USING ROUTINELY COLLECTED DATA TO DEFINE THE OPTIMAL TIMING TO INITIATE RENAL REPLACEMENT THERAPY IN AKI PATIENTS

Abstract BACKGROUND AND AIMS The optimal moment to start renal replacement therapy (RRT) in patients with acute kidney injury (AKI) in the intensive care unit (ICU) remains a challenging problem. Evidence is lacking that, in absence of absolute criteria for pH or serum potassium, timing of RRT initiation (early versus delayed) affects survival. The optimal cut-offs for these absolute criteria are however poorly defined. We used routinely collected observational data from tertiary ICUs to investigate whether applying pre-specified dynamic strategies for RRT initiation based on time-updated levels of serum potassium and pH, next to persisting oliguria, could further minimize 30-day ICU mortality in patients with stage 2 KDIGO-AKI. METHOD Based on cut-off values applied in large key RCTs on this topic, we investigated different pH thresholds ranging from 7.0 to 7.2 and serum potassium thresholds ranging from 5.5 to 6.5 mmol/L to identify the treatment strategy resulting in the lowest 30-day ICU mortality. Patients were followed from the time when one of the stage 2 KDIGO-AKI criteria was satisfied for the first time (considered as day 0) until ICU death, discharge or day 30, whichever occurred first. We evaluated decision rules in a hypothetical setting where a decision about RRT initiation is made every 24 h starting from day 0 based on the information available up to that point. If the treatment decision based on the considered strategy did not coincide with the observed treatment, the patient was censored for this regime from this time-point onward. We applied inverse probability of censoring (IPC) weighting to deal with potential selection bias due to this artificial censoring. We calculated the cumulative ICU mortality under each strategy using the IPC-weighted Aalen–Johansen estimator. We considered the best RRT initiation regime to be the one providing the lowest 30-day ICU mortality. We moreover estimated the number of patients actually initiated on RRT under the different regimes. RESULTS Of the 13403 available, potentially eligible ICU admissions between 1 January 2013 and 31 December 2017 (62.2% male, 60.8 ± 16.8 years of age, SOFA 7.0 ± 4.1), 4769 individual patients (66.4% male, 63.3 ± 15.6 years of age, SOFA 8.4 ± 4.3) met our in and exclusion criteria. Figure 1 presents the cumulative ICU mortality since day 0 for the treatment strategies: ‘Initiate RRT if creatinine stage 2 KDIGO-AKI condition has been met and at least one of the events occurred: pH < x, K > 6.0 mmol/L or oliguric stage 3 KDIGO-AKI condition has been met’ (left panel) or based only on pH measurement (so irrespective of serum potassium): ‘Initiate RRT if creatinine stage 2 KDIGO-AKI condition has been met and at least one of the events occurred: pH < x or oliguric stage 3 KDIGO-AKI condition has been met’ (right panel). The black line (obs) corresponds to the cumulative ICU mortality observed under the current standard of care. The best performing out of considered RRT initiation strategies improved 30-day mortality from 12.4% for the current standard of care to 10.8% (12.9% relative decrease) with a corresponding reduction of the number of patients initiated on RRT from 333 to 279 (16.2%). CONCLUSION Dynamic decision rules for delayed RRT initiation strategies based on lower (versus higher) thresholds of pH and/or higher (versus lower) thresholds of serum potassium in addition to persisting oliguria, resulted in lower 30-day ICU mortality and fewer patients initiated on RRT as compared with the current standard practice. Our results confirm absence of benefit of early start and add details on specific safety thresholds for absolute indications to initiate RRT. Our analysis should be considered through the nonrandomized nature of the study and the resulting conclusions might be potentially prone to uncontrolled residual confounding or selection bias.