The association between proton pump inhibitor use and risk of post-hospitalization acute kidney injury: a multicenter prospective matched cohort study

Study population

The Assessment, Serial Evaluation, and Subsequent Sequelae in AKI (ASSESS-AKI) is a prospective cohort study of participants with and without an episode of AKI, who attended a baseline study visit 3 months after the index hospitalization [16, 17]. Participants were enrolled from four North American centers between December 2009 and February 2015. Follow-up in-person study visits were conducted 3 and 12 months after the index hospitalization and annually after that through November 2018, with a determination of estimated glomerular filtration rate (eGFR) and serum creatinine concentration (SCr) [18]. In addition, telephone visits were conducted six months after each annual visit. If participants were hospitalized, then medical records including all in-patient creatinine determinations were obtained. Medical history, study events, and use of medications were updated at each in-person visit or phone contact. Detailed study design and eligibility criteria were included in the supplement and have been published previously [16].

For this analysis, we matched participants with history of PPI use to participants without PPI use during the follow-up period (Fig. 1). The exact matching strategy was performed first with matching factors of center and baseline AKI status. Under each subgroup, we used propensity score matching strategy to deal with further potential confounders [19]. The propensity score analysis was conducted using a multivariable logistic regression to model PPI use during the follow-up period as a function of 13 covariates, including age, gender, race, intensive care unit (ICU) history, creatinine at baseline, diabetes mellitus history, cardiovascular disease history, hypertension history, and six drugs use at baseline (Angiotensin-converting enzyme (ACE) inhibitors, angiotensin receptor blockers (ARB), anti-hypertensive agents, diuretics, insulin, and statins). The nearest-neighbor matching was used, and 1:4 matching was performed with the “without replacement” sampling method. We regarded standardized mean difference (SMD) as a measure to evaluate the matching results [20].

Fig. 1figure 1

Assembly of matched cohort of adults surviving a hospitalization with and without PPI use

Institutional review boards at the participating centers (Data Coordinating Center: Pennsylvania State University, and Clinical Research Centers: Yale University, Kaiser Permanente of Northern California, Vanderbilt University, University of Washington) approved the ASSESS-AKI study, and all methods were conducted in accordance with relevant guidelines and regulations.

Assessment of proton pump inhibitor use

Participants were identified as PPI users if they met both of two criteria: (1) participant’s self-reported or medical-recorded use of PPI (Supplemental Table 1) at any follow-up study visits; (2) at least one visit with PPI use occurring earlier than the incidence of AKI.

Ascertainment of outcomes

Post-hospitalization AKI was counted as one time if participants were in hospitalization with AKI records, which was defined as the percentage increase from the nadir to peak inpatient SCr value was ≥ 50% and/or absolute increase ≥ 0.3 mg/dL (26 µmol / L) in peak inpatient serum creatinine compared with baseline outpatient serum creatinine. For two consecutive episodes of AKI to be considered distinct episodes, they have to meet criteria for non-AKI (i.e., minimum of two serum creatinine in between and ≤ 0.2 mg/dL change above baseline) between episodes. Meanwhile, two consecutive episodes should be separated by at least 30 days [16].

We defined the incidence of AKI when participants without AKI at baseline were diagnosed with AKI during the follow-up period. Recurrent AKI was defined as participants with AKI at baseline who were diagnosed as AKI at least once during the follow-up period.

Progression of kidney disease was included in this study as a secondary outcome. Progression of kidney disease was defined as the occurrence of ESKD (a recipient of outpatient maintenance dialysis or a kidney transplant) or halving of eGFR since the time of the baseline study visit [16].

Assessment of covariates

Demographic information and comorbidities were collected at the baseline study visit by participants’ self-report. All SCr results were performed using an isotope dilution mass spectrometry–traceable assay. eGFR was estimated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) estimating equation [21]. Drug use information on a participant at baseline was collected during index hospitalization.

Power justification

We performed a simulation study to assess the effect size that can be detected with 80% empirical power [24]. Utilizing binomial and negative binomial distributions, we simulated 1:4 matched datasets comprising 340 participants, mirroring our actual study. We subsequently applied the Zero-Inflated Negative Binomial (ZINB) model to analyze these simulated datasets. This simulation process was repeated 10,000 times, and we assessed the effect size (rate ratio) with statistically significant p-values (< 0.05) on approximately 8,000 occasions.

Statistical analysis

For the 340 participants in this cohort study (68 PPI users and 272 PPI non-users), we summarized baseline participant characteristics across PPI users and non-users groups, with mean (SD) values for continuous variables, number and percentage for categorical variables.

For the primary analysis, after examining the hyper-Poisson variability of AKI counts (mean [SD] counts 0.71 [2.15]) and the dispersion statistic ((Pearson statistic)/(degree of freedom) = 1.66), an overdispersion of AKI counts was confirmed and negative binomial regression model would be an appropriate method [22]. Zero-inflation was confirmed in that 237 (69.7%) participants displayed zero counts of AKI incidence (Supplemental Table 2), and a significant positive Vuong statistic was presented (Z = 6.54, P < 0.0001). Therefore, we extended the negative binomial regression model to the ZINB regression model in order to account for the over-abundance of zeros [23].

After confirming no violations of the proportional hazards assumption, Cox proportional hazards regression models were conducted to examine the association between PPI use and the risk of progression of kidney disease, stratified by the matched covariates. For progression of kidney disease, participants were right-censored for death, loss to follow-up, or end of the study, whichever came first.

Sensitivity analyses

We performed four distinct sensitivity analyses to evaluate the robustness of our findings under various scenarios.

Scenario 1: Sensitivity analyses were performed by including covariates into the regression models, conducting multivariable ZINB regression analysis for post-hospitalization AKI, and multivariable Cox-proportional hazards regression analysis for progression of kidney disease.

Scenario 2: Comorbidities were considered as time-dependent covariates, and we aimed to evaluate the impact of newly developed comorbidities during the follow-up period on the hazard ratios (HR). In this context, post-hospitalization AKI was analyzed as time-to-event data, and a multivariable stratified Cox proportional hazards regression model was employed to examine the HR associated with PPI use and AKI. Concurrently, diabetes, cardiovascular disease, and hypertension were incorporated into the regression model as time-dependent variables [25].

Scenario 3: We matched six drug use histories, although there were additional medications, such as non-steroidal anti-inflammatory drugs (NSAIDs), aspirin, vasopressors, immunosuppressants, corticosteroids, and chemotherapeutics, that might influence the relationship between PPI use and AKI. To ensure optimal matching outcomes, we excluded these medications during the matching process. However, we were able to incorporate these drug use histories into the multivariable ZINB regression model.

Scenario 4: For individuals who developed AKI 6 months after discontinuing PPI use, we excluded these PPI users (N = 3) and their four matched pairs (N = 12) from our regression analysis.

We used R software version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria) for the power simulation and matching process, and SAS software (Version 9.4, SAS Institute, Cary, NC) for further analyses with a two-tailed alpha level of 0.05.

留言 (0)

沒有登入
gif