The Association between Baseline Proton Pump Inhibitors, Immune Checkpoint Inhibitors, and Chemotherapy: A Systematic Review with Network Meta-Analysis

Simple Summary Proton pump inhibitors (PPIs) are the most commonly used gastric acid suppressants in cancer patients. However, long-term use of PPIs can cause dysbiosis effects disrupting gut microbiota with subsequent impairment of the effectiveness of immune checkpoint inhibitors (ICIs). Our study demonstrates that, in advanced non-small cell lung cancer and urothelial carcinoma, patients receiving ICIs with concomitant PPIs are associated with poorer survival outcomes, when compared not only with those without PPIs but also with patients treated with chemotherapy, implying that PPIs could compromise the effectiveness of ICIs, causing them to be less effective than chemotherapy. As a result, we suggest that clinicians should avoid unnecessary PPI prescription in these patients. Conversely, there is little survival association with PPI in patients with melanoma, renal cell carcinoma, hepatocellular carcinoma, and squamous cell carcinoma of the neck and head. Nevertheless, future high quality prospective studies including more cancer types are warranted. Abstract (1) Although emerging evidence suggests that proton pump inhibitor (PPI)-induced dysbiosis negatively alters treatment response to immune checkpoint inhibitors (ICIs) in cancer patients, no study systematically investigates the association between PPIs, ICIs, and chemotherapy; (2) Cochrane Library, Embase, Medline, and PubMed were searched from inception to 20 May 2022, to identify relevant studies involving patients receiving ICIs or chemotherapy and reporting survival outcome between PPI users and non-users. Survival outcomes included overall survival (OS) and progression-free survival (PFS). Network meta-analyses were performed using random-effects models. p-scores, with a value between 0 and 1, were calculated to quantify the treatment ranking, with a higher score suggesting a higher probability of greater effectiveness. We also conducted pairwise meta-analyses of observational studies to complement our network meta-analysis; (3) We identified 62 studies involving 26,484 patients (PPI = 8834; non-PPI = 17,650), including non-small cell lung cancer (NSCLC), urothelial carcinoma (UC), melanoma, renal cell carcinoma (RCC), hepatocellular carcinoma (HCC), and squamous cell carcinoma (SCC) of the neck and head. Eight post-hoc analyses from 18 randomized–controlled trials were included in our network, which demonstrated that, in advanced NSCLC and UC, patients under ICI treatment with concomitant PPI (p-score: 0.2016) are associated with both poorer OS (HR, 1.49; 95% CI, 1.37 to 1.67) and poorer PFS (HR, 1.41; 95% CI, 1.25 to 1.61) than those without PPIs (p-score: 1.000). Patients under ICI treatment with concomitant PPI also had poorer OS (HR, 1.18; 95% CI, 1.07 to 1.31) and poorer PFS (HR, 1.30; 95% CI, 1.14 to 1.48) in comparison with those receiving chemotherapy (p-score: 0.6664), implying that PPIs may compromise ICI’s effectiveness, making it less effective than chemotherapy. Our pairwise meta-analyses also supported this association. Conversely, PPI has little effect on patients with advanced melanoma, RCC, HCC, and SCC of the neck and head who were treated with ICIs; (4) “PPI-induced dysbiosis” serves as a significant modifier of treatment response in both advanced NSCLC and UC that are treated with ICIs, compromising the effectiveness of ICIs to be less than that of chemotherapy. Thus, clinicians should avoid unnecessary PPI prescription in these patients. “PPI-induced dysbiosis”, on the other hand, does not alter the treatment response to ICIs in advanced melanoma, RCC, HCC, and SCC of the head and neck.


Introduction
Although clinical trials have shown that immune checkpoint inhibitors (ICIs) provide significant efficacy over non-ICIs comparators in many advanced cancers reshaping their therapeutic landscape [1][2][3][4], only a certain proportion of cancer patients have demonstrated a meaningful treatment response to ICIs. Considering metastatic castration-resistant prostate cancer as an example, pembrolizumab, the only FDA approved ICI for prostate cancer, is used predominantly in patients with high microsatellite instability (MSI-H) [5]. Therefore, identifying prognostic predictors and biomarkers for treatment response in patients taking ICIs is crucial, as it helps determine the subgroups of cancer patients who would benefit most from ICIs, improving the precision of therapeutic management. In fact, a variety of clinicopathological features, including age, gender, the Eastern Cooperative Oncology Group (ECOG) performance status, tumor mutation burden, and programmed death ligand 1 (PD-L1) expression, have been discovered as potential modifiers of treatment response in patients treated with ICIs [6][7][8]. Concomitant use of various medications has also been investigated to determine whether they alter the effectiveness of ICIs [9,10]. Notably, PPIs, commonly prescribed gastric acid suppressants for cancer patients, are well-known for their dysbiosis effects that disrupts gut microbiota, which theoretically impairs patients' response to both ICIs and chemotherapy [11,12].
Six meta-analyses [13][14][15][16][17][18] have investigated the survival impacts of concomitant PPI on the effectiveness of ICI, but their results are contradictory. One [14] demonstrated negligible survival influence of PPI on ICI. Another three [15,17,18] alluded that PPI could negatively affect ICI in advanced cancers, while the remaining two syntheses [13,16] implied melanoma patients may derive survival benefits due to lower disease recurrence. Despite these six meta-analyses, there has been little investigation and discussion on the interaction of PPI against ICI and chemotherapy. Although a subgroup analysis of trials of IMvigor 211 [19] and IMpower150 [20] showed that PPIs significantly compromised the benefit of ICIs over chemotherapy, another exploratory analysis of OAK [21] and POPLAR [22] demonstrated insignificant interaction.
Given the significant increased indications of ICIs due to actively broadening cancer types and the extensive use of PPIs, an updated synthesis is urgently warranted. Thus, our study contains two main objectives; firstly, we aim to investigate the association among PPIs, ICIs, and chemotherapy by conducting a network meta-analysis (NMA); secondly, we will update the comparative survival outcomes of PPI users and non-users in patients receiving ICIs and also in patients taking chemotherapy with a wider range of cancers.

Methods
We performed the present systematic review and meta-analysis based on the Cochrane Handbook for Systematic Reviews of Interventions [23] and reported results in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Network Meta-analysis extension statement, and Meta-analysis Of Observational Studies in Epidemiology guidelines ( Supplementary Method 1-3). The study was registered on PROSPERO (CRD 42021258800).

Study Selection
PubMed, Medline, Embase, and Cochrane Library were searched, from inception until 20 May 2022. Three investigators (W.Y.L, Y.C, and Y.C.C) independently identified relevant studies, and discrepancies were addressed by reaching a consensus with senior reviewers (K.Y.C and Y.N.K). Search details are presented in Supplementary Method 4.

Eligibility Criteria
Three predefined criteria for evidence selection were as follows: (1) randomizedcontrolled trials (RCTs), prospective, or retrospective cohort studies; (2) studies involving adult patients aged over 18 with cancers receiving ICIs or chemotherapy without concurrent radiotherapy; (3) studies reporting at least one comparative survival outcome between PPI users and non-users, overall survival (OS), or progression-free survival (PFS) (PPI versus non-PPI users), irrespective of indications.

Data Extraction
Three investigators (H.C.C., E.A, and T.H.W) independently extracted relevant information from eligible articles. Details are available in Supplementary Method 5.

Quality Assessment
Two investigators (T.H.T and K.Y.C) independently completed a critical appraisal of the included literature by using the Cochrane Risk of Bias tool 2.0 for RCTs, and the Risk of Bias in Non-randomised Studies-of Interventions (ROBINS-I) tool for non-RCTs. Any discrepancy was addressed through discussion with the third investigator (Y.N.K).

Assessment of Transitivity Assumption
A pivotal concept of NMA is that patients in the network are jointly randomizable. One plausible assessment of transitivity assumption is to place included trials under scrutiny to examine whether important effect modifiers are similarly distributed throughout the network [24]. We pre-specified age, gender, ECOG performance status, and PDL-1 expression status as effect modifiers since these factors are known prognostic factors for cancer patients.

Main Outcomes and Statistical Analysis
OS and PFS were pooled by obtaining the unadjusted hazard ratio (HR) extracted directly from each reference. When studies did not report the HR but presented Kaplan-Meier survival curves instead, we acquired an estimated HR from the curves through a well-established method [25], i.e., by using a calculation spreadsheet developed by Tierney and colleagues [26]. For baseline effect modifiers, we used weighted mean difference (WMD) and risk ratio (RR) through an inverse variance method to pool continuous and binary characteristics, respectively. When continuous outcomes are not reported as mean with standard difference but instead median with interquartile or range format, which cannot be used for quantitative syntheses, we utilized a well-established and well-validated tool [27] to convert the data to an appropriate format for syntheses. All estimated effects were presented with a 95% confidence interval (CI).
Network and head-to-head meta-analysis were conducted using RStudio with the 'netmeta' and 'meta' package, respectively (Supplementary Method 5). For the NMA, we produced a network graph to illustrate the structure of evidence followed by league tables for summary of frequentist NMA using a random-effect model. Regarding the league table, interventions were ranked by their p-scores with the netrank function; pscores were a value between 0 and 1, with a higher score suggesting a higher proba- bility of greater effectiveness. Forest plots of HRs and 95% CIs were generated with "chemotherapy" as the reference. Inconsistency was evaluated through the netsplit function https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2773396, accessed on 6 June 2022-zoi200800r45 and displayed via heat plots with the netheat command. We also performed sensitivity analyses by excluding trials of potential source of inconsistency, unknown PDL-1, and treatment line. Function 'netmeta::funnel' was used for depicting the comparison-adjusted funnel plot, which is a common method for depicting publication bias in NMA.
For pairwise meta-analysis, the pooled estimate was based on random-effects with the restricted maximum likelihood method due to inevitable between-trial variance. Heterogeneity was assessed using I-square [28], with values of I 2 < 25%, 25% < I 2 <50%, and I 2 > 50% indicating low, moderate, and high heterogeneity, respectively. Pre-specified sensitivity analyses include subgroup analyses based on cancer types as well as treatment line, and exclusion of studies subject to high risk of bias. Determination of statistical significance in these analyses followed common threshold (p < 0.05). Additionally, function 'funnel' and 'metabias' are used for examining publication bias when a head-to-head meta-analysis with more than 10 studies.

Network Meta-Analysis
Eight post hoc analysis [35,36,[44][45][46]53,54,78] of 18 RCTs [19][20][21][22]32,33,39,43,48,51,63,66,71,73,[81][82][83] enrolling advanced cancers (advanced NSCLC, UC, melanoma, CRC, and gastroesophageal adenocarcinoma) are included in the network. Through the visualization of network plot (Figure 2), patients under ICI treatment with concomitant PPI (red node) are associated with poor OS and PFS, compared not only with those under ICI without PPI, but also with those with chemotherapy. Details of NMA are presented in Supplementary  Figures S2 and S3. However, both the netsplit and netheat plots demonstrated significant inconsistency throughout the comparisons (Supplementary Figures S4-S7 Figures S8-S10), with a substantial decrease in inconsistency ( Supplementary Figures S11-S14). According to the net league table with pscores, ICI without PPI is ranked highest, followed by chemotherapy without PPI, ICI with PPI, and chemotherapy with PPI ( Table 2). The funnel plot (Supplementary Figures S15 and S16) demonstrated general symmetry through inspection, which was further supported by the Egger's test (p = 0.71), indicating no publication bias from small study effects.   Network plot for comparative association among PPI, ICI, and chemotherapy in terms of (A) overall survival and (B) progression-free survival. The thickness of the connecting line corresponds to the number of trials among comparators. We specifically highlight the arm ICI with baseline PPI as red node, the red arrow as its significant comparison with ICI, and chemotherapy without baseline PPI to reiterate our main findings. Blue arrows also indicate significant survival association between two nodes. Conversely, gray arrows suggest little association between two arms. PPI, proton pump inhibitor; ICI, immune checkpoint inhibitor; and HR, hazard ratio.

Pairwise Meta-Analysis for ICI cohorts (Figure 3 and Supplementary Results 7)
In patients with NSCLC, the use of PPIs poses a 36% higher risk of death (HR, 1.36; 95% CI, 1.28 to 1.45; I 2 = 0%) and disease progression (HR, 1.36; 95% CI, 1.26 to 1.47; I 2 = 0%) than those without PPIs. The same trend is observed in patients with UC regarding OS (HR, 1.71; 95% CI, 1.49 to 1.96; I 2 = 0%) and PFS (HR, 1.55; 95% CI, 1.37 to 1.75; I 2 = 0%). There is no significant difference in either of the survival outcomes with respect to Figure 2. Network plot for comparative association among PPI, ICI, and chemotherapy in terms of (A) overall survival and (B) progression-free survival. The thickness of the connecting line corresponds to the number of trials among comparators. We specifically highlight the arm ICI with baseline PPI as red node, the red arrow as its significant comparison with ICI, and chemotherapy without baseline PPI to reiterate our main findings. Blue arrows also indicate significant survival association between two nodes. Conversely, gray arrows suggest little association between two arms. PPI, proton pump inhibitor; ICI, immune checkpoint inhibitor; and HR, hazard ratio.

Pairwise Meta-Analysis for ICI Cohorts (Figure 3 and Supplementary Results 7)
In patients with NSCLC, the use of PPIs poses a 36% higher risk of death (HR, 1.36; 95% CI, 1.28 to 1.45; I 2 = 0%) and disease progression (HR, 1.36; 95% CI, 1.26 to 1.47; I 2 = 0%) than those without PPIs. The same trend is observed in patients with UC regarding OS (HR, 1.71; 95% CI, 1.49 to 1.96; I 2 = 0%) and PFS (HR, 1.55; 95% CI, 1.37 to 1.75; I 2 = 0%). There is no significant difference in either of the survival outcomes with respect to advanced melanoma, RCC, HCC, and SCC of the head and neck. Scatters in the funnel plot with Egger's test suggest no publication bias in either OS or PFS (Supplementary Figures S31 and S32). Sensitivity analysis of excluding studies subject to high risk of bias yields the same association with the significant decrease in the statistical heterogeneity ( Supplementary Figures S33 and S34).
advanced melanoma, RCC, HCC, and SCC of the head and neck. Scatters in the funnel plot with Egger's test suggest no publication bias in either OS or PFS ( Supplementary  Figures S31 and S32). Sensitivity analysis of excluding studies subject to high risk of bias yields the same association with the significant decrease in the statistical heterogeneity ( Supplementary Figures S33 and S34). Horizontal lines indicate the 95% CI of each study; diamond, the pooled estimate with 95%; CI, confidential interval; HCC, hepatocellular carcinoma; HR, hazard ratio; NSCLC, non-small cell lung cancer; PPI, proton pump inhibitor; RCC, renal cell carcinoma; SCC, squamous cell carcinoma; and UC, urothelial carcinoma.

Pairwise Meta-Analysis for Chemotherapy Cohorts (Supplementary Results 9)
The use of PPIs is associated with a significant 12% higher all-cause mortality in patients with advanced NSCLC receiving chemotherapy, with a marginally significant higher progression rate ( Supplementary Figures S35 and S36). For GI cancers, using a PPI also confers a significantly higher risk of death and disease progression rate (Figures S35 and S36) but with substantial heterogeneity.
When CRC cancer patients are stratified based on chemotherapeutic regimens, concomitant PPI use negatively affects the effectiveness of these regimens in patients with FU-based regimens ( Supplementary Figures S37 and S38). Conversely, survival outcomes are similar between PPI users and non-users in patients receiving capecitabine-based regimens ( Supplementary Figures S37 and S38). Horizontal lines indicate the 95% CI of each study; diamond, the pooled estimate with 95%; CI, confidential interval; HCC, hepatocellular carcinoma; HR, hazard ratio; NSCLC, non-small cell lung cancer; PPI, proton pump inhibitor; RCC, renal cell carcinoma; SCC, squamous cell carcinoma; and UC, urothelial carcinoma.

Pairwise Meta-Analysis for Chemotherapy Cohorts (Supplementary Results 9)
The use of PPIs is associated with a significant 12% higher all-cause mortality in patients with advanced NSCLC receiving chemotherapy, with a marginally significant higher progression rate (Supplementary Figures S35 and S36). For GI cancers, using a PPI also confers a significantly higher risk of death and disease progression rate (Figures S35 and S36) but with substantial heterogeneity.
When CRC cancer patients are stratified based on chemotherapeutic regimens, concomitant PPI use negatively affects the effectiveness of these regimens in patients with FU-based regimens ( Supplementary Figures S37 and S38). Conversely, survival outcomes are similar between PPI users and non-users in patients receiving capecitabine-based regimens ( Supplementary Figures S37 and S38).

Meta-Analysis Using Adjusted HR (Supplementary Results 10)
Supplementary Table S5 provides adjusted variables for adjusted outcomes and Supplementary Figures S39-S44 provide the results of the meta-analysis using adjusted HR. No significant discrepancy is noted between unadjusted and adjusted outcomes. We fi-nally summarize the meta-analyses findings with both unadjusted and adjusted results in Supplementary Table S6.

Discussion
Although various meta-analyses have investigated the influence of PPIs on the effectiveness of ICIs, no synthesis, to date, has explored the interaction among PPIs, ICIs, and chemotherapy. Our NMA demonstrated that baseline PPI use not only has a negative prognostic influence on advanced cancer patients treated with ICIs but is shown to compromise the effectiveness of ICI, causing it to be even worse than chemotherapy. It is also noteworthy that trials in the network not only included advanced cancer patients with ECOG PS < 2 but explicitly excluded patients with medical conditions warranting long-term antiplatelets or anticoagulants that may require baseline PPI to prevent GI bleeding, such as coronary artery disease and thromboembolism, which reflect high burdens of co-morbidities and which can obviously confound PPI's survival impact. Therefore, there is a low risk of unmeasured confounding bias for our network, although included trials contain no pertinent information on the indications of PPI. However, through the visualization of netheat plots (Supplementary Figures S6 and S7), significant inconsistency exists in the network, with melanoma deemed to be the source of the inconsistency, which decreased substantially after the removal of melanoma trials. This, together with the head-to-head meta-analyses, provides a plausible reason as to why melanoma contributes to the source of inconsistency, as melanoma was shown to be neutrally affected by PPI. Conversely, only advanced NSCLC and UC patients treated with ICIs are negatively affected by PPI. Although effect modifiers were comparably distributed and no significant inconsistency was detected in our NMA after the removal of melanoma trials, conceptual heterogeneity was inevitable as different cancers encompass distinct histopathological features, clinical behaviors, and responses to therapy. Ideally, for the most precise synthesis, every cancer type should possess their own network. However, the number of trials to date are too limited for us to construct an individual network for respective cancer types. We acknowledge this as our major limitation, and the findings of our NMA should be interpreted together with pairwise meta-analyses supplemented by real-world evidence.
On the basis of our syntheses, "PPI-induced dysbiosis" serves as a significant modifier of treatment response to ICIs in both advanced NSCLC and UC, and various basic science studies have already laid a solid foundation for this clinical observation. The higher diversity of gut microbiota correlates with a higher response to ICIs owing to its positive correlation with T-cell numbers and activity [11]. Notably, PPI users were found to have lower diversity, regardless of indications, compared with non-users, with Firmicutes being the most strongly affected species [84]. Moreover, Routy et al. [11] indicated that NSCLC patients responded well to anti-PD1 agents that had an over-presentation of Firmicutes and higher memory T-helper cell reactivity against commensals in the peripheral blood. Another possible explanation for the negative association between the effectiveness of PPIs and ICIs may be related to H. pylori infection. Oster and colleagues [85] unveiled that H. pylori infection negatively altered the response to ICI in a pre-clinical setting. They also demonstrated that NSCLC patients with seropositive H. pylori were associated with significantly lower OS and PFS when compared with those with H. pylori seronegativity. Consequently, PPI could be a potential surrogate marker for H. pylori infection causing the negative association observed in our study. Another study also implied that dysbiosis effects were minimal when regimens were administered in a later-line context [45], due to the development of immunosurveillance evasion in advanced cancers [86]. However, we demonstrate that the detrimental effects of PPIs may be independent of the treatment line on the basis of the prespecified sensitivity analyses.
On the other hand, "PPI-induced dysbiosis" does not seem to not alter the treatment response to ICIs in patients with advanced melanoma, RCC, HCC, and SCC of the head and neck. This result contradicts previous syntheses [13,16] that demonstrated the potential benefit of PPIs for melanoma patients treated with ICIs. There are two feasible explanations for such a distinctive observation. Firstly, only two studies were included in previous syntheses, which underlies the convincing mechanism, as a small body of evidence can underpower the result, giving rise to a false positive association. Secondly, the quality assessment tool used in their syntheses was the Newcastle-Ottawa Scale, which is now considered outdated and unreliable due to the advent of ROBINS-I, which is the goldstandard tool in current evidence-based medicine. When ROBINS-I was used in the appraisal of these two studies, it was discovered that the study [40] contributing to the positive effect of PPI on melanoma was subject to high risk of bias. Note that after excluding this study [40], statistical heterogeneity decreased substantially, as we demonstrated in Supplementary Results 8. In fact, pre-clinical studies have shed much light on potential dysbiosis effects on melanoma. For those treated with ICI, responders were found to have a significantly higher alpha diversity of gut microbiota than non-responders [87,88], and using antibiotics during the treatment window was associated with a shorter PFS [89,90]. However, PPIs were shown to exert a pro-apoptotic effect on melanoma cells through the inhibition of vacuolar ATPase, an ATP-dependent proton pump [91,93], which disturbs pH gradient across melanoma cells, resulting in caspase-dependent cell death [94]. We assumed that these two aforementioned mechanisms counterbalance each other and contribute to neutral survival impacts of PPI on melanoma. For patients with advanced RCC, our result resonates with a large pooled analysis [95] of individual data from clinical trials illustrating that PPIs conferred to negligible survival influence on RCC patients treated with tyrosine kinase inhibitors (TKIs), presumably due to insignificant effects on bioavailability of TKIs. The same neutral association is observed in patients with HCC, and SCC of the head and neck as well. However, such association in these cancers currently lacks a plausible mechanism, and clinicians should keep in mind that it exhibits a false negative due to small number of studies and the small sample size. We anticipate more RCTs or large observational studies of these cancers for larger pooled analysis in the future.
Regarding the chemotherapeutic modalities, they varied in CRC among included trials. We believe that the high variance in therapeutic modalities among CRC trials accounts for the main origins of heterogeneity, not only in pairwise analysis but also in NMA. Nonetheless, our study shows that patients treated with capecitabine-based therapy are not affected by baseline PPIs. Capecitabine is almost completely metabolized to fluorouracil after being absorbed through the GI walls, and it was believed that PPI hampers the solubility, absorption, and distribution of capecitabine by increasing the intragastric pH level [96]. However, in vitro data appear not to support this interaction, as capecitabine does not become ionized until it reaches a pH level of 8.8, when it is poorly absorbed, which is obviously higher than that induced by PPI [97,98]. In contrast, fluorouracil-based agents are negatively affected by concomitant PPI, although with considerable heterogeneity. This appears to arise from complex combinations with other chemotherapeutic agents and target therapy. Unfortunately, to date, scarce pharmacokinetic studies explore the effect of PPIs on oxaliplatin, irinotecan, and target therapy, and our findings suggest that interactions between PPIs and these modalities should be addressed in the future, warranting further clinical and pharmacokinetic investigations.
One caveat of incorporating CRC into the network is that ICIs currently have limited roles in the CRC treatment field. Although a recent trial, KEYNOTE-177 [99], has established the antitumor efficacy of ICIs in treatment-naïve advanced CRC, only patients harboring deficient-MMR/MS-instable tumors are eligible for using ICIs, with the rationale being that a high mutational burden from deficient-MMR/MS-instability results in overexpression of immunogenic antigens and upregulation of immune checkpoint proteins [100]. On the other hand, the role of ICI in advanced CRC with preserved-MMR/MS-stable diseases is still under exploration, with multiple active recruiting trials investigating the combination of ICI and traditional regimens, which is hypothesized to evoke immunogenic responses.
Another limitation to our network is in regard to the PDL-1 status. Although PDL-1 expression is well-balanced across our network, and sensitivity analyses demonstrated no significant impact of PDL-1 status on our NMA, readers should bear in mind that they suffer from three limitations. Firstly, different cancers contain their own exclusive immune profile and present distinctive responses to different ICIs with a varying correlation to the PDL-1 expression. For instance, in POPLAR and OAK trials, advanced NSCLC patients with higher PDL-1 expression treated with atezolizumab derived higher survival benefits over chemotherapy [21,22] Conversely, in advanced UC, the issue of whether ICI has survival superiority over chemotherapy varied among different agents. Pembrolizumab was shown to result in better survival benefits than chemotherapy, [101,102] whereas atezolizumab did not [19]. The role of PDL-1 expression remains undetermined in UC to date, with nivolumab (CheckMate 275) [103], durvalumab [104], and pembrolizumab (KeyNote 045) [101,102] predicting a better survival response rate using PDL-1, unlike atezolizumab (IMvigor 211) which does not. Furthermore, for CRC patients, Checkmate 142 [105] suggested the PDL-1status is not a predictive biomarker, but a mismatch repair/DNA microsatellite instability (MMR/MS) is. Secondly, PDL-1 expression can be evaluated by different methodologies. There are four commonly used immunohistochemistry assays, including 22C3, 28-8, SP142, and SP263, to stain tumors, and these scoring systems were not directly comparable metrics of PDL-1 expression level. Thirdly, it should be noted that studies of unknown PDL-1 status correspond exactly to CRC studies, which are also a source of heterogeneity. Thus, it implies our sensitivity analysis could be potentially biased.
In summary, the present study contains following novelties. Firstly, we investigated the interaction of PPIs against ICIs versus chemotherapy, which has allowed us to gain further insight into the detrimental magnitude of PPIs in affected cancers. Secondly, only NSCLC and UC are negatively affected by PPI-induced dysbiosis and melanoma, whereas RCC, HCC, and SCC do not seem to be affected. Therefore, when cancer patients have indications of using gastric acid suppressants, such as antiemesis for ICI and chemotherapy, peptic ulcers, and reflux esophagitis, H 2 -blockers may be considered given PPI's potential detrimental effects on ICI in certain cancers and on specific chemotherapeutic regimens. On the other hand, several limitations should be addressed in the future. Firstly, constructing respective networks for different cancers would provide greater knowledge of the interaction between PPIs and therapeutic modalities. Secondly, the included patients were mostly ECOG-PS 0-2, which hinders the generalizability to other cancer patients. Thirdly, information on PDL-1 expression is scarce, so we are unable to look further into the magnitude of PPIs' effects on ICI-treated patients with different PDL-1 expressions. Last but not least, there is still inadequate data for probing the impact of PPIs on other types of malignancy, such as prostate cancer, CRC, and HCC, as ICIs are still looking for their niche in these cancers.

Conclusions
"PPI-induced dysbiosis" serves as a significant modifier of treatment response in both advanced NSCLC and UC that are treated with ICIs, compromising the effectiveness of ICIs to be less than that of chemotherapy. Thus, clinicians should avoid unnecessary PPI prescription in these patients. "PPI-induced dysbiosis", on the other hand, does not alter the treatment response to ICIs in advanced melanoma, RCC, HCC, and SCC of the head and neck. Future high quality prospective studies including more cancer types, and more detailed PDL-1 status are warranted. Pairwise meta-analyses using adjusted hazard ratios (HR) (Figures S39-S44); Table S1. Eligibility criteria of trial patients included in NMA; Table S2. Details of uncategorized cancers; Table S3. Effect modifiers across the network; Table S4. Details of proton pump inhibitors; Table S5. Covariates of studies reporting adjusted estimates; Table S6. Summary of findings of our meta-analysis.