IT is devastating to both patients and anesthesiologists when patients, with no evidence of renal dysfunction preoperatively, develop acute renal failure (ARF) after surgery. What could have been done differently to avoid this complication? This question has been explored extensively in cardiac surgical patients, and many preoperative risk stratifications and intraoperative management protocols for this unique patient population have been proposed and tested. However, until now, no study has evaluated risk factors for ARF in a large general surgical population. In this issue of the Journal, Kheterpal et al.1
attempt to identify risk profiles in the large general surgical population of the University of Michigan and have reported their analyses of perioperative data collected over a recent 3-yr period.
This interesting study is based on information retrieved from a computerized database used for all surgical patients at the university between 2003 and 2006. A total of 65,043 patient charts were screened, and 15,102 episodes of care were included in the study. Kheterpal et al.1
evaluate the propensity of patients with certain risk factors or exposures to specific intraoperative variables to develop ARF. After extensive analyses, they developed propensity scores related to specific preoperative risk factors or their combinations that might help to predict the risks of unique patients to develop ARF after surgery. They also observed that the perioperative onset of ARF in patients with previous normal renal function was associated with increased postoperative mortality. The risk profiles for perioperative ARF and the association between perioperative ARF and mortality have been previously reported in cardiac surgical patients but not in a general, noncardiac surgical population.
This information is useful if it can help us to assess the preoperative risk of our patients and, hopefully, prevent their development of ARF. An appraisal of the value of this study to clinical practice requires critical scrutiny of their data and a general evaluation of the nature of observational studies. These are the important questions that need to be asked.
First, are these results applicable beyond the University of Michigan? By the nature of their dominant referral practices, academic medical centers tend to attract surgical populations that do not match well with nonacademic practices. In this case, the authors have reported their results as observations. A commonly understood strength of observational study designs and reports is their higher likelihood of generalizability.2
This aspect is particularly good in this study because preoperative renal function was assessed and found to be within normal ranges in the final selected population. There are, however, a few areas of concern. No preoperative renal function measures were made for 6,534 of the patients in the database and available for study; therefore, they were excluded from the study. If these patients had a low risk of developing ARF, their exclusion may have falsely increased the calculated risk of developing ARF in a general surgical population. Neither American Society of Anesthesiologists physical status, a simple method of assessing generalizability, nor ethnicity, hypothesized to affect a person’s risk for ARF, is reported. In addition, definitions for certain comorbidities and risk factors used during the study period do not seem to have been very precise, and there is no evidence that practitioners at the university have undergone specific training in their application. For example, what constituted “coronary artery disease” in these patients? How was pulmonary hypertension defined? Validity testing of this database has not been reported; therefore, the results must be interpreted cautiously. Fortunately, the large patient population and the lack of nonrenal limitations for inclusion into the study population provide readers with a sense of reassurance that the study’s results are reasonably generalizable.
Second, are the application and interpretation of statistical tests appropriate for this type of study? Propensity score methodology, an approach generally used to strengthen causal claims in observational studies, was applied to determine seven independent risk factors for developing postoperative ARF and, ultimately, a risk stratification tool. As with all procedures for statistical evaluation, the analysis of variability with propensity scores has limitations. In the case of propensity score methodology, the analysis is limited by the omission of unknown independent variables or risk factors. It is not possible to know what other variables existed that should have been included in the models. However, the observational design used in this study is a reasonable approach for this type of study, and propensity score modeling is an appropriate test for statistical evaluation. As more institutions develop and acquire the ability for computerized medical records and share their information, we can expect more opportunities for large observational studies. We will need to continue to better understand the nature and limitations of these types of studies as we attempt to determine the applicability of their results to various clinical practices.
The most useful conclusions of the article by Kheterpal et al. seem to center around the identification of unique risk factors, one of which is body mass index greater than 32 kg/m2, with this factor being independent of diabetes or hypertension. Obese patients come with their own unique set of challenges, and this finding reinforces the overall increased risk associated with obesity. Liver diseases and chronic obstructive pulmonary disease necessitating chronic bronchodilator therapy are also identified as independent risk factors; however, neither of these factors, as they have been analyzed, is defined sufficiently well enough to help distinguish between the perioperative ARF risks of patients with varying degrees of dysfunction (e.g., modest compared with severe) or types of disease (e.g., cirrhosis compared with hepatitis for liver disease). Not surprisingly, risk of postoperative ARF is found to rise significantly with increasing number of risk factors. The hazard ratio for developing ARF with three or more of the seven most highly associated risk factors was found to be 16.0 (95% confidence interval, 8.9–28.8), a clinically significant ratio. Even with the noted limitations and a modest sensitivity and specificity (i.e., an area under the curve of 0.73 ± 0.03; a very sensitive and specific association would have a value approaching 1), the proposed scoring system has the potential to improve our patient care. The risk identification and stratification can be applied as a guide for preoperative discussions with patients, as well as a tool to guide further quality assessment and improvements within individual practices.
Finally, if this study only identifies potential risk factors, why should any of these findings matter to busy practitioners? The identification of potential risk factors in this observational study provides information that clinical researchers can use to design additional prospective studies that will help to elucidate the etiology of ARF in unique patient groups and generate hypotheses and clinical trials on how to avoid this complication in the future. Reducing the surprisingly high frequency of ARF in our general noncardiac surgical populations is important; the authors have shown a significantly increased frequency of postoperative death within 1 yr in their surgical patients who developed postoperative ARF. Yes, there are issues that weaken their mortality conclusions. Patients who developed ARF had significantly higher body mass indices and proportionately more chronic obstructive pulmonary disease than those who did not develop ARF. Nonetheless, their finding that all-cause 30-day, 60-day, and 1-yr mortalities are quite a bit higher in patients with perioperative ARF is compelling and worthy of additional study. Kheterpal et al. should be congratulated for carefully using their institution’s large databases and reporting potential risk factors that we hope will lead to further investigations and, someday in the future, improvements that may decrease this major perioperative complication.
Pamela C. Nagle, M.D.,*
Mark A. Warner, M.D.†
*Department of Anesthesiology, Wake Forest University School of Medicine, Winston-Salem, North Carolina. firstname.lastname@example.org. †Department of Anesthesiology, Mayo Clinic College of Medicine, Rochester, Minnesota.
1. Kheterpal S, Tremper KK, Englesbe MJ, O’Reilly M, Shanks AM, Fetterman DM, Rosenberg AL, Swartz RD: Predictors of postoperative renal failure after noncardiac surgery in patients with previously normal renal function. Anesthesiology 2007; 107:892–902
2. Kirk RE: Experimental Design: Procedures for the Behavioral Sciences. New York, Brooks/Cole Publishing, 1995
© 2007 American Society of Anesthesiologists, Inc.