(124) Risk Modeling to Obtain Individualized Estimates of Ibuprofen Associated Acute Kidney Injury in Hospitalized Patients

Background: Ibuprofen is an essential component of multimodal analgesia in hospitalized patients, where minimization of opioid exposure is a key goal. However, ibuprofen can cause acute kidney injury (AKI) in susceptible patients, and few risk stratification tools are available to individualize treatment decisions. Risk modeling is an increasingly popular approach to identifying heterogeneity of beneficial effects in the analysis of clinical trials. Whether this approach can be successfully applied to examine heterogeneity of adverse effects using real-world data is unclear.

Objectives: To determine whether risk modeling is useful for identifying heterogeneity of ibuprofen’s effect on AKI risk in hospitalized patients.

Methods: We conducted a new-user active comparator cohort study of patients admitted to four hospitals in Philadelphia, Pennsylvania, comparing patients treated with ibuprofen versus those treated with oxycodone. The primary outcome was the incidence rate of AKI though day 14, defined according to the Kidney Disease Improving Global Outcomes creatinine criteria. The analysis of treatment effect heterogeneity involved four steps. First, we propensity score matched ibuprofen patients to controls in a ratio of 1:4 using nearest neighbor caliper matching based on 67 baseline covariates. Second, we derived an AKI risk model from the same set of baseline covariates in the matched population using LASSO logistic regression with 10-fold cross validation. Third, we created subgroups according to predicted AKI risk (low [ < 2.5%]; low-moderate [2.5%-5%]; moderate [5%-7.5%)]; and high [>7.5%]) and addressed potential residual covariate imbalance within risk subgroups with stratified balance weights. Lastly, we estimated stratified incidence rate differences with weighted Poisson regression.

Results: We included 31218 patients in the matched cohort (ibuprofen n=6310; oxycodone n=24908) with a median follow up of 3 days in each group. The final LASSO logistic regression model had a c-statistic of 0.74 and Brier score of 0.042. The overall rate difference associated with ibuprofen treatment was 3.33/1000 days (95% CI 1.37, 5.33), but this estimate varied more than 10-fold across categories of baseline predicted risk: low: -0.15/1000 days (95% CI -2.54, 2.25)]; low-moderate [1.83/1000 days (95% CI -0.38, 4.05)]; moderate [7.06/1000 days (95% CI 1.17, 12.96)]; and high [12.57/1000 days (95% CI 3.85, 21.29)]).

Conclusions: The effect of ibuprofen on AKI rate in hospitalized patients varied substantially across categories of baseline predicted AKI risk. Risk modeling may be a useful tool for developing individualized analgesia regimens in hospitalized patients.