A randomized clinical trial assessing the effect of automated medication-targeted alerts on acute kidney injury outcomes

Abstract

Acute kidney injury is common among hospitalized individuals, particularly those exposed to certain medications, and is associated with substantial morbidity and mortality. In a pragmatic, open-label, National Institutes of Health-funded, parallel group randomized controlled trial (clinicaltrials.gov NCT02771977), we investigate whether an automated clinical decision support system affects discontinuation rates of potentially nephrotoxic medications and improves outcomes in patients with AKI. Participants included 5060 hospitalized adults with AKI and an active order for any of three classes of medications of interest: non-steroidal anti-inflammatory drugs, renin–angiotensin–aldosterone system inhibitors, or proton pump inhibitors. Within 24 hours of randomization, a medication of interest was discontinued in 61.1% of the alert group versus 55.9% of the usual care group (relative risk 1.08, 1.04–1.14, p = 0.0003). The primary outcome — a composite of progression of acute kidney injury, dialysis, or death within 14 days — occurred in 585 (23.1%) of individuals in the alert group and 639 (25.3%) of patients in the usual care group (RR 0.92, 0.83–1.01, p = 0.09).

Introduction

Acute kidney injury (AKI) — an abrupt decline in kidney function — is a major cause of morbidity and mortality among hospitalized patients, associated with an increased hospital length of stay and a 10-fold increase in inpatient mortality. AKI also carries an independent risk for development of chronic kidney disease (CKD), end stage kidney disease, and death.

Despite current international best practice guidelines for management of AKI focusing on avoidance of nephrotoxic exposures, studies have shown that AKI, often asymptomatic in presentation, frequently goes unnoticed, and appropriate workup and treatment is inconsistently performed. For example, a retrospective study of 9534 hospitalized patients with severe AKI found that the rate of discontinuation of potentially nephrotoxic medications is low in the early stages of AKI, revealing an opportunity for early intervention that could potentially prevent progression to more severe forms of AKI.

Medications that affect kidney function are common contributors to AKI in hospitalized patients. However, international guidelines vary with respect to discontinuation of non-steroidal anti-inflammatory drugs (NSAIDs), renin–angiotensin–aldosterone system inhibitors (RAASi), and proton pump inhibitors (PPIs). Observational studies have shown that NSAIDs, potentially by increasing kidney vasoconstriction, increase the risk of AKI both in and out of the hospital. International guidelines strongly recommend the discontinuation of NSAIDs in the context of AKI. RAASi lower hydrostatic pressure at the glomerulus, thus decreasing the glomerular filtration rate and inhibiting the clearance of uremic toxins. Interestingly, RAASi may also increase kidney peritubular blood flow, potentially protecting against ischemic damage. There is thus debate regarding the utility of RAASi discontinuation during AKI, and this is reflected in guideline recommendations to "consider" discontinuation. PPIs have long been associated with the development of acute interstitial nephritis (AIN), a form of AKI, and have been linked to the progression of CKD. As AIN is likely underdiagnosed in hospitalized patients, and PPIs are likely overprescribed, there is significant interest in whether empiric discontinuation of PPI would affect the course of AKI in this population. To date, international guidelines do not address empiric discontinuation of PPI among patients with AKI. The ELectronic Alerts for AKI Amelioration 2 (ELAIA-2) investigators set out to evaluate the aggregate and individual effect of prompting discontinuation of NSAID, RAASi, and PPI within a single trial framework to provide higher quality evidence to clinicians caring for patients with AKI exposed to these agents.

Clinical decision support and best practice alerts have proliferated with the adoption of the electronic health record (EHR), often without rigorous evidence-based support. Trials of alerts for patients with AKI, including our own, have led to mixed results, though the most successful efforts couple alerts to specific, actionable information. In the ELAIA-2 trial, we set out to evaluate the effectiveness of a medication-targeted AKI alert across multiple hospitals within a large health system. Here, we show that alerts can increase the rate of cessation of medications of interest, but have limited effect on key clinical outcomes.

Results

Discussion

This pragmatic, randomized trial of medication-targeted electronic alerts for AKI found that alerts had a significant impact on reducing exposure to the medications of interest, but no significant effect on the primary outcome of progression of AKI, dialysis, or death. However, the alert was associated with clinical benefit among individuals receiving PPI at the time of randomization; PPI-exposed individuals comprised the majority of the study cohort.

Prior research into the efficacy of alerting for AKI has shown mixed results. In part, this is due to the variety of study designs employed to evaluate AKI alerts (which range from retrospective before/after analyses, to stepped-wedge designs, to individual-level randomized trials). In addition, the nature of the alerts studied varies significantly, ranging from e-mail based notifications to electronic "pop ups" of the sort tested in this study. Our group has conducted two prior randomized trials evaluating purely informational AKI alerts which, while showing a modest effect on process outcomes, did not demonstrate an improvement in clinical outcomes.

Several studies show that AKI alerts have substantial promise when executed appropriately. A stepped-wedge trial by Selby et al. showed that AKI alerts coupled to a care bundle and an educational AKI package led to a significantly reduced hospital length of stay, but no change in overall mortality. A large before / after study of a clinical decision support system for AKI including more than 500,000 patients found that hospital mortality decreased by around 10% among those with AKI compared to no change in those without AKI between the pre- and post-intervention period, but this study design precluded a causal analysis. The Nephrotoxic Injury Negated by Just-in-Time Action (NINJA) program has successfully demonstrated how alerts targeting medications can improve outcomes in children with AKI. Together with those studies, our study suggests that alerts coupled with actionable processes may improve outcomes for certain patients with AKI.

We were surprised to see no clinical benefit of alerts in the NSAID subgroup. We targeted this class of medications given the broad consensus and international guidelines that suggest discontinuing NSAIDs in the setting of AKI given their known potent adverse biological effects on kidney health. While the alert did numerically increase the rate of cessation of NSAIDs, the effect was not statistically significant. The control rate of NSAID cessation was the highest of any class, indicating that many physicians are stopping NSAIDs during AKI. Alerts can only translate to clinical benefit when they induce a substantial change in provider behavior.

Our observation that alerts reduced the primary outcome among those treated with PPIs has several potential explanations. Although this was a pre-specified subgroup analysis, there is the possibility that this signal is due to chance. Alternatively, there may be a causal link between PPI-use and AKI which is ameliorated by cessation in the context of alerting. Prior research, predominantly observational, has elucidated associations between PPI and AKI, and particularly acute interstitial nephritis. Research has also suggested a link between PPI use and CKD. We also recognize that the alert may have effects on AKI diagnosis, management, and treatment beyond medication cessation alone. In that context, the observed benefit among those receiving PPIs may be due to the fact that PPI use flags a distinct population of patients that benefit from AKI alerts in general. We note, for example, that those receiving PPIs were in more acutely ill than those who were not receiving PPI — with a higher proportion randomized in the ICU, and overall worse outcomes. Our mediation analysis, though underpowered, suggests that there are multiple pathways of alert benefit in this population, with some, but likely not the majority, benefit mediated by actual cessation of the PPI. Further studies to elucidate the mechanism of PPI-associated AKI and the benefit of withholding these medications in the setting of changes in kidney function are needed.

We were also intrigued to examine the data surrounding the alert targeting those on RAASi. Substantial debate exists about the utility of stopping these agents during AKI. RAASi reduce glomerular hydrostatic pressure and thus GFR which may lead to harm by decreasing clearance of uremic byproducts. On the other hand, they increase kidney perfusion which theoretically may reduce ongoing ischemia and allow for more rapid AKI recovery. While the alert led to significantly greater cessation of RAASi, we saw no effect on clinical outcomes, suggesting that these two forces may be in balance. Future studies may examine the benefit of stopping (or continuing) RAASi in key patient populations — such as those with AKI and heart failure.

Limitations of this study include the fact that only three medication classes were targeted. There are many other medications, including aminoglycoside antibiotics, certain types of chemotherapy, and others that are potentially nephrotoxic. However, we designed this trial not only to inform the benefit of alerting for AKI in the setting of medication exposures, but to evaluate the impact of cessation of key often-used medication classes in the setting of AKI. This has allowed us to conduct careful sub-group analyses that suggest that, for example, cessation of PPI may be reasonable among those with AKI. Other limitations include the fact that this trial was randomized at the patient level, which introduces the possibility of contamination, as providers who receive alerts may learn how better to care for AKI patients over time. Alternative designs, such as cluster-randomized or stepped-wedge designs might ameliorate this, but simulation studies have shown that, in the setting of potential contamination, statistical power is maximized through inflation of sample size (as done here) more efficiently than by conversion to a cluster or stepped-wedge design. In addition, the lack of effect modification by study time suggests that there was limited contamination of the usual care arm of the study. This study was conducted within a single health system in the Northeastern US, potentially limiting generalizability, but the hospitals included serve a demographically and economically diverse population. In addition, our alerts did not exist in a vacuum, but in competition with all the other clinical alerts and distractions that are enabled by the EHR. Alert fatigue may diminish alert performance. Moreover, in order to meet requirements set out by the IRB, the language of the alert was limited in how stringently it could recommend medication cessation; more forceful wording could have increased cessation rates, and, potentially, clinical effectiveness. Finally, AKI is a heterogenous disease; alerts may have different efficacy depending on the underlying mechanism (for example, alerts to stop NSAIDs might be particularly useful in the setting of hemodynamic AKI). However, the ability to determine the underlying cause of AKI in real time is difficult even for trained nephrologists, more so for electronic algorithms.

Strengths of this study include its sample size, which was adequate to detect a clinically meaningful impact of alerting, its use of a clinical outcome, focus on agents commonly used in the inpatient setting, randomization, and diverse patient mix.

In conclusion, this pragmatic, randomized trial of automated, electronic alerting for individuals with AKI exposed to common kidney-relevant medications suggests that there is unlikely to be broad-based benefit to all patients, but those exposed to PPIs may benefit.

Methods

Study design

The protocol was carried out according to the principles in the Declaration of Helsinki and approved by the Yale Human Investigations Committee (HIC #2000025786) which approved a waiver of informed consent. A reliance agreement was obtained from the Bridgeport Hospital and Greenwich Hospital IRBs as well. A waiver of informed consent was granted on the basis that the intervention was minimal risk, that the rights and welfare of participants were protected, and that the study could not be feasibly conducted with consent (as the act of consent would serve to be an AKI alert of sorts). ELAIA-2 was a multicenter, parallel-group, pragmatic, open-label randomized trial of a real-time EHR alert for AKI among individuals exposed to one or more of three classes of kidney-relevant MOIs: NSAIDs; RAASi and PPIs. These three classes were chosen as consensus recommendations regarding their empiric discontinuation in patients with AKI vary. The trial was pre-registered at clinicaltrials.gov/NCT02771977 on May 03, 2016 prior to any patient enrollment and no protocol modifications occurred after the study was launched.

Participants

Enrollment was automated, with inclusion and exclusion criteria built into the alert system itself. Eligible patients were adults ≥ 18 years old who were admitted to one of four participating study hospitals, had AKI as defined by the Kidney Disease: Improving Global Outcomes (KDIGO) serum creatinine criteria (at least 50% increase in creatinine within 7 days or a 0.3 mg/dL increase within 48 h), and had an active order for one or more of the three MOIs. Patients were excluded if their initial creatinine after admission was ≥ 4.0 mg/dL or if they had received dialysis within the year prior to meeting the AKI definition, if they had been admitted to a hospice service or had an active "comfort measures only" order, if they had a diagnosis code consistent with end-stage kidney disease, or if they had a kidney transplant within the 6 months prior to randomization.

Randomization and masking

Allocation was in a 1:1 ratio, achieved via simple randomization within the EHR. Participants and providers were necessarily not masked to the intervention, but the study investigators were.

Intervention

The intervention was a "pop-up" electronic alert embedded within the EHR (Epic systems, Verona WI). The alert would display when a provider (MD, DO, PA, or APRN, regardless of whether a primary or consulting provider) opened the order entry tab of the patient's chart. Alerts would display once per provider on a 24-h basis, provided the patient still met AKI criteria and continued to have an active prescription for one of the MOIs. The alert prompted providers to "consider clinical indication for the following medications!" It included information regarding the patient's creatinine level and trend and enumerated the class of MOI and specific MOI that was currently ordered. It also provided a direct link to the order entry system to allow for potential discontinuation of the listed medication. Patients randomized to the control group were identified via a silent alert (not shown to providers, but logged in the system), and received usual care.

Outcomes

The primary outcome was a composite of progression of AKI, receipt of dialysis, or death within 14 days of randomization or hospital discharge. Progression of AKI was defined by achieving a higher KDIGO AKI stage than the stage present at initial randomization (stage 2 represents a doubling of creatinine, stage 3 a tripling of creatinine). Secondary outcomes included the components of the primary outcome (at 14 days and during the entire admission), the proportion of patients who did not receive one or more MOIs for which they were enrolled within 24 h and while AKI was still present, the duration of AKI, the rate of readmission at 30-days, and hospital costs (as determined by internal cost accounting independent of insurance status).

Strength of guidance regarding empiric discontinuation of the MOIs differed by medication class, as well as by the mechanisms by which they influence renal function, prompting us to pre-specify subgroup analyses stratified by the MOI being received at randomization. We further identified key safety outcomes that may be associated with discontinuation of each class of MOI — all within 14 days or hospital discharge. For NSAIDs, we assessed the rate of prescription of opioid analgesics and maximum pain scores. For RAASi, we assessed the maximum achieved systolic and diastolic blood pressure and the rate of intubation (a proxy for flash pulmonary edema). For PPIs, we assessed the rate of blood transfusion (as inappropriate cessation of PPI may increase the risk of gastrointestinal bleeding), the minimum hemoglobin achieved, and maximum pain scores to capture epigastric pain consequent to PPI cessation.

Based on data from our prior trial, we expected the primary outcome to occur in 18.2% of control patients and considered a relative reduction of 20% to 14.6% would be clinically significant. The target sample size of 5060 individuals provided 90% power to detect a difference at least this large, accounting for two interim analyses and potential 10% contamination of intervention across study arms.

Statistical analysis

We present data as median (interquartile range — IQR) or count (proportion) as appropriate. We compared categorical variables across the treatment groups with the use of the Cochran–Mantel–Haenszel chi-squared test stratified by the four study hospitals. We assessed continuous outcomes using the Van Elteren test, accounting for clustering by study hospital. All data was analyzed using the intention-to-treat principle. Subgroup analyses presented herein were pre-specified. Due to the interim analysis, the two-sided p-value threshold for statistical significance of the primary outcome was 0.0498 based on the Hwang–Shih–DeCani spending function. Secondary outcomes are evaluated at a two-sided p-value threshold of 0.05 without multiplicity adjustment and should be considered hypothesis generating. Subgroup analyses, while pre-specified, should be considered exploratory.

We determined whether observed clinical effects were driven by cessation of target agents by performing mediation analyses treating cessation of the MOI as a binary mediator. The product of coefficients method was used to evaluate whether MOI cessation mediates the relationship between intervention and composite outcome. Baseline covariates, including age, sex, ICU at randomization, creatinine, hemoglobin, platelet counts, the use of other MOIs and time from admission to randomization, were adjusted to control for mediator-outcome confounding, and bootstrapping was used to calculate the 95% confidence interval for proportion mediated.

All statistical tests were performed in Stata version 15 (College Station, TX), SAS v9.4 (Cary, NC), and R v4.0.1 (Boston, MA).

Data availability

The data generated in this study have been deposited in the DataDryad database under accession code doi.org/10.5061/dryad.kh189327p. This data can be accessed and used for scientific research without permission from the authors. The DSMB has approved the release of this deidentified dataset.

References

Inline citation numbers have been omitted for readability in this interactive companion; the full bibliography below matches the published paper.

1. Hoste, E. A. J. et al. Epidemiology of acute kidney injury in critically ill patients: the multinational AKI-EPI study. Intensive Care Med. 41, 1411–1423 (2015).
2. Silver, S. A., Long, J., Zheng, Y. & Chertow, G. M. Cost of acute kidney injury in hospitalized patients. J. Hosp. Med. 12, 70–76 (2017).
3. Wonnacott, A., Meran, S., Amphlett, B., Talabani, B. & Phillips, A. Epidemiology and outcomes in community-acquired versus hospital-acquired AKI. Clin. J. Am. Soc. Nephrol. 9, 1007–1014 (2014).
4. Coca, S. G., Singanamala, S. & Parikh, C. R. Chronic kidney disease after acute kidney injury: a systematic review and meta-analysis. Kidney Int. 81, 442–448 (2012).
5. Acosta-Ochoa, I. et al. Impact on outcomes across KDIGO-2012 AKI criteria according to baseline renal function. J. Clin. Med. 8, 1323 (2019).
6. Zeng, X., McMahon, G. M., Brunelli, S. M., Bates, D. W. & Waikar, S. S. Incidence, outcomes, and comparisons across definitions of AKI in hospitalized individuals. Clin. J. Am. Soc. Nephrol. 9, 12–20 (2014).
7. Ishani, A. et al. Acute kidney injury increases risk of ESRD among elderly. J. Am. Soc. Nephrol. 20, 223–228 (2009).
8. Ftouh, S. & Thomas, M. Acute kidney injury: summary of NICE guidance. BMJ 347, f4930 (2013).
9. Kellum, J. A. et al. KDIGO clinical practice guideline for acute kidney injury. Kidney Int. Suppl. 2, 1–138 (2012).
10. Srisawat, N. et al. Variation in risk and mortality of acute kidney injury in critically ill patients: a multicenter study. Am. J. Nephrol. 41, 81–88 (2015).
11. Tawadrous, D. et al. Use of clinical decision support systems for kidney-related drug prescribing: a systematic review. Am. J. Kidney Dis. 58, 903–914 (2011).
12. Finlay, S. et al. Identification of risk factors associated with acute kidney injury in patients admitted to acute medical units. Clin. Med. 13, 233 (2013).
13. Moledina, D. G. et al. Variation in best practice measures in patients with severe hospital-acquired acute kidney injury: a multicenter study. Am. J. Kidney Dis. 77, 547–549 (2021).
14. Chou, C. I. et al. Adverse effects of oral nonselective and cyclooxygenase-2-selective NSAIDs on hospitalization for acute kidney injury: a nested case–control cohort study. Medicine 95, e2645 (2016).
15. Lafrance, J. P. & Miller, D. R. Selective and non-selective non-steroidal anti-inflammatory drugs and the risk of acute kidney injury. Pharmacoepidemiol. Drug Saf. 18, 923–931 (2009).
16. Clavé, S., Rousset-Rouvière, C., Daniel, L. & Tsimaratos, M. The invisible threat of non-steroidal anti-inflammatory drugs for kidneys. Front Pediatr. 7, 520 (2019).
17. Arora, P. et al. Preoperative use of angiotensin-converting enzyme inhibitors/angiotensin receptor blockers is associated with increased risk for acute kidney injury after cardiovascular surgery. Clin. J. Am. Soc. Nephrol. 3, 1266–1273 (2008).
18. Chaumont, M. et al. Acute kidney injury in elderly patients with chronic kidney disease: do angiotensin-converting enzyme inhibitors carry a risk? J. Clin. Hypertens. 18, 514–521 (2016).
19. Hsu, C.-y. et al. Renin-angiotensin system blockade after acute kidney injury (AKI) and risk of recurrent AKI. Clin. J. Am. Soc. Nephrol. 15, 26–34 (2020).
20. Brar, S. et al. Association of angiotensin-converting enzyme inhibitor or angiotensin receptor blocker use with outcomes after acute kidney injury. JAMA Intern. Med. 178, 1681–1690 (2018).
21. Tomson, C. & Tomlinson, L. A. Stopping RAS inhibitors to minimize AKI: more harm than good? Clin. J. Am. Soc. Nephrol. 14, 617 (2019).
22. Lazarus, B. et al. Proton pump inhibitor use and the risk of chronic kidney disease. JAMA Intern. Med. 176, 238–246 (2016).
23. Antoniou, T. et al. Proton pump inhibitors and the risk of acute kidney injury in older patients: a population-based cohort study. CMAJ Open 3, E166–E171 (2015).
24. Yang, Y. et al. Proton-pump inhibitors use, and risk of acute kidney injury: a meta-analysis of observational studies. Drug Des. Devel. Ther. 11, 1291 (2017).
25. Hart, E., Dunn, T. E., Feuerstein, S. & Jacobs, D. M. Proton pump inhibitors and risk of acute and chronic kidney disease: a retrospective cohort study. Pharmacother. 39, 443–453 (2019).
26. Moledina, D. G. & Perazella, M. A. PPIs and kidney disease: from AIN to CKD. J. Nephrol. 29, 611–616 (2016).
27. Eid, S. M. et al. Patterns and predictors of proton pump inhibitor overuse among academic and non-academic hospitalists. Intern. Med. 49, 2561–2568 (2010).
28. Ksidzyna, D., Szelg, A. & Paradowski, L. Overuse of proton pump inhibitors. Pol. Arch. Med. Wewn. 125, 289–298 (2015).
29. Ramirez, E. et al. Overuse of PPIs in patients at admission, during hospitalisation, and at discharge in a terciary Spanish hospital. Curr. Clin. Pharm. 5, 288–297 (2010).
30. Wilson, F. P. et al. Automated, electronic alerts for acute kidney injury: a single-blind, parallel-group, randomised controlled trial. Lancet 385, 1966–1974 (2015).
31. Wilson, F. P. et al. Electronic health record alerts for acute kidney injury: multicenter, randomized clinical trial. BMJ 372, m4786 (2021).
32. Selby, N. M. et al. An organizational-level program of intervention for AKI: a pragmatic stepped wedge cluster randomized trial. J. Am. Soc. Nephrol. 30, 505–515 (2019).
33. Goldstein, S. L. et al. A prospective multi-center quality improvement initiative (NINJA) indicates a reduction in nephrotoxic acute kidney injury in hospitalized children. Kidney Int. 97, 580–588 (2020).
34. Lachance, P. et al. Association between e-alert implementation for detection of acute kidney injury and outcomes: a systematic review. Nephrol. Dial. Transpl. 32, 265–272 (2017).
35. Al-Jaghbeer, M., Dealmeida, D., Bilderback, A., Ambrosino, R. & Kellum, J. A. Clinical decision support for in-hospital AKI. J. Am. Soc. Nephrol. 29, 654–660 (2018).
36. Chen, G. et al. Acute kidney injury following the use of different proton pump inhibitor regimens: a real-world analysis of post-marketing surveillance data. J. Gastroenterol. Hepatol. 36, 156–162 (2021).
37. Leonard, C. E. et al. Proton pump inhibitors and traditional non-steroidal anti-inflammatory drugs and the risk of acute interstitial nephritis and acute kidney injury. Pharmacoepidemiol. Drug Saf. 21, 1155–1172 (2012).
38. Moledina, D. G. & Perazella, M. A. Proton pump inhibitors and CKD. J. Am. Soc. Nephrol. 27, 2926–2928 (2016).
39. Milne, B., Gilbey, T., Ostermann, M. & Kunst, G. Pro: we should stop ACE inhibitors early before cardiac surgery to prevent post-operative acute kidney injury. J. Cardiothorac. Vasc. Anesth. 34, 2832–2835 (2020).
40. Tempe, D. K. & Hasija, S. Con: does preoperative discontinuation of angiotensin-converting enzyme inhibitors/angiotensin II receptor blockers reduce postoperative acute kidney injury. J. Cardiothorac. Vasc. Anesth. 34, 2836–2838 (2020).
41. Torgerson, D. J. Contamination in trials: is cluster randomisation the answer. BMJ 322, 355–357 (2001).
42. Hemming, K., Taljaard, M., Moerbeek, M. & Forbes, A. Contamination: How much can an individually randomized trial tolerate. Stat. Med. 40, 3329–3351 (2021).
43. Mehta, S. et al. The prognostic importance of duration of AKI: a systematic review and meta-analysis. BMC Nephrol. 19, 1–10 (2018).
44. Hwang, I. K., Shih, W. J. & De Cani, J. S. Group sequential designs using a family of type I error probability spending functions. Stat. Med. 9, 1439–1445 (1990).
45. Wilson, F. Automated, medication-targeted alerts for Acute Kidney Injury — A randomized trial, Dryad, doi.org/10.5061/dryad.kh189327p (2023).