Immune checkpoint inhibitors efficacy across solid cancers and the utility of PD-L1 as a biomarker of response: a systematic review and meta-analysis

Abstract

Background

Immune checkpoint inhibitors (ICPI) are a tumor agnostic treatment. However, trials of their use have been site specific. Here we summarize the trial data and explore the utility of programmed death-ligand 1 (PD-L1) expression as a biomarker to direct their pan-cancer use.

Method

A systematic review of literature, following PRISMA guidelines, was performed. Medline, Embase, Cochrane CENTRAL, NHS Health and Technology, and Web of Science were searched from their conception to June 2022 limited to the English language. The search terms and method were devised by a specialist medical librarian. Studies were limited to adults with solid cancers (excluding melanomas) treated with ICPIs. Only phase III randomized control trials (RCT) were included. The primary outcome was overall survival and secondary outcomes were progression free survival, PD-L1 expression, quality of life outcomes and adverse event data. Where present in eligible clinical trials, hazard ratios (HR), risk ratios (RR), standard error (SE) and 95% confidence intervals (CI) were extracted or calculated. Heterogeneity across studies was described with the use of an I² score (Low: 25, 50%: moderate, 75% low heterogeneity). HR pools inverse variance methods were adopted by Random Effects (RE). Means were standardized across any heterogenous scale limits.

Results

In total 46,510 participants were included in the meta-analysis. Overall, meta-analysis favored the use of ICPIs with an overall survival (OS) HR of 0.74 (95% CI 0.71 to 0.78). Lung cancers showed the most benefit in OS [HR 0.72 (95% 0.66–0.78)] followed by head and neck cancers [HR 0.75 (95% CI 0.66–0.84)] and gastroesophageal junction cancers [HR 0.75 (95% CI 0.61–0.92)]. ICPIs seem to be efficacious at both primary presentation and recurrence [OS HR 0.73 (95% CI 0.68–0.77)] vs. [OS HR 0.79 (95% CI 0.72 to 0.87)] respectively. Interestingly, subgroup analysis comparing studies in which most cancers demonstrated PD-L1 expression vs. those studies in which a minority of cancer demonstrated PD-L1 expression reported similar effect of ICPI use on OS; oddly the data favored ICPI use in studies with a minority of PD-L1 expression. Specifically, studies with minority PD-L1 expression had an HR 0.73 (95% CI 0.68–0.78) vs. studies with majority PD-L1 expression HR 0.76 (95% CI 0.70–0.84). This was maintained even when studies exploring the same cancer site were directly compared. Subgroup analysis was performed comparing the impact on OS subdivided by the specific ICPI used. Where meta-analysis was performed, Nivolumab led to the greatest impact [HR 0.70 (95% CI 0.64–0.77)] with Avelumab failing to reach significance [HR 0.93 (95% CI 0.80–1.06)]. However, overall heterogenicity was high (I² = 95%). Finally, the use of ICPIs led to an improved side effect profile when compared with standard chemotherapy [RR 0.85 (95% CI 0.73–0.98)].

Conclusion

ICPIs improve survival outcomes in all cancer types. These effects are seen in the primary, recurrent, chemotherapy sensitive, chemotherapy resistant disease. These data support their use as a tumor agnostic therapy. Furthermore, they are well tolerated. However, PD-L1 as a biomarker for the targeting of ICPI use seems problematic. Other biomarkers such as mismatch repair or tumor mutational burden should be explored in randomized trials. In addition, there are still limited trials looking at ICPI use outside of lung cancer.

Background

Cancer is the second leading cause of death in the world; only behind cardiovascular disease (1). Despite advancements in treatments, the mortality rate for many cancers remains high (2). For patients with advanced cancer, chemotherapy and radiotherapy are the primary treatment options. However, due to the systemic nature of chemotherapy, there have been issues with toxic side effects as well as drug resistance. Targeted therapies are therefore of intrinsic value as they seek to reduce treatment toxicity and resistance (3). One such targeted therapy are immune checkpoint inhibitors (ICPIs).

Immune checkpoint inhibitors are a class of treatment that exploit a common mechanism of cancer immune escape: the programmed death-1/programmed death ligand (PD-1/PD-L1) ligand/receptor interaction (4). Indeed, cancers that arise due to a defective mismatch repair system commonly exploit the PD-1/PD-L1 pathway (5). Cancers which overexpress PD-L1 inhibit cytotoxic T cells (6). These deactivated T cells remain in the tumor microenvironment as they are continuously recruited through the production of cancer related neo-antigens (7). ICPIs are monoclonal antibodies that act to block the PD-1/PD-L1 axis and reverse the induced T cell exhaustion to prevent cancer immune escape (8). They lead to a re-activation of the recruited tumor associated lymphocytes and tumor containment or eradication.

ICPIs have been trialed in numerous cancer sites with generally encouraging results (9). They were the first class of drug to receive approval from the Federal Drug Agency (FDA) based on a molecular characteristic within the tumor (microsatellite instability (MSI) or high mutational burden (HMB)) as opposed to the anatomical cancer site (10, 11). This approval was based on pooled data of single arm cohorts from trials as no meta-analysis existed (12). However to date, most trials have used PD-L1 as a biomarker for ICPI use and not MSI or HMB (13).

The use of PD-L1 as a biomarker for ICPI is based on its use in the initial trials done in melanomas (14). This is despite the original study that explored ICPI use in melanoma reporting that PD-L1 did not predict those in whom ICPI would lead to significant improved survival (15). There remains limited data synthesis as to the utility of PD-L1 as a biomarker of ICPIs effectiveness across all cancer sites. These data are important given the clinical application of ICPIs is not based on cancer site but tumoral biomarkers. This analysis is also prudent given the increasing use of ICPIs across multiple cancer sites, despite a relatively limited evidence base for their use in that specific cancer site (16). If pan-cancer analysis supported their application based on the expression of a biomarker, clinicians could be more confident in trialing ICPIs in these lesser studied cancer types based on the molecular profile of the cancer. In addition, if PD-L1 expression proved to be an accurate predictor of ICPIs treatment efficacy across multiple cancer sites, its sustained use as such a biomarker in trials would become clinically meaningful.

The aim of this study was to perform a systematic review and meta-analysis to synthesis the existing trial data evaluating the ICPIs use in all solid cancer types. Our hope was to provide evidence as to their cross-cancer utility and help inform their current application based on molecular characteristics as opposed to anatomical cancer site. In addition, we will explore the utility of PD-L1 as a biomarker of ICPI treatment efficacy.

Method

Search strategy and study identification

A systematic review of literature, following PRISMA guidelines, was performed (17). Medline, Embase, Cochrane CENTRAL, NHS Health and Technology, and Web of Science were searched from their conception to May 2022. The search terms and method were devised by a specialist medical librarian. In addition, we searched for non-published trial data via www.controlled-trials.com/rct and www.cancer.gov/clinicaltrials. Initial search results were supplemented by citation searching. Non-electronic and grey literature were excluded. The search methods are detailed in Supplementary material section 1.

Selection criteria

The protocol for this systematic review was preregistered with the PROSPERO database registration (ref: CRD420202219410). Only studies published in English were included. Studies were limited to adults with solid cancers (excluding melanomas) treated with ICPIs. Melanoma was excluded due to its exceptionally high expression of PD-L1 and the extensive evidence indicating ICPI therapeutic efficacy along with standardized PD-L1 immunohistochemistry protocols (18). These factors make it a distinct clinical entity that would be problematic to include in a pan-cancer meta-analysis. Only phase III randomized control trials were included. The primary outcome was overall survival with secondary outcomes being progression free survival, PD-L1 expression, quality of life outcomes and adverse event data. Full selection criteria are detailed in Supplementary material section 1.

Data extraction

Titles and abstracts were collated and screened using the Rayyan software¹. Screening was done independently by two authors (TF and NS), with any discrepancies reviewed by a third party (NAJR). Studies that were identified as meeting the inclusion criteria underwent full paper review which was conducted by two authors (TF and NS), with issues resolved through discussion and consensus with a senior author (NAJR) who made the final decision. A bespoke data collection tool was designed to ensure complete capture of all primary and secondary outcome data points (available on request). Demographic, tumor characteristics, therapeutic and outcome data were collected independently by two authors (TF and NS) and crosschecked (NAJR). The key outcomes are detailed in Supplementary Table S2. In studies in which there were multiple treatment arms, data was only extracted from the relevant arm in which an ICPI has been directly compared to a standard therapy or placebo.

Assessment of bias

Risk of bias was assessed using the International Cochrane Collaborations guidelines. This tool uses a 3-point scale to assess the following: selection, performance, detection, attrition, and reporting (19). Using scores from each of these 6 domains, an overall risk of bias for each trial was calculated. Definition of overall bias was calculated: ‘low’ if 4 or more domains were scored as low; ‘high’ if 3 or more domains were scored as high; ‘medium’ if neither of the above.

PD-L1 status

Studies were grouped into two groups; namely those in which the majority of tumors demonstrated high PD-L1 expression and those in which minority of tumors demonstrated high PD-L1 expression. The definition of high PD-L1 expression was taken from the authors’ definition. A majority was defined as >50% of the study population treated with ICPIs had tumors with high PD-L1 expression. In addition, the effect of significant PD-L1 expression (again as defined by the authors) was explored in different subgroup analysis.

Statistical analysis

A Priori power analyses were conducted for Random-Effects (RE) models with a conservative OR of 0.85 and between-study heterogeneity gradings at “low,” “medium,” and “high” with study size n-1 the predicted dataset.

Where present in eligible clinical trials, hazard ratios (HR), risk ratios (RR), standard error (SE) and 95% confidence intervals (CI) were extracted or calculated from source data. Means were standardized across any heterogenous scale limits. See Supplementary material for more detail.

All analysis was performed using R version 4.1.0² with the following libraries: tidyverse, meta version 5.0, metafor, and dmetar (20–22).

Results

Search results

Our search of the medical databases yielded 3,567 articles. In addition, searching the trials registry resulted in 695 additional studies. Therefore, after the removal of duplicates (n = 258), 4,004 studies underwent initial abstract screening. Four further studies were identified through citation searching. In total 81 papers underwent full manuscript review of which 44 were excluded (see Supplementary Table S4). Therefore 37 studies (23–59) were included in this systematic review and meta-analysis. This process is summarized in Figure 1.

Figure 1.

Prisma flow diagram.

Study characteristics

In total 46,510 participants were included in the meta-analysis. Four studies (31, 53, 54, 56) were found to have no risk of bias (see Supplementary Figure S1). Included studies were conducted in North America (n = 25), China (n = 5), Europe (n = 4), and Japan (n = 3). These studied lung (n = 18), gastro-esophageal junction (n = 6), head and neck (n = 3), breast (n = 3), uroepithelial (n = 2), ovarian (n = 2), liver (n = 1), mesothelioma (n = 1), and renal cell (n = 1) cancers. Studies were either in the primary (n = 15) or relapsed (n = 22) setting. The majority (n = 27) reported the number of tumors with PD1/PD-L1 expression. A range of ICPIs were used: Pembrolizumab (n = 16), Nivolumab (n = 11), Atezolizumab (n = 6) or Avelumab (n = 3) Durvalumab (n = 1). Most controls were non-ICPI agents, however five studies had a sole placebo control arm. These data are summarized in Table 1. Results detailing analysis power, small study effects, and influential studies can be found in the Supplementary material section 1.4.

Table 1.

Included studies characteristics.

Author	Study year	Country	Type of cancer	Number of participants	Intervention	Control	Randomization ratio	Primary or relapsed disease	Number of patients with PD1/PDL1 tumors
Gandhi et al.	2018	United States	NSCLC	616	Pembrolizumab	Placebo	2:1	Primary	388
Borghaei et al.	2015	USA	NSCLC	792	Nivolumab	Docetaxel	1:1	Relapsed	246
Zhou et al.	2017	China	NSCLC	425	Pembrolizumab	Docetaxel	1:1	Relapsed	227
Mok et al.	2019	USA	NSCLC	1,274	Pembrolizumab	Carboplatin & Paclitaxel OR Pemetrexed	1:1	Primary	1,274
Reck et al.	2016	China	NSCLC	305	Pembrolizumab	Carboplatin and Paclitaxel OR Gemcitabine and Cisplatin OR Gemcitabine & Carboplatin	1:1	Primary	205
Paz-Ares et al.	2018	Spain	NSCLC	559	Pembrolizumab	Placebo and Paclitaxel OR Nab-paclitaxel	1:1	Primary	353
Vokes et al.	2018	USA	NSCLC	352	Nivolumab	Docetaxel	1:1	Relapsed	42
Barlesi et al.	2018	Germany	NSCLC	792	Avelumab	Docetaxel	1:1	Primary	529
Rittmeyer et al.	2017	USA	NSCLC	1,255	Atezolizumab	Docetaxel	1:1	Primary	463
Carbone et al.	2017	USA	NSCLC	1,325	Nivolumab	Gemcitabine and Paclitaxel OR Pemetrexed	1:1	Relapsed	423
Wu et al.	2019	China	NSCLC	639	Nivolumab	Docetaxel	2:1	Relapsed	252
Pujade-Lauarine et al.	2021	USA	Ovarian	556	Avelumab OR Avelumab and Pegylated Liposomal Doxorubicin	Pegylated Liposomal Doxorubicin	1:1	Relapsed	288
Emens et al.	2021	USA	Breast	902	Atezolizumab and Nab-Paclitaxel	Nab-Paclitaxel and Placebo	1:1	Primary	369
Owonikoko et al.	2021	USA	SCLC	555	Nivolumab	Placebo	1:1	Primary	Unknown
Miles et al.	2021	USA	Breast	651	Atezolizumab and Paclitaxel	Paclitaxel and Placebo	2:1	Primary	292
Spigel et al.	2021	USA	SCLC	803	Nivolumab	Topotecan OR Amrubicin	1:1	Relapsed	Unknown
Shitara et al.	2020	USA	Gastric/GOJ	763	Pembrolizumab OR Pembrolizumab and Cisplatin OR 5-Fluorouracil OR Capecitabine	Placebo and Cisplatin OR 5-Fluorouracil OR Capecitabine	1:1:1	Primary	763
Shitara et al.	2018	USA	Gastric/GOJ	570	Pembrolizumab	Paclitaxel	1:1	Relapsed	380
Kojima et al.	2020	Japan	Oesophageal/GOJ	628	Pembrolizumab	Paclitaxel, Docetaxel, Irinotecan	1:1	Relapsed	222
Cohen et al.	2019	USA	HNSCC	495	Pembrolizumab	Methotrexate OR Docetaxel OR Cetuximab	1:1	Relapsed	295
Burtness et al.	2019	USA	HNSCC	882	Pembrolizumab OR Pembrolizumab and Cisplatin OR carboplatin and 5-Fluorouracil	EXTREME regimen	1:1:1	Primary	754
Ferris et al.	2016	USA	HNSCC	502	Nivolumab	Methotrexate OR docetaxel OR cetuximab	2:1	Relapsed	Unknown
Kang et al.	2017	Japan	Gastric/GOJ	493	Nivolumab	Placebo	2:1	Relapsed	Unknown
Bang et al.	2018	USA	GOJ	371	Avelumab and BSC	Irinotecan OR Paclitaxel & BSC	1:1	Relapsed	Unknown
Kim et al.	2019	China	Oesophageal/GOJ	123	Pembrolizumab	Paclitaxel OR Docetaxel OR Irinotecan	1:1	Relapsed	54
Fradet et al.	2019	Canada	Urothelial	542	Pembrolizumab	Paclitaxel OR Docetaxel OR Vinflunine	1:1	Relapsed	230
Winer et al.	2021	USA	Triple Negative Breast	622	Pembrolizumab	Capecitabine OR Eribulin, Gemcitabine OR Vinorelbine	1:1	Relapsed	605
Rudin et al.	2020	USA	SCLC	453	Pembrolizumab & Etoposide	Placebo & Etoposide	1:1	Primary	Unknown
Finn et al.	2020	USA	Hepatocellular	413	Pembrolizumab	Placebo and BSC	2:1	Relapsed	Unknown
Powles et al.	2018	United Kingdom	Urothelial Bladder	931	Atezolizumab	Paclitaxel OR Docetaxel OR Vinflunine	1:1	Relapsed	Unknown
Hamanishi et al.	2021	Japan	Ovarian	316	Nivolumab	Gemcitabine OR Pegylated Liposomal Doxorubicin	1:1	Relapsed	123
Jassem et al.	2021	USA	NSCLC	572	Atezolizumab	Cisplatin OR Carboplatin and Pemetrexed OR Gemcitabine	1:1	Primary	554
Motzer et al.	2015	USA	Renal Cell Carcinoma	1,068	Nivolumab	Everolimus	1:1	Relapsed	181
Horn et al.	2018	USA	SCLC	403	Atezolizumab & Carboplatin & Etoposide	Placebo and Carboplatin and Etoposide	1:1	Primary	Unknown
Antonia et al.	2018	USA	NSCLC	713	Durvulumab	Placebo	2:1	Relapsed	Unknown
Fennell et al.	2021	United Kingdom	Mesothelioma	332	Nivolumab	Placebo	2:1	Relapsed	86
Wu et al.	2021	China	NSCLC	262	Pembrolizumab	Carboplatin and Paclitaxel OR Pemetrexed	1:1	Primary	262

GOJ, gastro-oesophageal junction adenocarcinoma; HNSCC, head and neck squamous cell carcinoma; SCLC, small cell lung cancer; NSCLC, non-small cell lung cancer; EXTREME regimen, cetuximab, cisplatin and/or carboplatin, 5-fluorouracil; BSC, best supportive care.

Overall survival

All cancers

In total, 34 studies informed the pan-cancer overall survival (OS) meta-analysis, with four studies (26, 40, 44, 47) contributing data from more than one arm. Overall, meta-analysis favored the use of ICPIs with a HR of 0.74 (95% CI 0.71–0.78). Heterogeneity was high [I² 95% (95% CI 94–96)]. These data are summarized in Figure 2.

Figure 2.

Pan cancer overall survival analysis ICPIs vs Control.

Cancer site

Subgroup analysis was performed by cancer site as to explore its effect on OS. The subgroup analysis included studies investigating lung (n = 15), gastro-esophageal junction (n = 6), breast (n = 3), head and neck (n = 3), ovarian (n = 2), urothelial cancers (n = 2), liver (n = 1), renal (n = 1), and mesothelioma (n = 1). Overall, all cancer sites favored the use of ICPIs [HR 0.78 (95% CI 0.74–0.81)]. However, only lung, head and neck, liver, mesothelioma, renal and gastro-esophageal junction cancers demonstrated a significant benefit in their subgroup analysis. Mesothelioma demonstrated the most significant effect [HR 0.69 (95% 0.60–0.80)] however this was based on one study. Where meta-analysis was possible, lung cancers showed the most benefit [HR 0.72 (95% CI 0.66–0.78)] followed by head and neck cancers [HR 0.75 (95% CI 0.66–0.84)] and gastro-esophageal junction cancers [HR 0.75 (95% CI 0.61 to 0.92)]. These data are summarized in Figure 3.

Figure 3.

Subgroup meta-analysis comparing the efficacy of ICPI use on OS by cancer site.

Looking to lung cancer in more detail, a subgroup analysis was performed comparing the effect of ICPIs in both small cell and non-small disease. This was due to the higher levels of tumor mutational burden (TMB) in small cell cancers versus non-small cancers (11). Both histotypes demonstrated a significant benefit favoring the use of ICPIs [HR 0.73 (CI 95% 0.68–0.78)] however the effect was more pronounced in non-small cell cancers vs. small cell cancers (HR 0.69 vs. 0.77 respectively). These data are summarized in Supplementary Figure S2.

Disease specific characteristics

Immune checkpoint inhibitors seem to be efficacious at both primary presentation and recurrence {[HR 0.73 (95% CI 0.68 to 0.77)] vs. [HR 0.79 (95% CI 0.72 to 0.87)] respectively}. In advanced and metastatic cancers, the use of ICPIs led to an improved OS [HR 0.73 (95% CI 0.68–0.79)]. This was less pronounced in the recurrence setting [HR 0.80 (95% CI 0.72–0.89)] however fewer studies informed this sub meta-analysis when compared to the primary presentation. These data are presented in Supplementary Figure S3. See 1.4 in Supplementary material.

Treatment characteristics

Subgroup analysis was performed comparing the impact on OS subdivided by the specific ICPI used. Where meta-analysis was performed, Nivolumab lead to the greatest impact [HR 0.70 (95% CI 0.64 to 0.77)] with Avelumab failing to reach significance [HR 0.93 (95% CI 0.80 to 1.06)]. However, overall heterogenicity was high (I² = 95%). These data are presented in Figure 4. Furthermore, we explored the impact on OS in studies whereby a single ICPI was used vs. studies in which more than one ICPI was used. The impact was similar in both scenarios [single agent HR: 0.68 (95% CI 0.62–0.75) vs. multiple agent HR: 0.68 (95% CI 0.57–0.81)]. These data are shown in Supplementary Figure S6.

Figure 4.

Subgroup meta-analysis comparting the overall survival grouped by studies that used different ICPI agents. Abbreviations: ATZ: Atezolizumab; AVE: Avelumab; NIV: Nivolumab; PMB: Pembrolizumab.

Study characteristics

The impact on OS was compared in studies deemed to be low bias vs. those studies deemed to have a high bias. There was limited impact on OS [low bias HR 0.72 (95% CI 0.66–0.78) vs. high bias HR 0.76 (95% CI 0.71–0.81)]. However, the low bias subgroup had a lower degree of heterogeneity (I² 77% vs. 94%). These data are presented in Supplementary Figure S7. Furthermore, we explored the impact of a placebo arm. When compared to a placebo or other treatment, ICPI performed well leading to significant improvement in OS HR 0.74 (95% CI 0.71–0.78). This was, as expected, more pronounced in the placebo arm with an of HR 0.67 (95% CI 0.56–0.82) vs. HR 0.75 (95% CI 0.71–0.80). Only a limited number of studies that included a placebo arm (n = 5); these data are summarized in Supplementary Figure S8.

Progression free survival

All cancers

In total 32 studies informed the pan-cancer progression free survival (PFS) analysis. Once more, four studies contributed data from more than one study arm (26, 40, 44, 47). The use of ICPIs did improve PFS across all cancer sites [HR 0.80 (95% CI 0.74–0.87)]. Heterogeneity across studies was high (I² 98%). These data are presented in Figure 5.

Figure 5.

A meta-analysis exploring the impact of ICPIs on progression free survival across all cancer sites.

Cancer site

Studies examining the impact of ICPIs on the PFS of the lung (n = 15), gastro-esophageal junction (n = 6), breast (n = 3), head and neck (n = 2), ovarian (n = 2), liver (n = 1), renal (n = 1), mesothelium (n = 1), and urothelial (n = 1) cancers underwent subgroup analysis to explore the bearing of cancer site on the impact of ICPIs on PFS. Of those in which a meta-analysis was possible, only lung cancers showed a significant improvement in PFS. Looking at lung cancer in more detail, non-small cell studies (n = 11) reported a significant improvement [HR 0.73 (95% CI 0.63–0.85)] in PFS with ICPI use. Studies reporting small cell lung cancers (n = 4) also reached significance [HR 0.76 (95% CI 0.59–0.97)]. These data are shown in Supplementary Figures S9, S10.

Disease specific characteristics

Progression free survival was significantly improved using ICPIs in the primary presentation setting [HR 0.78 (95% CI 0.71–0.86)] however, although there was a trend to improvement in the recurrence setting it did not reach significance [HR 0.85 (95% CI 0.71–1.02)]. These data are outlined in Supplementary Figure S11. PFS was improved with the use of ICPIs in both studies reporting participants who had and had not responded to primary routine treatment {[HR 0.78 (95% CI 0.67–0.89)] and [HR 0.81 (95% CI 0.73–0.91)] respectively – see Supplementary Figure S12}. As seen in OS, the degree of PD-L1 expression did not greatly impact on PFS {[majority PD-L1 HR 0.83 (95% CI 0.71–0.97) vs. minority PD-L1 HR 0.78 (95% CI 0.71 to 0.87)] – see Figure S13}. This is explored in more detail in Supplementary material section 1.5.

Treatment characteristics

Regarding specific ICPI use, the PFS mirrored OS with Nivolumab demonstrating the greatest impact [HR 0.75 (95% CI 0.65–0.87)] and Avelumab failing to reach significance [HR 0.97 (95% CI 0.57–1.67)] as shown in Supplementary Figure S14. PFS only showed a significant improvement in the meta-analysis single agent studies [HR 0.78 (95% CI 0.71–0.86)]. In combination ICPI therapy there was a trend toward improved PFS, however this failed to reach significance [HR 0.86 (95% 0.70–1.06)]. Single agent studies had a higher heterogenicity (I² 98%) compared with studies with that explored multiple agent use (I² 45%). These data are shown in Supplementary Figure S15.

Study characteristics

Subgroup analysis demonstrated that overall PFS was improved by ICPI use in studies judged to be high and low bias [HR 0.80 (95% CI 0.74–0.86)]. However, low bias studies demonstrated a greater effect (HR 0.73 vs. 0.85) but with a higher heterogenicity (I² 99% vs. 77%). These data are summarized in Supplementary Figure S16. Studies that compared ICPIs with placebo found a significant improvement in PFS [HR 0.65 (95% CI 0.56–0.76)]. In the meta-analysis of ICPIs compared to standard treatment, a significant but less pronounced improvement was seen [HR 0.83 (95% CI 0.77–0.90)]. Five studies made up the placebo meta-analysis (32, 33, 37, 47, 54). These data are shown in Supplementary Figure S17.

Side effect profile

Overall, a significantly improved side effect profile with ICPI use however was demonstrated [RR 0.80 (95% CI 0.68–0.95)]. A summary meta-analysis is shown in Figure 6 and described in more detail in Supplementary material section 1.4. It should be noted the former side effects were only reported by a few studies.

Figure 6.

Summary forest plot showing the meta-analysis of reported side effects when comparing ICPI use vs standard treatment/placebo in all cancers.

Quality of life outcomes

Too few studies reported quality of life outcomes for a meta-analysis or an informative narrative analysis. Only Pujade-Lauraine et al. made specific reference to quality-of-life outcomes, stating that “treatment related symptom burden was generally similar across all groups” (40).

Discussion

Immune checkpoint inhibitors are considered tumor-agnostic therapies (60). However, to date, their efficacy has only been studied in RCTs in single tumor sites with limited attempts to describe their therapeutic effect across multiple cancer sites. Furthermore, PD-L1 has been accepted as the biomarker of choice for directing IPCI therapy without any pan-cancer level analysis. To the authors’ knowledge, here we present the most comprehensive review of ICPI use as a tumor-agnostic treatment. We found the use of ICPIs lead to improved survival outcomes across numerous tumor sites with limited toxicity. This was most pronounced in lung, head and neck and gastro-esophageal junction cancers. Our data would suggest the use of ICPI is most effective in the primary treatment setting however it was still beneficial in recurrent disease. Furthermore, ICPI are of benefit in those who have, and have not responded to first line treatment. These findings do not seem to be influenced by the number of ICPIs or the specific ICPI that is used. Finally, PD-L1 as a biomarker of ICPI treatment efficacy would seem problematic.

The variation in ICPI efficacy by cancer site that is seen within our data can be explained mechanistically. ICPIs prevents tumor escape through the PD1/PD-L1 axis which is commonly utilized by cancers with a high mutational burden, as these malignancies express high levels of neoantigens that stimulate a cytotoxic immune response (7, 61) We observed an improved ICPI efficacy in lung, head and neck and gastro-esophageal junction cancers which all have a relatively high mutational burden (62). The improved immunotherapeutic effect in non-small cell lung cancer is well documented (63) and was seen in our data; these cancers also have a hypermutated phenotype (62). Furthermore, we noted primary metastatic cancers had a better response ICPIs; again, metastatic disease is known to have a high mutational burden (64).

Not all studies reported a favorable outcome. Bang et al. (23), Carbone et al. (27), and Miles et al. (35) reported no significant improvement in survival. These outliers can be explained by methodological issues within the studies. Bang et al. stained for PD-L1 with several different antibodies. Of note, subgroup analysis found those cases in which 22C3 antibodies were used for defining PD-L1 expression had better survival outcomes (65); this could speak to a methodological issue rather than a true negative result. Carbone et al. suffered from significant treatment cross over with over 60% of their control group receiving an ICPI. Mature survival data from Miles et al. did find that ICPI use was associated with an improved survival (66).

The use of biomarkers to direct treatment is a tenant of personalized therapy. We explored the effect of using PD-L1 as a biomarker for ICPI use. In studies in which the authors had preselected their treatment cohort based on positive PD-L1 expression, an improved OS HR was noted. However, in head-to-head comparison between studies in which most cancers had a high PD-L1 expression vs. those with a minority where PD-L1 positive, our analysis would indicate that using PD-L1 as a ICPI treatment biomarker was of deleterious effect, although this did not reach significance. Given these data are not consistent it may suggest PD-L1 may not be as reliable a predictor of ICPI effectiveness as commonly held. The immunohistochemistry for PD-L1 is known to be difficult and often open to high levels of result inconsistency (67–69). Issues with fixation, antibody binding, clonal expression, and interpretation are established issues (70, 71) The included studies within this meta-analysis used a range of immunohistochemical platforms, antibodies, scoring methods, thresholds and number of pathologists involved in forming a consensus opinion. This therefore reflects the real word situation in which PD-L1 testing and scoring is heterogeneous (72). The authors therefore recommend that PD-L1 is not preferentially used as to direct ICPI treatment as it does not seem to improve survival outcomes. Other biomarkers, with higher levels of test consistency, such as TMB and or MMRd either via immunohistochemistry or MSI testing should be explored as potential biomarkers. Meta-analysis exploring the utility of TMB as a biomarker of ICPI treatment concluded it led to a significant improvement in survival outcomes, however this analysis relied on retrospective trial data (73). Tumor mutational analysis has been used to direct ICPI treatment in trials and has been shown to predict significantly improved survival outcomes (73). We were unable to do this in this meta-analysis due to a lack of study data; only one RCT used tumor mutational burden prospectively in our analysis (39).

To put our work in context, Sun et al. published a systematic review and meta-analysis on ICPI use in advanced and metastatic cancers including 35 studies in adult and non-adult populations (13). They too found potential issues with PD-L1 as a biomarker for directing ICPI use, however they were unable to draw clear conclusions do to methodological issues. Studies included in this analysis included trials with ICPIs and other biological treatments which makes conclusions on ICPI effect difficult. Subgroup analyses within this work relied on small numbers. In addition, this meta-analysis included a large proportion of melanomas (8%); these cancers are known to demonstrate consistently high PD-L1 expression and therefore could impact on the conclusions of PD-L1 as a biomarker of ICPI efficacy in other cancer sites (74). Overall response rate was included in their analysis, and as noted by the authors, such a measure is thought unhelpful in immunotherapy as cancers often demonstrate a pseudo-progression on ICPI as they increase in size secondary to immune infiltrate (75). Finally, the inclusion of phase II trials is problematic as these studies tend to be underpowered to explore survival outcomes and present immature data which add to the heterogeneity of analysis and make summary statistics more problematic (76). Therefore, our work does add to the current knowledgebase given it applies a more robust inclusion/exclusion criteria and more mature survival data.

A strength of our work is it followed PRISMA guidance throughout (17). Our thorough search and screening lead to the inclusion of 37 phase III randomized control trials for our meta-analysis. This has enabled a comprehensive summary of the current evidence and enabled robust conclusions as to the clinical efficacy of ICPIs across numerous tumor sites and populations. Furthermore, we limited subgroup analysis to remain within our published protocol. These subgroups were comprised of large cohorts of study subjects giving reliability to the analysis. In addition, throughout the meta-analysis, multiple sensitivity studies were conducted to ensure the robustness of results.

Our work is not without its limitations. We used study level data as our attempts to contact authors to collect individual level data were unsuccessful. Furthermore, most studies included in this analysis were not free from methodological bias. We recognize the studies included in this meta-analysis are heterogeneous; the effects of this were mitigated by the use random effects modeling. However, meaningful comparisons in heterogeneous data can be difficult. Of note, because of the available studies, there is a predominance of lung cancers, the use of Pembrolizumab and North American populations which could impact on the generalizability of this meta-analysis and its conclusions. In addition, we were unable to include studies without an English translation which introduces selection bias within our study. In addition, there are several studies that have or are due to publish exploring ICPI use in novel cancer sites that had not published at the time of our search; these are therefore not included in this analysis. These factors, individually and combined, decrease the reliability of this analysis. However, even with these limitations, our work remains the most comprehensive meta-analysis of ICPI use in solid cancers published to date.

We did not include melanomas in this study. This was to reduce the confounding effect of their inclusion on this analysis; melanomas have an excellent response to ICPIs which is well established (18). In addition, the staining for PD-L1 has been extensively explored in this cancer meaning that PD-L1 interpretation benefits from established protocols (77, 78). PD-L1 is often expressed at high levels in melanoma which also aids in the interpretation of this biomarker in a way that is not so in other cancer sites (77–79). Therefore, through the exclusion of melanomas, our meta-analysis can better explore the impact of ICPIs in non-melanoma solid cancers which we believe have distinct clinical characteristics. However, we recognize it does mean the findings of this study cannot be applied to melanomas.

Conclusion

In conclusion, herein we present the most comprehensive review of ICPIs as a tumor agnostic therapy. These data confirm that their use improves survival outcomes across a range of tumor sites and are well tolerated. This benefit is seen regardless of specific ICPI used, whether in primary or recurrent disease and where there had and had not been a good response to primary treatment. Of note, the use of PD-L1 immunohistochemistry to direct ICPI use would seem potentially problematic and other biomarkers such as TMB or mismatch repair status should be explored in more depth. In addition, trials should be conducted in cancers that have been so far under investigated.

Data availability statement

The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding authors.

Author contributions

TF, NS, TW and NR contributed to the data acquisition, data interpretation and data analysis. NR drafted the manuscript. All authors contributed to the literature search, study concept and design, and critical revision to the manuscript.

Funding

EC was supported by the NIHR Manchester Biomedical Research Centre (IS-BRC-1215-2007) and an NIHR advanced fellowship (NIHR300650). DE is an NIHR Senior Investigator (NF-SI-0513-10076).

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Glossary

Abbreviations

CI

confidence Interval

FDA

Federal Drug Administration

HMB

high mutational burden

HR

hazard ratio

ICPI

immune checkpoint inhibitors

MSI

microsatellite instability

OS

overall survival

PD-1

programmed death 1

PD-L1

programmed death-ligand 1

PFS

progression free survival

PRISMA

preferred reporting items for systematic reviews and meta-analyses

RCT

randomized control trial

RE

random effects

RR

risk ratio

SE

standard error

Supplementary material

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmed.2023.1192762/full#supplementary-material