Proton pump inhibitors and the risk of acute kidney injury in cancer patients receiving immune checkpoint inhibitors: A Danish population‐based cohort study

Previous studies have suggested that the use of proton pump inhibitors (PPIs) more than doubles the risk of acute kidney injury (AKI) in cancer patients receiving immune checkpoint inhibitors (ICIs). However, this association may be confounded. Therefore, we conducted a register‐based cohort study to examine the risk of AKI in users and nonusers of PPIs among cancer patients treated with ICIs in Denmark from 2011 through 2021 while accounting for a comprehensive range of potential confounders. PPI use was determined based on redeemed prescriptions of PPIs before ICI initiation. We identified laboratory‐recorded AKI events within the first year after ICI initiation. We estimated the risks and hazard ratios (HRs) of AKI while accounting for a comprehensive range of confounders (including comorbidities and comedication) by propensity score weighting. Furthermore, we performed an additional per‐protocol analysis while accounting for informative censoring by weighting. We identified 10 200 cancer patients including 2749 (27%) users, 6214 (61%) nonusers, and 1237 (12%) former users of PPIs. PPI users had an increased risk of AKI compared to nonusers (1‐year risk, 24.7% vs 19.9%; HR, 1.42 [95% confidence interval (CI), 1.29‐1.56]); however, this association attenuated when accounting for confounders (weighted 1‐year risk, 24.2% vs 23.8%; weighted HR, 1.06 [95% CI, 0.93‐1.21]). In the per‐protocol analysis, the crude HR was 1.86 (95% CI, 1.63‐2.12), while the weighted HR was 1.24 (95% CI, 1.03‐1.49). Thus, the association between PPI use and AKI could largely be explained by confounding, suggesting that previous studies may have overestimated the association.


| BACKGROUND
Immune checkpoint inhibitors (ICIs) are a type of cancer immunotherapy that inhibits checkpoint proteins on cancer cells from binding to T-cells, thereby inhibiting the immune tolerance of the cancer cells.In 2011, the first ICI was approved by the United States Food and Drug Administration, and since then, ICIs have been increasingly used in several types of advanced cancers including lung cancer, malignant melanoma, and urologic cancers due to their evident survival benefits. 1,2However, ICIs are associated with a wide range of immunerelated adverse events including acute kidney injury (AKI) from acute interstitial nephritis. 36][7][8][9][10][11][12] Moreover, AKI in patients receiving ICIs is associated with a more than 2-fold increased mortality. 44][15][16][17][18] Recent studies have found a more than 2-fold increased risk/odds of AKI in patients treated with ICIs, 6,11,[19][20][21] leading to the recommendation of prompt cessation of ICIs and PPIs in the management of stage 2-3 AKI. 22However, the studies were limited by incomplete control for potential confounders including confounding by the underlying indication for PPI treatment 6,11,12,[19][20][21] and information bias due to misclassification of exposure. 20erefore, we conducted a population-based cohort study to estimate and compare the risk of AKI in users and nonusers of PPIs among cancer patients receiving ICIs while accounting for a comprehensive range of potential confounders.

| Setting and data sources
We conducted a nationwide cohort study using data from the Danish medical and administrative databases.The Danish National Health Service provides tax-supported health care to the Danish population (5.9 million inhabitants), ensuring free access to general practitioners and hospitals.A unique personal identifier is assigned to all Danish residents, enabling unambiguous individual linkage between registries. 23

| Study population
We identified adult patients (aged ≥18 years) with at least one administration of ICIs during 2011-2021, and with a diagnosis of primary cancer or cancer of unknown primary before the first ICI administration recorded in the Danish National Patient Registry (patient registry; see Table S1 for codes).The patient registry contains administrative and clinical data, including information on diagnoses, surgeries, and other procedures, for example, anticancer drug treatment and radiotherapy. 24The index date was defined as the date of the first administration of ICIs.Data on plasma creatinine measurements were retrieved from the population-based Danish laboratory databases, covering the inpatient, emergency room, outpatient clinic, and general practice setting. 25,26[27] Furthermore, to increase the specificity of the outcome, we excluded patients with AKI within 7 days before or at the index date (see the definition of AKI below), patients receiving chronic dialysis within 10 years before the index date as recorded in the patient registry, and patients with kidney failure (defined by a baseline estimated glomerular filtration rate [eGFR] below 15 mL/min/1.73m 2 ).We calculated eGFR based on plasma creatinine measurements using the 2009 Chronic Kidney Disease Epidemiology Collaboration equation assuming non-black race.Baseline eGFR was defined as the median of eGFR values recorded within 90 days before the index date.

| Exposure
We obtained data on redeemed prescriptions from the Danish National Prescription Registry (prescription registry).The prescription registry contains information on all prescriptions redeemed at community pharmacies. 28I users were defined as patients with at least one redeemed prescription of PPIs within 90 days before or at the index date, while PPI nonusers were defined as patients without any redeemed prescription of PPIs within 365 days before or at the index date (see Figure 1 for illustration).To test the robustness of our findings, we defined groups of new, long-term, and former users.New users were defined as patients with at least one redeemed prescription of PPIs within 90 days before or at the index date but without any redeemed prescriptions 365 to 91 days before the index date.Long-term users were defined as patients with at least one redeemed prescription of PPIs both 365 to 91 days before the index date and within 90 days before or at the index date.Former users were defined as patients without any redeemed prescriptions within 90 days before or at the index date but with at least one prescription redeemed 365 to 91 days before the index date.

| Outcome
We used plasma creatinine measurements from the laboratory databases to assess AKI events within 1 year after the index date.In alignment with guidelines from Kidney Disease: Improving Global Outcomes, we defined AKI as an increase in plasma creatinine of 26.5 μmol/L or more within 48 hours, a 1.5-fold or more increase in plasma creatinine within 7 days, or a 1.5-fold or more increase in plasma creatinine compared to the most recent creatinine measurement within the prior 8 to 365 days.

| Covariates
We obtained data on variables potentially associated with both PPI use and the risk of AKI (see Table 1 for variables included and Table S1 for coding appendix for codes).Information on sex and age at index date was retrieved from the Civil Registration System. 29In case of more than one type of cancer, we included only the most recent cancer before the first ICI administration.Prescriptions for comedication redeemed within 90 days before or at the index date were obtained from the prescription registry.The number of admissions within 90 days before or at the index date (0, 1, 2, or >2) and comorbidities within 10 years before or at the index date were obtained from the patient registry.Furthermore, we included information on whether the patient had undergone any surgery, received non-ICI anticancer drug treatment or radiotherapy within 90 days before or at the index date, and intensive care unit admissions and use of palliative care within 1 year before or at the index date from the patient registry.Finally, we obtained information on baseline eGFR from the laboratory databases.

| Statistical analysis
We followed patients from the index date until the occurrence of AKI, death, emigration (obtained from the Civil Registration System), 1 year after the index date, or at the end of the study period (end of 2021), whichever came first.
We computed propensity scores (PS) for exposure to PPIs using a logistic regression model including all covariates listed in Table 1.We used the PS for standardized mortality ratio (SMR) weighting to balance the covariate distribution among exposed and unexposed.We trimmed patients above the 97.5th percentile of the PS observed among the  nonusers and subsequently reestimated PS to increase the covariate balance in the remaining patients.We reported covariate balance before and after SMR weighting including absolute standardized mean differences (ASMD).We estimated the crude and SMR-weighted 1-year risks of AKI and plotted the SMR-weighted risks of AKI using the Aalen-Johansen estimator considering death as a competing event.
We estimated the crude and SMR-weighted hazard ratios (HRs) of AKI within 1 year after the index date comparing PPI users with nonusers using Cox proportional hazards analysis.The 95% confidence intervals (CIs) were obtained through robust variance estimators.The assumptions of proportional hazards were checked using log minus log plots.

| Additional analyses
We repeated the analysis in new, long-term, and former users of PPIs, respectively.
Furthermore, we repeated the analysis while defining AKI as a 2-fold or more increase in plasma creatinine and while restricting to AKI events that persisted >48 hours (see Supplemental Methods for description).
To mitigate the effect of misclassification due to switch in exposure status during follow-up, we performed a per-protocol analysis where we censored PPI users if they discontinued PPI treatment

| RESULTS
We identified 10 461 cancer patients treated with ICIs, of whom 10 200 met the study criteria and were eligible for analysis (see Figure 2).Among the included patients, 2749 (27%) redeemed a prescription of PPIs within 90 days before or on the day of ICI initiation and were accordingly considered users, while 6214 (61%) did not redeem a prescription for PPIs within 365 days before or on the day of ICI initiation and were accordingly considered nonusers.

| Descriptive characteristics
Patients' characteristics at the time of ICI initiation are shown in After SMR weighting of the nonusers, all measured covariates were well-balanced between users and nonusers with ASMD <0.10 for all covariates (Table 1).

| Acute kidney injury
During the first year after the index date, we observed 624 cases of AKI in PPI users and 1125 in nonusers based on a median of 18 creatinine measurements (interquartile range [IQR], 11-26) in PPI users and a median of 18 creatinine measurements (IQR, 10-25) in nonusers.
Subgroup analyses by cancer type, ICI type, and eGFR showed that the 1-year risks of AKI varied between the groups, with the highest risk among patients with eGFR <60 mL/min/1.73m 2 and in patients treated with a combination of CTLA-4 and PD-L1 inhibitors (Table 2).However, the HRs associated with PPI use were largely similar to those in the main analysis, except for the small subgroup

| DISCUSSION
Our data revealed a high risk of AKI in the first year after ICI initiation, especially in PPI users.However, when accounting for confounders, the risks of AKI were similar between patients using PPIs at ICI initiation and those who were not.The same was observed when analyzing new, long-term, and former users of PPIs.When we restricted the outcome to AKI defined by a 2-fold or more increase in plasma creatinine and when restricting to AKI persisting >48 hours, we observed lower absolute risks of AKI; however, the HRs between PPI users and nonusers remained at the same level.Among the subgroup of patients receiving CTLA-4 inhibitors, we did identify a 1.55-fold increased risk of AKI in users of PPIs compared to nonusers, though with low precision.Whether this association is causal or could be explained by residual confounding remains to be clarified.In the per-protocol analysis, PPI users had an increased risk of AKI compared to PPI nonusers.In the SMR and IPC-weighted analysis, the HR was attenuated, but remained slightly higher in PPI users compared to nonusers.

| Interpretation
Our study adds to the existing literature on the impact of PPIs on the risk of AKI by examining a high-risk cohort of cancer patients receiving ICIs.0][21] In case-control studies by Cortazar et al and Gupta et al, respectively, the authors showed that concomitant use of PPIs was associated with a 2.40-to 2.85-fold increased adjusted odds of AKI in patients receiving ICIs. 19,20In line with these results, Seethapathy et al showed in a cohort study that PPI use at ICI initiation was associated with an adjusted HR of 2.85. 6Since both ICIs and PPIs are drugs associated with acute interstitial nephritis, 19,30 PPIs are hypothesized to predispose to acute interstitial nephritis in patients receiving ICI by activating or reactivating drug-specific T-cell-mediated immune reactions. 19,20,31The results from previous studies finding an association between PPI and AKI in patients receiving ICI are in line with the crude results from our per-protocol analysis but in contrast to the weighted results from our analyses.There are several possible explanations behind these divergent results.First, the previous studies adjusted for only a limited number of potential confounders, whereas our study accounted for a large range of covariates likely to be associ- and AKI risk. 21In contrast, the outcome in our study was defined using creatinine measurements from laboratory databases, whereby the risk of notification bias was minimized.

| Limitations
Although our study takes advantage of the extensive Danish data on medical history, treatment exposure, and plasma creatinine measurements as well as complete follow-up on the included patients, some limitations should be considered.First, in the main analysis, patients with a prescription of PPIs redeemed within the 90 days before or on the day of ICI initiation were considered exposed, while patients with no redeemed prescription of PPIs within the 365 days before or on the day of ICI initiation were considered unexposed.However, patients could have initiated or discontinued PPI treatment during follow-up, which may have reduced the exposure contrast between users and nonusers during follow-up.If so, this intention-to-treat approach may have biased the results toward no effect, assuming the misclassification to be nondifferential.Addressing this, we conducted a per-protocol analysis, which showed a tendency toward PPI use being associated with an increased risk of AKI.However, this design is susceptible to selection bias, and despite our efforts to mitigate this issue by applying IPC weights, which attenuated the association, it is questionable whether we have adequately included variables both associated with AKI and switching of exposure to PPIs.Therefore, selection bias may, at least to some extent, explain this finding.Second, we observed that the HR moved from 1.42 to 1.06 in the main analysis and from 1.86 to 1.37 in the per-protocol analysis when accounting for confounders, indicating that confounding had a considerable impact on the observed association between PPI use and AKI.
Despite our efforts to control for confounders, residual confounding may have remained in the per-protocol analysis and caused an overestimation of the association between PPI use and AKI.Third, we did not have information on over-the-counter use of PPIs, However, only 2% of PPIs are sold over the counter in Denmark, 32 and thus, any misclassification due to over-the-counter use is assumed to be negligible.
Fourth, AKI was defined based on plasma creatinine measurements, meaning that, some of the captured AKI events may not represent true clinically acknowledged AKI events but kidney-independent fluctuation in plasma creatinine level.To mitigate this potential misclassification, we repeated the analysis while defining AKI as an at least 2-fold increase in plasma creatinine and as AKI persisting >48 hours, and we found similar HRs.Fifth, we included AKI events regardless of whether they were attributed to the ICI treatment.However, from a prognostic perspective, it is important to investigate the overall risk of AKI, as AKI directly attributed to ICIs only confers a fraction of the AKI events 5,6,12 and is associated with a lower mortality than AKI events not directly attributed to ICIs. 4 Moreover, ICIs may induce AKI through indirect pathways, for example, hemodynamic instability due to colitis or myocarditis, which will not be caught by the previously applied definitions of AKI directly attributed to ICI. 19,20

| Conclusion
Based on our findings, we suggest that the increased risk of AKI associated with PPI use at the time of initiation of ICIs is to a large extent explained by confounding.In the per-protocol analysis, we did observe an increased risk of AKI after adjusting for confounding; however, the effect estimate was lower than that reported in previous studies and may still be confounded by indication.Thus, concurrent treatment with PPIs and ICIs may be safe.

1
Illustration of definitions of users, nonusers, new users, long-term users, and former users.ICI, immune checkpoint inhibitor; PPI, proton pump inhibitor.T A B L E 1 Patients' characteristics.
ated with both PPI use and the risk of AKI, including comedication such as opioids, glucocorticoids, and paracetamol.Second, in the casecontrol study by Gupta et al, exposure to PPIs was ascertained in the 14 days preceding the AKI event in cases and at the time of initiation of ICIs in controls.This may have overestimated the association since the prevalence of PPI use may vary during the course of cancer.Furthermore, the PPI use in the weeks leading up to AKI may have been caused by clinical deterioration, which could have caused overestimation of the association between PPIs and AKI. 20In our study, we sought to align the definition of exposure by examining the exposure to PPIs before the ICI initiation.Third, the pharmacovigilance casecontrol study by Gérard et al may have been affected by notification bias, which may have overestimated the association between PPI use

3
Standardized mortality ratio-weighted risk of acute kidney injury in proton pump inhibitor users and nonusers.PPI, proton pump inhibitor.
Outcomes in proton pump inhibitor users and nonusers.