Survival outcomes of patients with nonsmall cell lung cancer concomitantly receiving proton pump inhibitors and immune checkpoint inhibitors

Recent evidence suggests that gut microbiota dysbiosis adversely affects the efficacy of immune checkpoint inhibitors (ICIs). Our objective was to investigate the association between concomitant use of proton pump inhibitors (PPIs) and ICIs, and poor prognosis in patients with nonsmall cell lung cancer (NSCLC). We conducted a cohort study using a completely enumerated lung cancer cohort from a nationwide healthcare database in South Korea. We identified 2963 patients treated with ICIs as second‐line or later therapy for stage ≥IIIB NSCLC. PPI use was ascertained within 30‐days before and on the date of ICI initiation, and nonuse was defined as no prescription of PPIs during this period. Using national vital statistics in South Korea, we assessed the risk of all‐cause mortality associated with concomitant PPI use through a propensity score‐matched Cox proportional hazard model. Among 1646 patients included after 1:1 propensity score‐matching, concomitant PPI use was associated with a 28% increased risk of all‐cause mortality, compared to nonuse (adjusted hazard ratio [HR] 1.28; 95% confidence intervals [CIs], 1.13‐1.46). We observed an increased risk when we restricted the analysis to new users of PPI (adjusted HR = 1.64; 95% CI = 1.25‐2.17). Subgroup analysis showed that PPI use was associated with high mortality risk among patients with viral hepatitis (adjusted HR = 2.72; 95% CI = 1.54‐4.78; Pinteraction = .048). Our study indicates that PPI use is associated with poor prognosis in NSCLC patients treated with ICIs. Further prospective studies are required to determine the risk‐benefit balance of concomitant use of PPIs and ICIs.

efficacy remains unknown. Here, the authors investigated associations between concomitant PPI and ICI therapy and poor prognosis among NSCLC patients in South Korea. Propensity score-matched cohort study shows that PPI use is associated with elevated risk of early mortality in ICI-treated NSCLC patients. Increased risk was most evident among patients with viral hepatitis.

| INTRODUCTION
Immune checkpoint inhibitors (ICIs) have emerged as a mainstay in the treatment of advanced nonsmall cell lung cancer (NSCLC); however, their clinical outcomes are highly heterogeneous with only 20% to 30% of patients achieving an objective response. 1 Emerging evidence shows that the gut microbiome modulates tumor response to ICIs, and an altered gut microbiota can negatively affect survival outcomes. 2 Concerns have been raised that the use of proton pump inhibitors (PPIs) may induce gut microbiota dysbiosis and thus negatively impact ICI efficacy by disturbing gastric acidity and targeting bacterial and fungal proton pumps. 3 Alternatively, it has been suggested that PPIs possess anticancer properties. 4,5 The acidic extracellular microenvironment of tumors promotes cancer progression and induces immune escape. 5 It has been hypothesized that PPIs can neutralize the acidity of the tumor microenvironment 4 and enhance the activity of tumorinfiltrating lymphocytes, 6 suggesting the clinical potential of PPIs in T cell-based cancer immunotherapy. A phase II randomized controlled trial (RCT) involving metastatic breast cancer patients showed a significant clinical benefit of concomitant esomeprazole use with chemotherapy in terms of the objective response rate (67.7% vs 46.9%, P = .049), compared to that of chemotherapy without PPIs. 7 In light of these two contradictory hypotheses, there is a high demand for further clinical research. To date, several studies have reported conflicting findings regarding the impact of PPI use on the outcomes of ICI therapy, including positive effects, 8 nonassociation 9,10 and detrimental effects. 11 These inconsistencies may be due to small sample sizes 9,10 and differences in cancer types. [8][9][10][11] Furthermore, the generalizability of the findings of post hoc analyses from RCTs 11 to a real-world setting is uncertain.
Given the lack of robust population-based studies, we analyzed a completely enumerated lung cancer cohort in South Korea to assess if the concomitant use of PPIs with ICIs is associated with reduced survival.

| Study population
We conducted a retrospective cohort study to determine if the concomitant use of PPIs and ICIs is associated with reduced survival among patients treated with ICIs for advanced NSCLC. We identified patients with pathologically confirmed stage IIIB-IV NSCLC who started receiving programmed cell death 1 (PD-1) and PD-L1 inhibitors as second-line or later therapy between 21 August 2017 (when PD-1 inhibitors were first reimbursed) and 31 December 2018. The study ICIs were the PD-1 inhibitors nivolumab, pembrolizumab and atezolizumab. Cohort entry was defined as the date of incident ICI use.
We excluded the following patients: (a) those <18 years of age at cohort entry and (b) those who initiated PD-1 inhibitors after 1 October 2018, to ensure at least 3 months of follow-up ( Figure S1). PPI users and nonusers were further matched in a 1:1 ratio using propensity scores to address the imbalance of potential confounders.

| Exposure
We ascertained exposure to PPIs in both inpatient and outpatient settings using an intention-to-treat approach. Exposure to PPIs was ascertained within 30 days before or on the date of ICI initiation (cohort entry). Patients who received PPIs during this period were classified as PPI users, while those who did not were classified as nonusers. To avoid immortal time bias, patients were followed-up from the cohort entry. In addition to the analysis of the entire cohort with all study population, we restricted other analysis to new PPI users to minimize potential carryover effect.
We applied 180-day new-user washout windows before the exposure ascertainment window ( Figure S2A). The new user analysis included patients who did not receive any PPI treatment during the 180-day washout window. Among these PPI naïve patients, new PPI users were those who received PPIs within 30 days before or on the date of ICI initiation (cohort entry), whereas PPI nonusers were those who did not.

| Outcome
Our study outcome was all-cause mortality, based on data linked to national vital statistics. In the national vital statistics database, all death records are coded based on medical death certificates or police reports that are transmitted to Statistics Korea. The date of death was defined as the date of the event. Each patient was followed up from cohort entry until the occurrence of the outcome of interest (all-cause mortality) or the end of the study period (31 December 2018).

| Potential confounders
We assessed demographic information (age at cohort entry, sex and income level). The use of comedications that may potentially modulate gut microbiota, 13 including antibiotics, metformin, corticosteroids and opioids, was assessed within 30 days before cohort entry. We assessed baseline characteristics within 1 year before cohort entry for patient medical history (including diabetes, cardiovascular disease, stroke, respiratory disease, sepsis, autoimmune disease and dementia; listed in Table S1), other comedications (chemotherapies, antibiotics, immunomodulators [conventional and targeted disease-modifying antirheumatic drugs and immunosuppressants], and opioids [listed in Table S2]) and Charlson comorbidity index. Additionally, we included lifestyle factors using up to 3 years of health screening records from cohort entry, including smoking (never, ex-smoker and current smoker), alcohol consumption

| Statistical analysis
We estimated the propensity scores for receiving PPIs by fitting a multivariable logistic regression model using all predefined covariates assessed before cohort entry. The propensity score was estimated by fitting a logistic regression model including age, sex, income level, smoking status, alcohol consumption, BMI, medical history and medications. PPI users were matched to nonusers in a 1:1 ratio with propensity scores, using the greedy matching macro. Descriptive statistics were used to summarize patient characteristics, with frequencies and percentages for categorical variables and means (SDs) for continuous variables. Potential imbalances in covariates were assessed using the absolute value of standardized difference, with a value ≥0.1 considered as significant.
We used a Cox proportional hazard model to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for mortality risk with the concomitant use of PPIs with ICIs, compared to nonuse, among patients treated with ICIs for NSCLC. The model was adjusted for age, sex, respiratory disease, viral hepatitis, antibiotics and corticosteroid use. These covariates were also used in the propensity score-matched cohort for a doubly robust estimation of the causal effect. We used the Kaplan-Meier method to estimate overall survival and assessed the median survival time. Kaplan-Meier survival curve was used to measure the probability of surviving over a period of time with the consideration of time in each small interval. 14 It involves computing the probabilities of event occurrence at a certain point of time and then multiplying all the prior survival probabilities to compute the final estimate (Supplementary Method S1).
Log-rank test was used to statistically compare the survival functions of two groups (PPI users vs nonusers).

| Subgroup, sensitivity and exploratory analyses
We conducted subgroup analyses using the interaction terms by age group, sex, preexisting respiratory disease, viral hepatitis, ICI subtypes and gene driver mutations (epidermal growth factor receptor [EGFR] and anaplastic lymphoma kinase [ALK]). A subgroup analysis by EGFR and ALK driver mutation was conducted to assess if these mutations altered the observed association between concomitant PPI use with ICIs and reduced survival. EGFR or ALK mutation was defined as prior exposure to the respective targeted therapies. Only those who received EGFR tyrosine kinase inhibitors as first-line treatment were considered as EGFR mutant as these treatments can be prescribed irrespective of the status of EGFR mutations in second-line or later treatment settings. On the contrary, ALK tyrosine kinase inhibitors were only indicated for those with ALK mutations.
We performed three sensitivity analyses. First, we redefined new PPI users, given the great variability in time between studies to restore the gut microbiome after PPI or antibiotic use. 15,16 We applied 30-day and 90-day washout windows before the exposure ascertainment window (30 days before and on the date of ICI initiation [cohort entry]; Figure S2B,C). Second, we redefined the exposure ascertainment period to 180 days before and on cohort entry to explore the long-term impact of PPI use on ICI efficacy. We then assessed the dose-duration response according to cumulative PPI duration (duration ≤ 14 days, 14 days < duration ≤ 30 days and duration > 30 days) and the cumulative defined daily dose of PPI (DDD; dose ≤ 14 DDD, 14 DDD < dose ≤ 30 DDD and dose > 30 DDD). Finally, we conducted a post hoc sensitivity analysis by calculating the E-value to quantify the effect of an unmeasured confounder that would have affected our findings (Supplementary Method S2). 17,18 In the exploratory analysis, we also evaluated the influence of histamine 2 receptor antagonists (H2RAs), which also can theoretically alter and induce the compositional changes of gut microbiota. 19 To investigate this association, an exploratory analysis with three exposure groups (PPI users, H2RA users and nonusers) was conducted. Exposure to either PPIs or H2RAs was assessed within 30 days before and on the cohort entry.
Patients who were exposed to both PPIs and H2RAs during this period were excluded for accurate exposure group assignment. Patients naïve to these drugs were classified as nonusers. Additionally, given that subsequent cancer treatments after the initial ICI treatment may modify survival in NSCLC patients, we conducted stratified analyses by subsequent post-ICI treatment options, including palliative radiation, single-agent chemotherapy and platinum-based doublet chemotherapy.
All statistical analyses were performed using SAS Enterprise software (version 7.1; SAS Institute, Cary, North Carolina). A two-tailed value of P < .05 indicated statistical significance.

| RESULTS
Of the 112 870 patients with lung cancer, we identified 2963 NSCLC patients who used ICI as second-line or later therapy, which comprised 936 concomitant PPI users (31.6%) and 2027 nonusers (68.4%; Figure S3). We included 1646 patients after propensity score-matching at a 1:1 ratio (C-statistic, 0.752). All variables were well-balanced after propensity score-matching (Table 1).  Figure 1A and  Figure 1B and Table 2).
We observed an increased risk of mortality among concomitant PPI users regardless of subgroup ( Figure 2 As indicated by the results of the exploratory analysis, H2RA users had an increased risk of mortality, compared to nonusers of either PPIs or H2RAs ( Figure S4).  20 Alternatively, PPIs may directly suppress the immune system through the exertion of antiinflammatory effects by reducing the secretion of proinflammatory cytokines and adhesion molecules expressed by inflammatory cells. 21  Another observational study involving 224 nivolumab users with NSCLC in the Czech Republic investigated the impact of several concomitant medications on mortality. However, the results were underpowered such that a significant association between PPI use and mortality risk could not be detected (HR = 1.22; 95% CI = 0.72-2.05). 28 The generalizability of the results of previous studies to ICIs as a whole has remained unclear, as the study populations were limited to either atezolizumab or nivolumab users only. 11,[26][27][28] The results of our study indicated that concomitant PPI use had a negative impact on survival in the large real-world study population of PD-1 inhibitor users. In conclusion, concomitant use of PPIs with ICIs indicated a negative prognostic association with survival outcomes in advanced NSCLC patients. Furthermore, PPI use was associated with a high mortality risk in patients with viral hepatitis, which warrants further investigation. However, given the limitation of retrospective data, there remains a need for additional prospective studies to investigate the causality of this association.

ACKNOWLEDGMENT
We sincerely thank National Health Insurance Service for their cooperation in providing access to the database (Data number: NHIS-2020-1-227).

CONFLICT OF INTEREST
Dr. Ju-Young Shin reports receiving grants from the Ministry of

ETHICS STATEMENT
Our study was approved by the Institutional Review Board of Sungkyunkwan University (approval number: SKKU-IRB-2019-011-001).
The requirement for written informed consent was not applicable as our study used anonymized subject data.