Variations in Mortality among Hospitalizations for Acute Kidney Injury

Health insurance, setting of care, geography, and primary disease all affect mortality among various populations in the United States.^1,2 If medical care for these different populations were truly comparable, then we might expect similar mortality outcomes and attribute differences to random variation. Unfortunately, population-to-population variations in mortality are frequently observed and persist even after controlling for individual patient characteristics. This suggests potentially remediable differences in the content, organization, and delivery of health care may be important in shaping true variation.^3–6 Such modifiable differences in health care–related mortality, if they exist, warrant special attention as clinical and public health problems.

This conceptual framework should be familiar to nephrologists. The national ESRD surveillance system, comprising the ESRD Networks and the US Renal Data System, consistently reports the likelihood of receiving a kidney transplant, mortality in hemodialysis patients, or incident ESRD differ in a nonrandom manner among patient groups, among geographic areas, and over time.⁷ Variations in patient care associated with these outcomes are reported as well,⁸ and interventions during the past 15 years to reduce variations in providing hemodialysis have substantially reduced mortality among patients with ESRD.^9–11

An important question is the extent to which this conceptual framework extends to other aspects of clinical nephrology. One area in which new ideas are being piloted is in pre-ESRD care for stage 4 chronic kidney disease (CKD). The Centers for Medicare and Medicaid Services recently piloted surveillance and quality improvement activities to test the applicability of reducing variations in care in earlier stages of CKD among primary care physicians.¹² If surveillance data can be used to direct targeted interventions to improve the care of patients with earlier stages of CKD, then nephrologists may be able to translate evidence-based clinical practice guidelines into better outcomes for kidney disease.

Another potential application of this conceptual framework comes from an important report in this issue of JASN. James et al.¹³ examine the mortality of patients who were treated in hospitals for acute kidney injury (AKI) regarding variations in time and place. They use an annual sample of hospitalizations, the Nationwide Inpatient Sample, designed in part to track and analyze national trends in health care quality and outcomes, to address two simple questions: Does mortality among patients who are hospitalized for AKI vary between weekday and weekend admissions, and does mortality risk for AKI vary by hospital size? They found patients who were admitted with a diagnosis of primary AKI on weekends experienced 7% greater mortality. This increased weekend risk was substantially greater in smaller (17% increase) compared with larger (7%) hospitals, a 45% relative risk differential.

A frequent concern about reported variations in mortality rates such as those reported by James et al.¹³ is that selection of study participants; errors in measuring either the outcome or group membership' and failure to account for disease severity, comorbidities, and other patient attributes may bias observed results. The authors' treatment of these issues is instructive and provides unexpected insight into hospital admissions for AKI.

They used a nationally representative sample of admissions to US hospitals, readily available to researchers, to reduce selection bias and increase the applicability of their results to the practicing nephrologist. In-hospital mortality was recorded in the same data source, reducing the likelihood that patients who met their AKI diagnosis in either high- or low-mortality hospitals were systematically excluded from the analyses. The authors used a validated, claims-based algorithm to ensure that each case of AKI was based on comparable documentation. Admissions were categorized as being for either primary and secondary AKI on the basis of the discharge diagnosis listed as the main cause for hospitalization. James et al.¹³ also reduced heterogeneity among the populations being compared across time and place by focusing their analyses on the remaining 22% of patients with primary AKI.

These steps substantially reduce but do not preclude the possibility that measurement bias may occur absent standard diagnostic criteria for AKI.¹⁴ Hospital-to-hospital variations in AKI documentation might have been further reduced, however, if more precise diagnostic criteria were available. In particular, identifying the type of injury that led to AKI,¹⁵ which is highly uneven and influences mortality, might have affected the data on day of the week or hospital size, and this issue might contribute to some of the mortality differences; however, information about the cause of injury is not reflected by the current International Classification of Diseases, Ninth Revision codes for acute renal failure (584.5, 584.6, 584.7, 584.8, and 584.9), which depend instead on the presumed location of injury, tubular necrosis, renal cortical necrosis, medullary (papillary) necrosis, other specified pathologic lesion in kidney, and unspecified causes.¹⁶

The reported mortality comparisons might also have been biased if differences in the severity of AKI and other patient attributes associated with increased risk for death varied during the week or among hospitals. They account for this possibility by controlling for age, gender, race, and comorbid conditions. Ideally, had the relevant information been available in the database, risk stratification criteria for AKI such as the recently published Risk, Injury, Failure, Loss, ESRD (RIFLE) criteria that is based on changes in urine output and serum creatinine could have further controlled for differences in the severity of renal injury.^17–19 This issue is important, because it is evident that even minor AKI confers increased risk for both short- and long-term mortality.^20–23

What then are we to make of the conclusion of this well-designed study that weekend admissions characterized by either primary or secondary AKI associate with increased risk for death that is greater in small compared with large hospitals? These data do need replication, as is true of any single report, and as noted by the authors' further explanatory studies. Notably, these results raise the important question of identifying potentially modifiable risk factors that contribute to mortality differences.

James et al.¹³ suggest a reasonable hypothesis that increased weekend mortality reflects delayed recognition and treatment among patients with AKI as a result of the limited resources, particularly timely initiation of nephrology consultation and hemodialysis. These two possibilities warrant close scrutiny and comment. That delayed hemodialysis may contribute to variations in AKI-related mortality reflects the assumption that renal replacement therapy plays a substantial role in the care of patients who are admitted with AKI.

The data in this study, which reveal an unexpected picture of hospitalizations for AKI, suggest otherwise. The frequencies of hemodialysis care on the weekend (8.4%) and weekday (8.9%) were unexpectedly low and nearly the same. Although we are not provided dialysis rates individually for primary and secondary AKI, it is evident that for the overwhelming majority of these patients, someone determined that dialysis was not necessary. The overall low dialysis rates for both weekend and weekday AKI admissions suggest that although delayed initiation of renal replacement therapy may contribute to the increased weekend mortality, it may not be the sole or even major explanatory factor.

What of the second possibility, that failure of timely nephrology referral contributes to increased weekend mortality? As discussed by the authors,¹³ the impact of “nephrology consultation and the provision (and timing of) renal replacement therapy” on patient mortality is not currently supported by evidence—the notion being “hypothetical” in their words. Furthermore, a contrarian might argue if ≥90% of patients who are admitted with AKI do not require hemodialysis, then what added benefit is derived from a nephrology consultation?

The process similar to that used in ESRD to improve hemodialysis care may be relevant in addressing the nondialytic contribution of nephrology care for AKI. The methods for producing evidence-based practice guidelines for nephrology care in AKI is well advanced,²⁴ and the clinical practice guidelines similar to those used in ESRD are being published (http://www.renal.org/Clinical/GuidelinesSection/AcuteKidneyInjury.aspx). Although the current guidelines appropriately focus mainly on dialysis therapy in AKI, current AKI guidelines do suggest nondialytic contributions to the care of patients who are admitted to the hospital for AKI.²⁵ Attributes of this care include timely nephrology consultation, appropriate assignment of cause and severity and need for dialysis, nondialytic management of volume, attention to antisepsis measures, metabolic control, adjustment of drug dosing, and modification of nutritional support. An international Kidney Disease: Improving Global Outcomes (KDIGO) working group is currently developing international AKI guidelines that address nondialytic nephrology care in more detail.²⁶

If an evidence database supporting the role of these or other variations in nephrology care for patients with AKI can be developed²⁷ and translated into quality indicators,²⁸ similar to those for hemodialysis practice, then it would be possible to draw inferences about their contribution to weekday-to-weekend and hospital-to-hospital variability in mortality of patients with AKI. In turn, this information could be used to address variations in care.²⁹

• Copyright © 2010 by the American Society of Nephrology