We included AKI as a continuous variable in a multivariate regression to determine the relationship of AKI and dialysis with hospital cost (Table 4). The model was adjusted for numerous clinical and demographic variables. Age, gender, race, autologous transplant, tumor grade, diabetes, and liver disease were not significant predictors of hospital cost in the final model. The need for dialysis was associated with a 21% increase in hospital cost. Each percent increase in serum creatinine was associated with a 0.16% increase in cost. An interaction was identified between mechanical ventilation and sepsis (25% increase in hospital cost).
Table 3. Multivariate Logistic Regression for Predictors of Death at 60 Days after ICU Admission (AKI Included as a Continuous Variable) (n = 2,398)
| VARIABLE | OR | 95% CI | P |
|---|---|---|---|
| Age ≥55 years | 1.4 | 1.1–1.9 | <0.001 |
| Percent increase in creatinine | 1.008 | 1.005–1.01 | <0.001 |
| ER admission | 5.4 | 3.8–7.7 | <0.001 |
| Pre-ICU length of stay (days) | 1.02 | 1.00–1.04 | 0.016 |
| SEER stage (distant vs. other) | 2.0 | 1.6–2.7 | <0.001 |
| Medical vs. surgical service | 2.2 | 1.5–3.2 | <0.001 |
| Vasopressors | 2.0 | 1.5–2.7 | <0.001 |
| Mechanical ventilation | 1.8 | 1.4–2.5 | <0.001 |
| Amphotericin | 1.8 | 1.1–3.2 | 0.031 |
| IV diuretics | 1.4 | 1.0–1.8 | 0.024 |
| Dialysis | 6.2 | 2.3–16.5 | <0.001 |
Likelihood ratio x2(11) = 815 (P < 0.001), positive predictive value 72%, negative predictive value 88%, area under the receiver operating curve = 0.88.
OR, odds ratio; ICU, intensive care unit; AKI, acute kidney injury; ER, emergency room.
Table 4. Multivariate Log-Linear Regression Predicting Hospital Cost (Log Dollars) (n = 2,398)
| VARIABLE | β | SE | P |
|---|---|---|---|
| Increase in creatinine (per 1%) | 0.00156 | 0.000257 | <0.001 |
| Dialysis | 0.213 | 0.0994 | 0.032 |
| Diuretics | 0.0831 | 0.0180 | <0.001 |
| Mechanical ventilation | 0.561 | 0.0299 | <0.001 |
| Allotransplant | 0.538 | 0.0960 | <0.001 |
| Medical vs. surgical service | 0.259 | 0.0381 | <0.001 |
| Liquid vs. solid tumor | 0.227 | 0.0433 | <0.001 |
| Distant vs. locoregional stage | 0.0717 | 0.0314 | 0.023 |
| Sepsis | 0.151 | 0.0622 | 0.015 |
| ER admission | −0.246 | 0.038 | <0.001 |
| Heart failure | 0.107 | 0.0469 | 0.023 |
| Hypertension | 0.0647 | 0.0271 | 0.017 |
| Mechanical ventilation × sepsis | 0.251 | 0.0853 | 0.003 |
| Constant | 10.8 | 0.0254 | <0.001 |
R2 = 0.32.
Discussion
The incidence of AKI in our study was 12.6% of all patients admitted to the ICU, and there was a progressive decrease in survival associated with worsening kidney injury. This association remained even after adjusting for covariates. AKI and the need for dialysis were also associated with increased hospital costs. To our knowledge, this is the largest single-center study to examine the RIFLE criteria for AKI in a critically-ill population with cancer.
A striking finding in our study is the significant effect that small elevations in serum creatinine may have on survival. An increase of 0.6 mg/dL in the RIFLE risk category increased the odds for mortality by a factor of 2.3 compared to patients without AKI. The median maximum creatinine in this group was only 1.3 mg/dL, which is still within the “normal” range for males in our institution. Criteria that define mild renal toxicity as a serum greater than “1.5 × the upper limit of normal” would exclude a significant number of patients in the RIFLE risk and injury categories, although their risk of mortality was significantly increased.[22] Other criteria that define AKI by glomerular filtration rate (GFR) are also problematic as estimating equations for GFR require serum creatinine to be in steady state. This is a false assumption to make in the setting of AKI, where serum creatinine may fluctuate daily. Serum creatinine is an insensitive marker of renal injury in patients with cancer, and more sensitive and specific biomarkers of AKI are currently under development.[23], [24] and [25] Until these markers are routinely available, renal injury in oncology practice and clinical trials may be better defined as a percentage rise in serum creatinine relative to baseline, similar to the RIFLE criteria.
Out of all variables examined, it is interesting that the need for dialysis had the greatest association with 60-day mortality (Table 3). Although we adjusted for other risk factors, there may still be residual confounding to explain the strong association of dialysis with mortality. However, it is also recognized that dialysis may promote a proinflammatory state[26] and that AKI, in itself, may lead to injury of distant organs via systemic cytokine release.[27] and [28] These deleterious effects may be amplified in patients with cancer, who frequently are neutropenic and have chronic inflammation (e.g. capillary leak syndrome, diffuse alveolar hemorrhage, graft-vs.-host disease). It is known that the need for dialysis after a stem cell transplant is associated with >70% mortality.[29] Although dialysis remains pivotal for volume and metabolic clearance, a true “therapy” for AKI has unfortunately remained elusive thus far.
Our overall incidence of 12.6% for AKI is lower than the reported incidence of 13%–42% in other studies of critically ill patients with cancer.[2], [30] and [31] We excluded patients who had a serum creatinine >1.5 mg/dL on admission to the ICU as we were interested in the development of AKI after ICU admission. This likely excluded patients who already had AKI on presentation, which may have contributed to the lower incidence of AKI and the need for dialysis in our study. Unlike previous studies, our cohort included a large number of patients on a surgical service who were electively admitted to the ICU for routine postoperative care and, therefore, were at lower risk of developing AKI. However, when limited to patients on a medical service, our incidence of 21% is consistent with the results of previous studies of patients in medical ICUs.
The prognosis of patients requiring dialysis was dismal, with an estimated 89% 60-day mortality. This is somewhat higher than the reported mortality of 66%–88% in previous studies.[32], [33], [34], [35] and [36] Given that our institution also serves as a referral cancer center for patients who have had progressive disease on standard therapy, it is possible that our patient population may have been more predisposed to complications from cancer therapy. Patients with hematological malignancies had a higher incidence of AKI and need for dialysis. However, underlying hematological malignancy and HCT were no longer significantly associated with 60-day mortality in the adjusted analysis. Similar to the findings of others, this would suggest that it is not the underlying malignancy itself but rather the complications of treatment and prolonged immunosuppression that lead to decreased survival in these patients.[37] and [38] Early goal-directed intensive life support should be considered for most patients,[39] but continuation of dialysis may not be of benefit, in terms of both survival and cost, in patients with hematologic malignancy who demonstrate minimal improvement.
Our study had certain limitations. Given the retrospective design, we cannot rule out selection bias or residual confounding. We were able to adjust for several variables specific to cancer and critical care as well as pre-ICU length of stay, which may be a surrogate marker for comorbidities and functional status. Nonetheless, our conclusions should be interpreted as hypothesis-generating. Second, our study is based on a single-center experience, which may limit its generalizability. Nonetheless, our study had a large sample size that was subjected to fairly uniform management. Third, we did not have data on end-of-life decisions, which may have impacted mortality and need for dialysis. Lastly, we were unable to obtain cost-to-charge ratios, which may limit the generalizability of our findings to other institutions. However, we reported on percent increases in cost as opposed to absolute dollar figures, which may adjust for some of this variation.
Conclusions
AKI as defined by the RIFLE criteria may be predictive of short-term mortality in critically ill patients with cancer. We have demonstrated that relatively small changes in serum creatinine are associated with higher mortality and that the need for dialysis entails a very poor prognosis. The mechanism behind the increased mortality in patients with hematological malignancies appears to be secondary to the associated complications of therapy, as opposed to the underlying cancer itself. We hypothesize that strategies to prevent the development of AKI and progression to dialysis dependence may improve survival. Whether the prevention of AKI translates to cost savings is also of interest.