a key role in restricting autoimmunity and in the negative regulation of cytokine production and T lymphocyte proliferation in the case of inflammatory response to infections [6].
PD-L1 is expressed on tumor cells or tumor-infiltrating immune cells (TICs) that bind to PD-1 on T cells and inhibit effector T cells [5, 7, 8]. Suppression of CD8+ T cells markedly decrease the effect of cytotoxicity, allowing cancer cells to avoid immune surveillance of T cells [9]. These findings suggest that PD-L1 could be a potential biomarker for estimating disease progression, prognosis, and therapeutic efficiency. Wang et al. [10] analyzed nine studies of breast cancer and discovered that overexpression of PD-L1 was associated with shorter OS. Accumulating evidence on the prognostic value of PD-L1 has been studied in solid cancers, such as non-small lung cancer, pancreatic cancer, and prostate cancer [11–14]. Since the predictive value of PD-L1 expression in cervical cancer remains controversial, we gathered eligible data and performed a meta-analysis to determine the prognostic and clinical value of PD-L1 in cervical cancer [15, 16].
MATERIAL AND METHODS
Literature search strategy
We searched for relevant studies in electronic databases, including PubMed, Web of Science, Embase, Ovid MEDLINE, and the Cochrane Central Register of Controlled Trials up to 2021. The following Medical Subject Headings (MeSH) terms: “PD-L1” OR “programmed cell death ligand-1” OR “B7-H1” OR “CD274” AND “cervical carcinoma” OR “cervical cancer” OR “cervical tumor” OR “expression” were used in our electronic search. We reviewed relevant review articles, reference lists of published trials, and conference abstracts [American Society of Clinical Oncology (ASCO), Annual Meetings, and the European Cancer Conference (ECCO)] were manually reviewed for potentially eligible studies.
Inclusion and exclusion Criteria
Studies were considered qualified if they met the following criteria: 1) patients were histologically identified as cervical cancer; 2) detection samples of PD-L1 expression were selected before PD-1/PD-L1 checkpoint inhibitor treatment; 3) PD-L1 expression was measured by immunohistochemistry (IHC) staining of tumor cells and tumor-infiltrating lymphocytes (TILs); 4) the studies analyzed and demonstrated a relationship between PD-L1 expression and prognosis (such as OS, progression-free survival [PFS] ); 5) hazard ratio (HR) or relative risks (RR) values could be extracted directly or calculated indirectly through Kaplan– Meier curves; and 6) articles were published in English. Studies that did not meet the inclusion criteria included literature reported in reviews, case reports, animal studies, and letters, and these were excluded. Two investigators (HW and XC) independently checked whether each study met the inclusion criteria, and discrepancies were resolved by judgement from a third, additional reviewer.
Data extraction and quality assessment
The investigators (XW and SX) extracted all information, including author name(s), publication year, study design, country, patient’s demographics, prognostic endpoint, treatment, International Federation of Gynecology and Obstetrics (FIGO) stage, IHC stained cells, and cut-off value of PD-L1 expression. Any disagreements were resolved by consensus. The primary endpoints were OS and PFS.
Quality assessment
Study quality assessment was estimated separately by two researchers (TH and XW) based on the Newcastle-Ottawa Scale (NOS). Total scores ranged from 0 to 9 points; trials considered ‘high quality’ were scored higher than 6.
Statistical analysis
Statistical analyses were performed using Stata statistical software (version 13.0; Stata Corp., College Station, TX, USA). The HRs and 95% CIs were used to evaluate the association between PD-L1 expression and prognosis. RR was used to analyze the relationship between PD-L1 expression and clinicopathological features. Chi-square tests and I2 were used to assess study heterogeneity. The Begg’s funnel plot and Egger linear regression test were used to investigate the possibility of publication bias [17, 18]. Differences were considered statistically significant at p < 0.05.
RESULTS
Literature selection
A total of 1209 relevant references were checked after the initial literature search. Among these, 1096 publications were excluded as 628 were review articles and other ineligible types of references, 165 were non-English language articles, 236 included experiments on non-human species, and 67 articles were duplicates (Fig. 1). The remaining 30 articles were retrieved for a more detailed assessment. After screening the full text, 17 studies were removed because the authors: explored other types of cancers (n = 6), presented insufficient data (n = 6), and did not use immunohistochemistry (n = 5). Finally, 13 studies were included in this meta-analysis [15, 19–30].
Characteristics of patients and studies
Thirteen eligible studies with 1422 patients were included in this study, all of which were published between 2009 and 2020. Four studies were performed in China, three in Japan, two in Korea, one in Canada, one in the USA, one in Brazil, and one in Belgium. Thirteen studies provided information on OS as an endpoint, and eight studies used PFS as the endpoint. Clinical points such as FIGO stage, tumor size, vascular invasion, and lymph node metastasis were also explored to determine the relationship with PD-L1 expression. All the included studies were of high quality and scored over 6. We treated PD-L1 expression in the area of tumor cells and TICs as two different IHC staining areas; thus, we extracted the information as two independent groups, which led to one study being analyzed twice (Tab. 1).
Table 1. Characteristics of including studies |
||||||||
Study |
Country |
No. of patients |
Age |
FIGO stage |
Endpoint |
PD-L1 expression |
NOS score |
|
Area by IHC |
Cut-off value |
|||||||
Karim 2009 [19] |
USA |
115 |
47 (24–87) |
I–II |
OS |
TICs |
> 0% |
9 |
Enwere 2017 [20] |
Canada |
120 |
44 (39–49) |
IB–IVA |
OS, PFS |
Tumor cells |
tAQUA score |
9 |
Kim 2017 [15] |
Kroea |
27 |
46 (36–71) |
IB1–IIA |
OS, PFS |
Tumor cells |
1% |
9 |
Feng 2018[21] |
China |
219 |
49 (26–75) |
I–IV |
OS |
Tumor cells, TICs |
5% |
9 |
Kawachi 2018 [22] |
Japan |
148 |
45 (30–72) |
I–II |
OS |
Tumor cells |
5% |
9 |
Wang 2018 [23] |
China |
90 |
46 (23–71) |
IB1–IIA2 |
OS, PFS |
Tumor cells |
H-score of 100 |
8 |
Grochot 2019 [25] |
Brazil |
155 |
44 |
I–IVB |
OS, PFS |
Tumor cells |
> 0% |
7 |
Chung 2019 [24] |
Korea |
98 |
46 (24–75) |
II–IVB |
OS, PFS |
Tumor cells + TICs |
Combined positive score |
7 |
Taruma 2019 [26] |
Japan |
20 |
50 (32–68) |
III–IV |
OS, PFS |
Tumor cells |
1% |
7 |
Chen 2020 [27] |
China |
222 |
49 (21–75) |
I–II |
OS, DFS |
Tumor cells, TICs |
Tumor cells > 1% TICs > 5% |
8 |
Lijima 2020 [28] |
Belgium |
33 |
N.A |
IIB–IVA |
OS, PFS |
Tumor cells |
1% |
9 |
Miyasaka 2020 [29] |
Japan |
71 |
60 (28–88) |
IB–IVA |
OS, PFS |
Tumor cells |
1% |
8 |
Tsuchiya 2020 [30] |
Japan |
104 |
46 (26–77) |
I–IV |
OS |
Tumor cells, TICs |
Score(tumor cells,0; TICs,3) |
9 |
FIGO — The International Federation of Gynecology and Obstetrics; IHC — immunohistochemistry; NOS — Newcastle-Ottawa Scale; OS — overall survival; PFS — progress-free survival; TICs — tumor-infiltrating immune cells |
Connections between PD-L1 expression and survival indicators in cervical cancer patients
Thirteen studies explored the association between PD-L1 expression and OS. The results showed that high PD-L1 expression predicts poor survival in OS (HR: 1.31; 95% CI 1.03–1.66, P = 0.025), as the heterogeneity was high (I2 = 81.3%, p = 0.00), random effects were chosen (Fig. 2). In the analysis of the association between PD-L1 expression and PFS, the combined effect measures identified no conclusive association between the level of PD-L1 expression and PFS (HR: 0.93; 0.73-1.19, p = 0.57) (Fig. 3).
Subgroup analysis
We performed subgroup analysis of OS based on sample size (> 100 or ≤ 100), race (Asian or non-Asian), IHC staining area (tumor cells, TICs, or tumor cells + TICs), and cut-off values (1%, 5% and others). According to the results, high levels of PD-L1 expression with a sample size of over 100 indicated a shorter OS (HR: 1.51; 95% CI 1.13–2.01). As for race, high level expression of PD-L1 in Asians represented a lower OS (HR: 1.52; 1.14–2.03). Overexpression of PD-L1 in tumor cells (HR: 1.57; 1.29–2.10) and TICs (HR: 1.75; 1.02–2.99) predicted poor OS. High levels of PD-L1 expression (HR: 4.04; 2.58–6.31) showed a lower effect of OS with a cut-off value of 5% (Tab. 2). However, in the subgroup analysis of PFS, PD-L1 expression showed no significant prognostic value in relation to sample size, race, IHC staining area, and cut-off value (Tab. 3).
Table 2. Subgroup analysis of overall survival |
|||||
Subgroup |
HR (95% CI) |
Pz-value |
I2 |
PH-value |
|
Number |
> 100 |
1.51 (1.13, 2.01) |
0.05 |
81% |
0.00 |
≤ 100 |
1.38 (0.71, 2.68) |
0.35 |
60.5% |
0.02 |
|
Race |
Asian |
1.52 (1.14, 2.03) |
0.82 |
0% |
0.00 |
Non-Asian |
0.95 (0.63, 1.44) |
0.00 |
84.5% |
0.83 |
|
IHC area |
Tumor cells |
1.57 (1.19, 2.10) |
0.00 |
77.9% |
0.00 |
TICs |
1.75 (1.02, 2.99) |
0.04 |
76.1% |
0.00 |
|
Tumor cells + TICs |
0.30 (0.15, 0.57) |
0.00 |
– |
– |
|
Cut off value of PD-L1 expression |
other |
0.82 (0.62, 1.10) |
0.18 |
73.8% |
0.00 |
1% |
1.28 (0.47, 3.51) |
0.63 |
0.0% |
0.40 |
|
5% |
4.04 (2.58, 6.31) |
0.00 |
80.9% |
0.00 |
|
HR — hazard ratio; IHC — immunohistochemistry; TICs — tumor-infiltrating immune cells; CI — confidence interval |
Table 3. Subgroup analysis of progress-free survival |
|||||
Subgroup |
HR (95% CI) |
Pz-value |
I2 |
PH-value |
|
Number |
> 100 |
1.01 (0.66, 1.54) |
0.96 |
0% |
0.95 |
≤ 100 |
0.89 (0.65, 1.21) |
0.46 |
89.4% |
0.00 |
|
Race |
Asian |
0.89 (0.65, 1.21) |
0.96 |
89.4% |
0.00 |
Non-Asian |
1.01 (0.66, 1.54) |
0.46 |
0% |
0.95 |
|
ICH area |
Tumor cells |
1.24 (0.91, 1.69) |
0.17 |
84% |
0.00 |
TICs |
1.81 (0.64, 5.12) |
0.26 |
0% |
– |
|
Tumor cells + TICs |
0.42 (0.26, 0.69) |
0.00 |
0% |
– |
|
Cut off value of PD-L1 expression |
other |
1.25 (0.74, 1.43) |
0.98 |
92.8% |
0.00 |
1% |
1.48 (0.47, 4.66) |
0.22 |
37% |
0.19 |
|
HR — hazard ratio; IHC — immunohistochemistry; TICs — tumor- infiltrating immune cells; CI — confidence interval |
Relationship between PD-L1 expression and clinical-pathological characteristics
We investigated tumor size, FIGO stage, lymph node status, and vascular invasion to determine the effect of PD-L1 expression on clinicopathological characteristics. We found no significant associations with these characteristics (Tab. 4).
Table 4. Relations between PD-L1 expression and clinical points |
|||||
Clinical point |
Studies Number |
RR (95% CI) |
PZ |
I2 |
PH |
Tumor size (≥ 4 cm vs < 4 cm) |
4 |
0.99 (0.77, 1.26) |
0.93 |
27.3% |
0.24 |
Lymph nodes |
8 |
1.02(0.82, 1.26) |
0.90 |
35.9% |
0.12 |
FIGO stage |
4 |
0.88(0.73, 1.06) |
0.18 |
76% |
0.02 |
Vascular invasion |
5 |
0.90(0.75, 1.07) |
0.45 |
0% |
0.18 |
FIGO — International Federation of Gynecology and Obstetrics; RR — relative risk; CI — confidence interval |
Publication bias
We assessed publication bias using Begg’s funnel plots and Egger’s linear regression test and found no publication bias in OS (Begg’s p = 0.59, Egger’s p = 0.79) and PFS (Begg’s p = 0.62; Egger’s p = 0.61). The details are shown in Table 5.
Table 5. The publications bias of the study |
||
Overall |
Begg’s p |
Egger’s p (95% CI) |
OS |
0.62 |
0.79 (–3.56, 2.77) |
PFS |
0.71 |
0.61 (–4.9, 1.73) |
Clinical points |
||
tumor size |
0.22 |
0.19 (–4.14, 1.27) |
vascular invasion |
1 |
0.93 (–3.79, 3.55) |
FIGO |
0.22 |
0.23 (–4.27, 12.02) |
Lymph nodes |
0.47 |
0.20 (–5.17, 1.27) |
OS — overall survival; PFS — progress-free survival; FIGO — International Federation of Gynecology and Obstetrics |
DISCUSSION
This study focused on the prognostic value of PD-L1 expression in cervical cancer. We updated the data and analyzed 13 studies with 1422 patients to identify the relationship between PD-L1 expression and survival. Our findings demonstrated that high levels of PD-L1 expression in cervical cancer were associated with poor OS survival. Moreover, in subgroup analysis, a high level of PD-L1 was associated with shorter OS in terms of race, sample size, IHC staining area, and cut-off value. According to our results, no association existed between PD-L1 expression and PFS, including estimates explored in the subgroup analysis.
Similar results were obtained in a previous study, which demonstrated that overexpression of PD-L1 had a prognostic value of lower OS [31]. However, we obtained different results in subgroup analysis, such that high PD-L1 expression indicated poor OS with sample size over 100, and we did not find that PD-L1 was a prognostic factor of PFS among Asians. Moreover, we analyzed PD-L1 expression in the IHC staining area and the connection between PD-L1 expression and cut-off values, which were not mentioned in the previous study. Overexpression of PD-L1 in tumor cells and TICs predicted a poor effect of OS, but in the mixture of TICs and tumor cells, PD-L1 expression indicated favorable results for OS. We deduced that the differences between the two studies may be due to several reasons, the first one being sample size. We included 13 studies with 1422 patients in this research; however, the previous research only included seven studies. Larger samples may offer more evidence to prove the prognostic value of PD-L1. The second determinant may be due to the method of data extraction, as we treated PD-L1 expression in tumor cells and TICs as two different datasets. Thus, we extracted and analyzed the information as two different groups and that would result in one study being analyzed twice.
As immune checkpoint inhibitors are a hot spot in cancer therapy, an increasing number of studies are focusing on the treatment of anti PD-1/PD-L1 antibodies in cervical cancer [26, 32–38]. However, how to make immune checkpoint inhibitors more efficient is still a problem because among patients, the same therapy strategy may have different effects. Thus, it is valuable to determine the characteristics of patients with respect to PD-L1 expression. The predictive value of PD-L1 expression has been shown in other types of cancers, such as lung cancer, gastric cancer, and colorectal cancer [39, 40]. Zhang et al. [41] analyzed the association between PD-L1 expression and gynecological cancers and found that a high level of PD-L1 expression had a negative effect on OS and was not significantly associated with PFS.
Our study has several limitations. Firstly, although we performed the subgroup analysis and analyzed the data using a fixed model, we still did not find the source of heterogeneity. Given the incomplete dataset, we did not conduct further research. Secondly, as we extracted the value of HR only from Kaplan-Meier curves, different methods or software for reading the graph may produce slightly different results. Thirdly, we limited our inclusion criteria to select articles only published in English and excluded all other non-English, written literature. Thus, considering the limitations mentioned above, the results should be interpreted carefully.
CONCLUSIONS
In conclusion, this study indicated that high expression of PD-L1 is associated with poor OS and has no significant relationship with PFS in cervical cancer. These findings indicate that PD-L1 may potentially serve as a valuable prognostic indicator of cervical cancer.
Authors contributions
XW contributed to conception and design of the research. TH contributed to data collection, and data analysis. HW contributed to data analysis. XC contributed to manuscript writing. SX and XW contributed to revising the manuscript critically for important intellectual content. All authors have read and approved the manuscript.
Acknowledgement
We would like to thank Editage (www.editage.cn) for English language editing.
Conflict of interest
The authors declare no conflict of interests.