Models to Predict Risk after Acute Kidney Injury

Among the one in seven hospitalized patients who develop acute kidney injury (AKI) during hospital admission, many continue to experience poor health outcomes following discharge, including a one in three risk of unplanned readmission within 90 days, as well as incomplete recovery of kidney function associated with new or progressive chronic kidney disease during the year after the initial discharge. There are few data available on which discharged hospital patients should receive follow-up care after AKI and for which reasons.

Current guidelines recommend follow-up of all individuals with AKI after 90 days. According to Simon Sawhney, MBChB, PhD, and colleagues, for some patients, an early period of increased surveillance and care may be beneficial; for others, additional monitoring and follow-up interventions may not be of benefit and may unnecessarily increase healthcare costs.

Two risk prediction models for outcomes of AKI after hospital discharge have been developed recently. The Grampian- Aberdeen (United Kingdom) tool predicts death or hospital readmission for all individuals (with and without AKI) in the early postdischarge period. The model, that includes AKI as a key predictor, could be used by primary care physicians to target those who may benefit from early surveillance. A second focused risk prediction model, Alberta (Canada), has been developed to predict the risk of progression to new CKD glomerular filtration rate (GFR) categories 4 and 5 (G4-G5) among survivors of AKI and can be used to identify patients who would benefit from referral to a nephrologist.

Dr. Sawhney et al. conducted a validation study of the two risk models of AKI outcomes to compare the net benefit of risk model-based clinical decisions following AKI. The outcomes of interest were death or readmission within 90 days for all hospital survivors, and progression to new CKD G4-G5 for patients who survived at least 90 days following AKI. Results were reported in the American Journal of Kidney Diseases [2021;78(1):28-37].

In both cohorts, AKI was defined and staged for severity using Kidney Disease Improving Global Outcomes (KDIGO) AKI criteria. The Aberdeen readmissions model established AKI using a validated KDIGO-based algorithm: AKI was present if there was (1) an increase in creatinine of >0.3 mg/dL within 48 hours; or (2) a 50% rise from the lowest creatinine in the past 7 days; or (3) a 50% rise from the median creatinine in the past 8-90 days, or 91-3265 days if no closer samples existed. In the Alberta CKD G4-G5 model, AKI was established based on a rise in creatinine during hospitalization of >0.3 mg/dL or >50% of the most recent outpatient prehospital baseline 7-365 days prior to admission.

The Aberdeen death or readmissions model was temporally validated in a cohort of all adult Grampian, North Scotland, residents admitted to hospital in Grampian in 2012 (with or without AKI). The Alberta CKD G4-G5 model was geographically validated using all adult Grampian residents admitted to hospital with AKI between 2011-2013 who had a baseline estimated GFR >45 mL/min/1.73 m2 and survived at least 90 days postdischarge.

The validation cohort for the Aberdeen death or readmissions model included 26,575 individuals (mean age, 66.1 years) and 2927 events. The validation cohort for the Alberta CKD G4-G5 model included 9382 individuals (mean age, 60.9 years) and 140 events.

Both models discriminated well in the external validation cohorts: area under the curve (AUC) of 0.77 for the Aberdeen death or readmissions model and AUC of 0.86 for the Alberta G4-G5 model. In both models, risks were overpredicted; following recalibration, that measure improved. The odds ratios were similar for refitted models with the exception of albuminuria in the CKD G4-G5 model, for which an unmeasured value of albuminuria had a protective effect in the external validation cohort.

For predicting death or readmission among all hospital survivors using the Aberdeen model, there was a positive net benefit of follow-up of those with AKI; a model-guided approach led to the greatest net benefit at the relevant risk thresholds. In decision curve analysis of the net benefit for predicting death or readmission where the analysis was restricted to people with AKI, the greatest net benefit occured by following the risk model, with limited benefit or potential net harm from following only those with severe AKI.

For predicting CKD G4-G5 progression at 1 year using the Alberta model among those who survived to hospital discharge, a model-guided approach provided a small gain in net benefit that remained superior to a strategy guided by discharge eGFR <30 mL/min/1.73 m2 at the prespecified 10% risk threshold. The superiority of a model-guided approach was consistent for both models regardless of whether the original, recalibrated, or refitted model was used.

In process mining of all hospital discharges, of the 105,461 individuals discharged from a hospital admission in Grampian between 2011-2014, 9% (n=9220) died or were readmitted within 90 days. Of the 9220, 41% (n=3776) were recovering after an episode of AKI.

Of 13,232 individuals discharged after AKI, 29% (n=3776) were readmitted or died within 90 days. Most of the monitoring of kidney function between 30 and 90 days postdischarge was conducted in primary care. An additional 10% (1369/13,232) patients were assessed in an outpatient specialty clinic within 90 days, and 10% (1325/13.232) attended an emergency department. A lack of any postdischarge monitoring between discharge and readmission was evident in 42% (1101/2401) of deaths/readmissions within 30 days, and 39% (1464/3776) of deaths or readmission within 90 days after AKI. Median times to the unmonitored adverse outcomes were 9 and 13 days.

In citing limitations, the researchers noted that both original models overstated risks, indicating the need for regular updating.

In conclusion, the authors said, “We have shown that risk prediction models for death or readmission and CKD have the potential to assist in prioritizing people who have had AKI within follow-up care planning and may be superior to alternative strategies such as prioritizing on AKI severity or kidney recovery alone. Further, many people with poor outcomes after AKI receive little or no postdischarge monitoring. A necessary next step is to design and trial risk model-assisted decisions that triage people into appropriate models of postdischarge care that provide the most appropriate level of specialist/nonspecialist input at the most appropriate timepoint.”

Takeaway Points

  1. There are few data available to guide prioritizing care following hospital discharge among patients who experiences acute kidney injury (AKI) during the hospital stay.
  2. Results of an external validation of two risk models for outcomes following AKI were reported. The models were the Grampian-Aberdeen readmission model and the Alberta model that predicts the risk of progression to chronic kidney disease stage 4-5.
  3. Both models provided net benefit superior to any other decision strategy.