Costs and Outcomes of Acute Kidney Injury in Critically Ill Patients with Cancer

Over the past several years, important advances have occurred in the treatment and supportive care of critically ill patients with cancer.^[1] However, acute kidney injury (AKI) remains a familiar complication and is a negative prognostic factor for overall survival.^{[2] and [3]} The development of AKI can limit further cancer treatment, increase toxicity of chemotherapy and reduce its delivery, and exclude patients from clinical trials. Further, patients with AKI have been shown to have longer hospitalizations and increased hospital costs.^{[4] and [5]} Recognized causes of AKI include acute tubular necrosis from medications or sepsis, volume depletion, tumor lysis syndrome, abdominal compartment syndrome, and obstruction from tumor or lymphadenopathy. Elevations in serum creatinine of as little as 0.3 mg/dL, which were previously considered insignificant, have been associated with a higher mortality rate in hospitalized patients.^[4] However, few of the numerous definitions of AKI used in the cancer literature incorporate these subtle declines in kidney function.

An increase in serum creatinine has traditionally been used as a reflection of AKI. However, it is well known that elevation in serum creatinine is a relatively late marker of kidney injury.^[6] In addition, patients with cancer often have decreased creatinine production secondary to cachexia, which may limit the sensitivity of creatinine as a marker of kidney injury. Other variables including total body volume, ethnicity, medications, and protein intake may also vary the serum creatinine level independent of renal function. Recent studies have demonstrated that a significant number of patients with cancer and normal serum creatinine have underlying chronic kidney disease (CKD) when renal function is estimated by the Cockcroft-Gault equation.^{[7] and [8]} Therefore, using an arbitrarily defined level of serum creatinine as an indicator of AKI (i.e. >1.5 or 2.0 mg/dL) may not be suitable.

What may be a more accurate measure of kidney injury is a classification system based on the percent increase in serum creatinine relative to baseline. One such model is the Risk, Injury, Failure, Loss, and End-Stage Kidney (RIFLE) classification, which defines three levels of severity of AKI (risk, injury, and failure).^[9] Previously, over 35 different definitions of AKI were used in the literature, which has made cross-comparisons between studies difficult.^[10] The RIFLE classification provides a uniform definition of AKI and has been validated in numerous studies.^{[11], [12], [13], [14], [15], [16], [17] and [18]} The aim of this analysis was to estimate the incidence, outcomes, and costs associated with AKI as defined by the RIFLE classification in critically ill patients with cancer.

Materials and Methods

The study included all patients ≥18 years of age who were admitted to the intensive care unit (ICU) at the University of Texas M.D. Anderson Cancer Center from December 2005 through December 2006. Patients with a baseline serum creatinine >1.5 mg/dL were excluded from the analysis. The protocol was approved by the institutional review board. Demographic and clinical data were obtained from the Department of Critical Care database, the Department of Pharmacy database, and the global institutional database (Enterprise Information Warehouse). The data were incorporated into a single spreadsheet using Excel 12.2 for Mac (Microsoft, Redmond, WA).

RIFLE categories for AKI were defined by the percent increase in serum creatinine from the time of ICU admission to the maximum creatinine at any point during the ICU stay: risk (≥50% rise in serum creatinine), injury (≥100% rise in serum creatinine), and failure (≥200% rise in serum creatinine). Consistent with the Acute Kidney Injury Network modifications of the original criteria, patients who required dialysis were classified into the RIFLE failure category, irrespective of the percent rise in serum creatinine.^[19] The modality for continuous renal replacement therapy used at our institution is continuous slow low-efficiency dialysis (c-SLED), which has been described previously.^[20] For patients who did not have an initial creatinine available within 24 hours after ICU admission, the most recent prior creatinine within the previous 48 hours was used.

Statistics

Descriptive data are presented as medians with interquartile ranges for continuous variables and absolute numbers with percentages for categorical variables. Survival of patients with AKI as defined by the RIFLE criteria was estimated by the Kaplan-Meier method. Patients were censored at death or last known follow-up, as determined by the clinical record. Statistical significance was determined by the log-rank test.

The primary end point for logistic regression was death at 60 days after ICU admission. Two separate models were developed, examining AKI as a categorical variable (RIFLE categories) and as a continuous variable (percent increase in creatinine from baseline). The variable “age” was significantly associated in a linear fashion with log odds of death but was dichotomized to provide a more meaningful odds ratio for the reader. Correlated data were assessed by correlation coefficients, and no variables were significantly correlated >0.6. Model reduction was achieved by variable elimination using the likelihood ratio test between nested models. Predictive ability and goodness-of-fit statistics were calculated, and the model was internally validated. No significant interactions were identified in either logistic regression model.

Lastly, a multivariate log linear regression model was developed to assess the relationship of AKI and dialysis with hospital cost. Cost was defined as hospital charges from the provider perspective. Log transformation of “cost” was used to account for skewness and heteroskedasticity. Coefficients in this model were multiplied by a factor of 100 to estimate a percent change in the dependent variable (cost) associated with a unit change in the independent variable.^[21]

A two-tailed P < 0.05 was considered statistically significant. No patients were excluded from the analysis because of missing data. Statistical analysis was performed with Stata 10 for Mac (StataCorp, College Station, TX).

Results

The data set included 2,398 patients. Patient characteristics are listed in Table 1. The median age was 59 years. The cohort was predominantly Caucasian (75%) and relatively balanced with respect to gender. The majority of patients on a medical service were admitted to the hospital from the emergency room (76%), compared to only 10% of patients on a surgical service. Sepsis was diagnosed in 23% of patients on a medical service vs. only 4% of patients on a surgical service. This is consistent with the large number of patients at our institution who were admitted to the ICU for routine monitoring after elective surgeries. A significant number of patients had underlying hypertension and diabetes (54% and 18%, respectively). One-third of patients had advanced malignancy by Surveillance, Epidemiology, and End Results (SEER) stage on initial presentation to our institution.