Acute renal failure (ARF) is a serious complication in hospitalized patients, with single-center studies reporting a rising incidence over the past two decades (1,2) and case fatality rates exceeding 50% among those who require dialytic support (3–10). Whereas these single-center reports might have excelled in the systematic collection of epidemiologic data, the results are not generalizable to the population at large and might reflect known and unknown biases and confounders inherent to the reporting institution. Consequently, accurate estimates of the incidence of ARF and its impact on short-term and intermediate outcomes in a national representative sample are lacking.
Large administrative databases can identify individuals and hospital admissions for population-based research endeavors and have been used in studies of various medical conditions, including sepsis (11), heart failure (12), and cirrhosis (13). The aim of this study was to provide a broad characterization of the epidemiology of ARF among adults (age ≥ 18 yr) who were hospitalized in the United States, using the National Hospital Discharge Survey (NHDS), a nationally collected database for use in epidemiologic estimates, and to identify specific groups with an altered susceptibility to the development of ARF. This study also examined whether ARF influences hospital length of stay (LOS), mortality, and patient disposition at time of hospital discharge, mainly transfers to short- and long-term care facilities.
Materials and Methods
Data Source
For this analysis, we used the 2001 NHDS database, which was acquired from the National Center for Health Statistics (14). In brief, this annual survey comprises a sample of all nonfederal acute care hospitals (with an average patient LOS of <30 d) in the United States and includes approximately 500 hospitals, with equal representation from all geographic regions (14). This survey excludes federal, military, and Department of Veterans Affairs hospitals; institutional hospital units (e.g., prison hospitals); and hospitals with fewer than six beds (14).
The database is constructed through the survey of discharge records for inpatients from each participating hospital, representing approximately 1% of all hospitalizations, or 330,210 discharges annually in the United States. The data set includes weighted frequencies, which can be generalized to the entire population. The weighted sample of approximately 1% hospital discharges projects to a national estimate of 29,039,599 discharges.
Discharge records are abstracted for demographic information (age, gender, ethnic background, geographic location, and marital status), seven diagnostic codes, four procedural codes, date of hospital admission and discharge, LOS, sources of payment, and disposition at discharge. The diagnosis and procedure codes are derived from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) (15).
Case Definitions
Cases were identified from discharge records in the NHDS, which included a code for ARF. ARF was defined by the presence of any of the following ICD-9-CM codes: 584.0 (ARF, unspecified), 584.5 (acute tubular necrosis), 584.6 (cortical ARF), 584.7 (medullary ARF), 584.8 (ARF with other specified pathologic lesion), and 584.9 (ARF not otherwise specified). Discharge records with an ICD-9-CM code for chronic dialysis status (V45.1, V56.0, V56.31, V56.32, or V56.8) were excluded. Requirement for acute dialysis was defined by one of two codes: 39.95 or 54.98.
Coexisting conditions were defined by the following ICD-9-CM codes: Diabetes (250 and 250.01 to 250.03), hypertension (401.0, 401.1, and 401.9), coronary artery disease (414, 414.02, 414.03, and 414.9), congestive heart failure (428), chronic kidney disease (403.11, 403.91, 404.13, 404.92, 404.93, 250.40 to 250.43, 581.0 to 581.3, 581.81, 581.89, 581.9, 582, 582.1, 582.2, 582.81, 582.89, 582.9, 583.0 to 583.6, 583.81, 583.89, 583.9, and V42.0), chronic lung disease (chronic obstructive lung disease [492.8] and emphysema [496]), cancer (colon [153.9], breast [174.9], lung [162.9], prostate [185], melanoma [172.9], myeloma [203], kidney [189], and bladder [188.9]), and HIV infection (042).
Acute hospital-related factors were defined by the following diagnosis and procedure ICD-9-CM codes: Sepsis (038, 020.0, 790.7, 117.9, 112.5, and 112.81) (11), cardiac catheterization (37.22 and 37.23), and cardiac surgery (coronary artery bypass graft surgery [36.10 to 36.16] and valve repair [35.1 to 35.2, 35.11 to 35.14, and 35.21 to 35.28]). Acute organ system dysfunction (AOSD) was defined using a previously published ICD-9-CM–based classification (11) with some minor modifications, and comprised the respiratory, cardiovascular, hepatic, hematologic, and neurologic organ systems (Appendix 1).
Validation of ARF ICD-9-CM Case Definitions
The accuracy of ICD-9-CM for ARF was validated by reviewing the records of all patients who were discharged from a tertiary medical center (Caritas St. Elizabeth’s Medical Center, Boston, MA) during 2001 (n = 13,412). The Human Investigation Review Committee approved this validation study. Admissions involving patients with ESRD (n = 175) were excluded from further analysis. We compared ARF diagnostic coding (defined by the presence or absence of ICD-9-CM codes 584.5 to 584.9) with well-established criteria for ARF that are based on changes in serum creatinine (1). Nadir and peak serum creatinine values over the course of the admission were used to establish the presence of ARF on the basis of a rise in serum creatinine level of 0.5, 1, or 1.5 mg/dl from nadir values of ≤1.9, 2.0 to 4.9, and ≥5.0 mg/dl, respectively.
Statistical Analyses
Continuous data are presented as medians with interquartile range, and nominal data are presented as percentages. All statistical analyses were performed using SAS (SAS Institute, Cary, NC) version 9.1. Following the NHDS guidelines, data with a sample size of <60 or with a relative SE >30% were not used for the analyses. Unless specified, all descriptive statistics and analyses were weighted to allow inflation to national estimates, based on the weight variable provided by the NHDS. Of note, because of this very large sample size, this study has extremely high power, and P values are expected to be significant, regardless of magnitude or clinical relevance.
Kaplan-Meier survival analysis was used to compare hospital LOS (in weeks) among patients with and without ARF. This analysis was stratified by presence or absence of other nonrenal AOSD. The log-rank statistic was used to test differences between groups. In this analysis, death was censored, and patients with missing discharge status were excluded.
Logistic regression analysis was used to explore the relationship between ARF and hospital death and patient disposition at time of hospital discharge. The analysis was adjusted in a stepwise manner for the following covariates: Demographic factors (age, gender, ethnic background, and source of payment), coexisting conditions, acute hospital-related factors, and number of AOSD. The end points were ascertained at time of hospital discharge, no censoring was required, and there was no loss of follow-up. Twenty-four percent of discharges were excluded from this analysis as a result of missing values for either the predictor or outcome variables. Most of the missing values (96%) were on the ethnicity variable. However, the results were not significantly different when missing ethnicity was included as an indicator variable.
A separate robust regression model explored the relationship of ARF and other AOSD to hospital LOS. This regression method is more appropriate than linear regression for outcome variables with a skewed distribution (16). In brief, this analysis uses the median of the squared residuals instead of the sum of the squared residuals. Admissions of <24 h were assigned a LOS of 1 d. Although records with missing data for either predictor or outcome variables were excluded from the final analysis, the results were very similar when those with missing ethnicity were included as an indicator variable.
Results
Cohort Characteristics
Approximately 1.9% of hospitalizations in the United States included a discharge diagnosis of ARF during the study period of 2001. The characteristics of patients with and without a discharge diagnosis of ARF are summarized in Table 1. ARF was more common in older patients, men, and black patients. A discharge diagnosis of ARF was also more commonly assigned to individuals with a coexisting diagnosis of congestive heart failure, chronic lung disease, chronic kidney disease, cancer, and HIV infection but less commonly to individuals with a coexisting diagnosis of coronary artery disease, diabetes, and hypertension. Patients with a diagnosis code of ARF were also more likely to have a diagnosis of sepsis or have undergone cardiac surgery. Finally, patients with an ARF diagnosis code were more likely to have other nonrenal AOSD. A total of 7.5% of patients with a diagnosis of ARF required dialysis. Whereas overall hospital mortality was 2.6%, patients who were discharged with a diagnosis of ARF had a higher mortality rate compared with those without ARF (21.3 versus 2.3%; P < 0.0001).
ARF and Hospital LOS
As shown in Table 1, hospital LOS was significantly higher among patients with a discharge diagnosis of ARF. Table 2 displays the LOS data for different diagnostic categories and is stratified according to survivor status. Compared with other single AOSD, a discharge diagnosis of ARF per se was associated with the longest median LOS (7 d), whereas acute cardiovascular system dysfunction was associated with the shortest LOS, which might be due to a high early fatality rate (see also Figure 1). The presence of ARF coupled to single or multiple, nonrenal AOSD was associated with the longest hospital LOS (median 10 d). Of note, the presence of ARF alone and without other nonrenal AOSD was associated with longer LOS than for patients with single nonrenal AOSD (Table 2). ARF was the second most frequent AOSD (1.9%), after acute respiratory system dysfunction (2.3%).
Kaplan-Meier survival analysis demonstrated that the sole presence of ARF, as well as its addition to other nonrenal AOSD, confers the most prolonged hospital LOS (P < 0.0001 by log rank test; Figure 2). In addition, dialysis-requiring ARF was associated with a more prolonged LOS compared with hospitalization in which ARF did not require dialysis (P < 0.0001 by log rank test; Figure 3).
On multivariate analysis (Table 3), a discharge diagnosis of ARF was associated with an estimated prolongation of hospital LOS by 2 d (P < 0.0001) and was surpassed only by cardiac surgery (3.9 additional days) and sepsis (2.6 additional days). Several other groups of diagnosis codes were also associated with LOS but to a lesser degree, including a discharge diagnosis of chronic kidney disease (0.2 additional day). Factors associated with a shorter LOS included a discharge diagnosis of coronary artery disease, hypertension, and cardiac catheterization.
ARF and Hospital Discharge Disposition
The characteristics of patients who had ARF and were discharged from the hospital alive are shown in Table 4 and are stratified according to disposition. Compared with patients who had ARF and were discharged home, those who were discharged to a short-term care facility were older and had a higher frequency of congestive heart failure, chronic lung disease, sepsis, and acute respiratory system dysfunction. However, compared with patients who had ARF and were discharged to a short-term care facility, those who were discharged to a long-term care facility were older by a median of 6 yr, more likely to be female, and more likely to have a coexisting diagnosis of HIV infection and acute cardiovascular or neurologic system dysfunction.
On multivariate analysis (Table 5), ARF was associated with a 2.0-fold higher odds for transfer to a short- or long-term care facility after adjustment for age, gender, ethnic background, payment source, coexisting conditions (including chronic kidney disease), acute hospital-related factors, and other AOSD.
ARF and Hospital Death
On multivariate analysis (Table 5), ARF was associated with an eight-fold higher odds for death after adjustment for age, gender, ethnic background, and payment source. This association was weakened but persisted with 4.1-fold higher odds for death after adjustment for age, gender ethnic background, payment source, coexisting conditions (including chronic kidney disease), acute hospital-related factors, and other AOSD.
ARF Coding Validation
We performed an audit of ARF coding practices by reviewing the electronic discharge records of all patients who were discharged from our tertiary medical center during the 2001 calendar year. After exclusion of the discharges of 175 patients with ESRD, 13,237 records were available for analysis. Using serum creatinine–based laboratory criteria, ARF supervened on 1584 hospital admissions (12.0%). An ICD-9-CM code for ARF was present in 347 (2.6%) discharge records during this period. Acute renal failure was confirmed by biochemical criteria in 304 of these cases, yielding a positive predictive value of 87.6%. However, the sensitivity of ARF ICD-9-CM coding was 19.2% (304 of 1584). Among the 12,890 discharge records without an ARF ICD-9-CM code, serum creatinine criteria for ARF were not met in 11,610, yielding a negative predictive value of 90.1%. The specificity of ARF coding was 99.6%.
Discussion
Administrative data sets such as the NHDS, although potentially less precise with regard to diagnostic accuracy and the relationship of cause and effect than patient-oriented clinical studies, can provide valuable, nationally relevant data that could improve our understanding of the impact of ARF on the entire US hospital system. Registries that are based on laboratory results would represent a much more precise method of capturing this information, but the development of such registries would require significant resources and a concerted effort of all regional health care provider systems. The analysis of existing administrative databases could provide an overview of the importance of ARF within a health care system such as that of the United States and could support the premise that ARF is an important factor for health care resource utilization.
In the 2001 NHDS cohort, the overall incidence of ARF was 1.9%, which is lower than that observed in studies of single academic centers (1,2). It is conceivable that academic tertiary hospitals likely provide care to patients with greater severity of illness and in whom ARF might be more likely to develop. However, these discrepancies are more likely due to differences in case definitions, contrasting clinical with administrative diagnosis of ARF. In contrast to results from single-center studies, the NHDS data are collected from a representative sample of hospitals that reflect a broad spectrum of care throughout the United States. The observed incidence of ARF in our study is considerably higher than the 0.4% reported by Liano et al. (17) in a population-based study from Madrid, Spain. This is likely due to the adoption of a more specific but less sensitive definition of ARF by these authors, which required a sudden elevation in serum creatinine concentration by 2 mg/dl from a baseline of <3 mg/dl with subsequent recovery of at least 50% of kidney function at time of hospital discharge. The NHDS analysis was based on ICD-9-CM codes, which likely enabled a more liberal definition of ARF and therefore cannot exclude disease misclassification.
The hospital mortality rate of patients with ARF was 21%, which is similar to the 19% rate reported by a large prospective cohort study that was conducted at a US urban academic center (2). Notably, this mortality rate was lower than that observed in several other studies, primarily focused on critically ill patients with a higher degree of disease severity and a less favorable outcome (3,7,18,19). Our results suggest that the high mortality associated with ARF is observed not only in tertiary care hospitals but also throughout the entire spectrum of acute care hospitals in the United States, and our data support an independent impact of ARF on mortality, after adjustment for several factors, including coexisting conditions and concomitant single or multiple nonrenal AOSD, and thus support the generalizability of the results of smaller, single-center, patient-oriented studies (3,6,20,21).
In this study, presence of a discharge diagnosis of ARF was associated with prolonged hospital LOS, surpassing acute respiratory system dysfunction (Figure 1). Furthermore, the combination of ARF and multiple AOSD was associated with the highest hospital LOS burden, particularly among survivors. Our study also demonstrates an association between assignment of an ARF diagnosis and discharge to short- and long-term care facilities. This suggests that ARF independently and adversely affects recovery of physical function after an acute illness, irrespective of kidney function recovery. Consequently, ARF may be associated with a substantial cost burden on the health care system that persists after hospital discharge. To our knowledge, this association has not been described previously.
The lower prevalence of hypertension among patients with a discharge diagnosis of ARF may be caused by less frequent coding for less threatening chronic conditions such as hypertension in the setting of a severe acute illness with numerous complications as described previously (22). The association of hypertension, coronary artery disease, and cardiac catheterization with a shorter LOS likely reflects the typical short hospitalizations associated with cardiac catheterizations for diagnostic or therapeutic purposes.
The primary limitation of this study was our dependence on ICD-9-CM codes to ascertain the diagnosis of ARF. As administrative coding is dependent on accurate documentation by health care professionals and hospital “coders,” there is an unquestionable potential for misclassification. As others have done for congestive heart failure (23) and sepsis (11), we attempted to validate ICD-9-CM coding for ARF at our institution. We demonstrated that ARF codes were effective at identifying patients with and without ARF (positive predictive value 87.6% and negative predictive value 90.1%, respectively). However, approximately 80% of patients with biochemical criteria for ARF were not identified by ICD-9-CM codes. If this is truly representative of coding practices across the United States, then the marked impact of ARF that we described may have been relatively conservative. When we projected the test characteristics from our single-center validation study to the entire NHDS population, the positive predictive value of ARF coding was only 50.1% with a negative predictive value of 98.5%. However, it is unclear whether extrapolating the test characteristics from a single institution to the entire NHDS cohort is appropriate.
There are other shortcomings that require mention. The NHDS collects data on hospital discharges and not individual patients. Thus, a patient may be captured on multiple occasions as a result of frequent admissions. The NHDS contains only seven diagnosis codes, which may or may not contain the relevant coexisting conditions for which we attempted to adjust. In addition, because of the large sample size, small differences in the characteristics of patients with and without ARF might be statistically significant but not of clinical significance. Therefore, care must be taken in the interpretation of these differences as to their clinical importance and relevance.
In summary, this study demonstrates that using the NHDS, a nationwide administrative database, the existence of ARF in a hospital discharge record is independently associated with adverse outcomes, including higher hospital-associated mortality. Furthermore and more importantly, hospitalizations in which ARF is reported are associated with a prolonged LOS and with a greater likelihood of discharge to short-term and long-term care facilities. These factors suggest higher costs associated with the care for patients with ARF and a significant burden that ARF places on the health care system. Future research should focus on the ascertainment of more accurate national estimates of ARF and better describe the relative contribution of ARF to the utilization of health care resources in the United States today. Such research endeavor would underscore the urgent need to improve health care delivery by allocating research budgets to the development of more effective preventive strategies and to the early diagnosis and timely management of ARF, particularly in high-risk populations.
;)
Figure 1: Median hospital length of stay (LOS) stratified by single acute organ system dysfunction (AOSD), including acute renal failure (ARF).
Figure 2: Kaplan-Meier analysis for the cumulative percentage of patients who remained hospitalized according to presence or absence of ARF with or without other single or multiple AOSD. (A) No ARF and no other single AOSD (hatched line), and ARF and no other single AOSD (solid line), P < 0.0001 by log rank test. (B) No ARF but presence of other single AOSD (hatched line), and ARF and no other single AOSD (solid line), P < 0.0001 by log rank test. (C) No ARF but presence of other single AOSD (hatched line), and ARF and presence of other single AOSD (solid line), P < 0.0001 by log rank test. (D) No ARF but presence of other multiple AOSD (hatched line), and ARF and presence of other multiple AOSD (solid line), P < 0.0001 by log rank test.
Figure 3: Kaplan-Meier analysis for the cumulative percentage of patients who had ARF and remained hospitalized according to dialysis requirement. ARF requiring dialysis (hatched line) and ARF not requiring dialysis (solid line), P < 0.0001 by log rank test.
Table 1: Characteristics of patients with and without a discharge diagnosis of ARF a
Table 2: Hospital LOS stratified by ARF, AOSD, and survival status
Table 3: Multivariate (robust regression) analyses examining the association of ARF and other covariates with hospital LOS a
Table 4: Characteristics of patients who had ARF and were discharged from the hospital alive and stratified by disposition a
Table 5: Multivariate (logistic regression) analyses examining the association of ARF with hospital death and discharge disposition a
Appendix: ICD: 9-CM–and CPT-based classification of AOSD (
11)
B.L.J. was supported by National Institutes of Health Grant K23 DK065102-01. O.L. was supported by the Inter-Institutional Clinical Research Fellowship Program of Tufts-New England Medical Center and Caritas St. Elizabeth’s Medical Center (Boston, MA).
This work was presented in part at the 37th Annual Meeting of the American Society of Nephrology, October 27 to November 1, 2004, in St. Louis, MO.
Published online ahead of print. Publication date available at http://www.cjasn.org.
O.L., R.W., and J.W.O. contributed equally to this study.
References
1. Hou SH, Bushinsky DA, Wish JB, Cohen JJ, Harrington JT: Hospital-acquired renal insufficiency: A prospective study. Am J Med 74: 243–248, 1983
2. Nash K, Hafeez A, Hou S: Hospital-acquired renal insufficiency. Am J Kidney Dis 39: 930–936, 2002
3. Shusterman N, Strom BL, Murray TG, Morrison G, West SL, Maislin G: Risk factors and outcome of hospital-acquired acute renal failure. Clinical epidemiologic study. Am J Med 83: 65–71, 1987
4. Kellum JA, Angus DC: Patients are dying of acute renal failure. Crit Care Med 30: 2156–2157, 2002
5. Brivet FG, Kleinknecht DJ, Loirat P, Landais PJ: Acute renal failure in intensive care units—Causes, outcome, and prognostic factors of hospital mortality; a prospective, multicenter study. French Study Group on Acute Renal Failure. Crit Care Med 24: 192–198, 1996
6. Levy EM, Viscoli CM, Horwitz RI: The effect of acute renal failure on mortality. A cohort analysis. JAMA 275: 1489–1494, 1996
7. Clermont G, Acker CG, Angus DC, Sirio CA, Pinsky MR, Johnson JP: Renal failure in the ICU: Comparison of the impact of acute renal failure and end-stage renal disease on ICU outcomes. Kidney Int 62: 986–996, 2002
8. Metnitz PG, Krenn CG, Steltzer H, Lang T, Ploder J, Lenz K, Le Gall JR, Druml W: Effect of acute renal failure requiring renal replacement therapy on outcome in critically ill patients. Crit Care Med 30: 2051–2058, 2002
9. Chertow GM, Lazarus JM, Christiansen CL, Cook EF, Hammermeister KE, Grover F, Daley J: Preoperative renal risk stratification. Circulation 95: 878–884, 1997
10. Mehta RL, Pascual MT, Soroko S, Savage BR, Himmelfarb J, Ikizler TA, Paganini EP, Chertow GM: Spectrum of acute renal failure in the intensive care unit: The PICARD experience. Kidney Int 66: 1613–1621, 2004
11. Martin GS, Mannino DM, Eaton S, Moss M: The epidemiology of sepsis in the United States from 1979 through 2000. N Engl J Med 348: 1546–1554, 2003
12. Koelling TM, Chen RS, Lubwama RN, L’Italien GJ, Eagle KA: The expanding national burden of heart failure in the United States: The influence of heart failure in women. Am Heart J 147: 74–78, 2004
13. Foreman MG, Mannino DM, Moss M: Cirrhosis as a risk factor for sepsis and death: Analysis of the National Hospital Discharge Survey. Chest 124: 1016–1020, 2003
14. National Hospital Discharge and Ambulatory Surgery Data. Available:
http://www.cdc.gov/nchs/about/major/hdasd/nhdsdes.htm. Accessed June 22, 2005
15. International Classification of Diseases, Ninth Revision, Clinical Modification, Salt Lake City, Medicode, Inc., 2004
16. Rousseeuw PJ: Least median of squares regression. J Am Stat Assoc 79: 871–880, 1984
17. Liano F, Pascual J: Epidemiology of acute renal failure: A prospective, multicenter, community-based study. Madrid Acute Renal Failure Study Group. Kidney Int 50: 811–818, 1996
18. Morgera S, Kraft AK, Siebert G, Luft FC, Neumayer HH: Long-term outcomes in acute renal failure patients treated with continuous renal replacement therapies. Am J Kidney Dis 40: 275–279, 2002
19. Mehta RL, Pascual MT, Gruta CG, Zhuang S, Chertow GM: Refining predictive models in critically ill patients with acute renal failure. J Am Soc Nephrol 13: 1350–1357, 2002
20. Chertow GM, Levy EM, Hammermeister KE, Grover F, Daley J: Independent association between acute renal failure and mortality following cardiac surgery. Am J Med 104: 343–348, 1998
21. Chertow GM LE, Hammermeister KE, Grover F, Daley J: Independent association between acute renal failure and mortality following cardiac surgery. Am J Med 104: 343–348, 1998
22. Iezzoni LI, Foley SM, Daley J, Hughes J, Fisher ES, Heeren T: Comorbidities, complications, and coding bias. Does the number of diagnosis codes matter in predicting in-hospital mortality? JAMA 267: 2197–2203, 1992
23. Lee DS, Donovan L, Austin PC, Gong Y, Liu PP, Rouleau JL, Tu JV: Comparison of coding of heart failure and comorbidities in administrative and clinical data for use in outcomes research. Med Care 43: 182–188, 2005
