- Email: [email protected]
Predicting Length of Stay Following Radical Nephrectomy Using the National Surgical Quality Improvement Program Database
Predicting Length of Stay Following Radical Nephrectomy Using the National Surgical Quality Improvement Program Database C. Adam Lorentz, Andrew K. Leung, Austin B. DeRosa, Sebastian D. Perez, Timothy V. Johnson, John F. Sweeney and Viraj A. Master* From the Departments of Urology and Surgery (SDP, JFS), Emory University, Atlanta, Georgia
Purpose: Length of stay is frequently used to measure the quality of health care, although its predictors are not well studied in urology. We created a predictive model of length of stay after nephrectomy, focusing on preoperative variables. Materials and Methods: We used the NSQIP database to evaluate patients older than 18 years who underwent nephrectomy without concomitant procedures from 2007 to 2011. Preoperative factors analyzed for univariate significance in relation to actual length of stay were then included in a multivariable linear regression model. Backward elimination of nonsignificant variables resulted in a final model that was validated in an institutional external patient cohort. Results: Of the 1,527 patients in the NSQIP database 864 were included in the training cohort after exclusions for concomitant procedures or lack of data. Median length of stay was 3 days in the training and validation sets. Univariate analysis revealed 27 significant variables. Backward selection left a final model including the variables age, laparoscopic vs open approach, and preoperative hematocrit and albumin. For every additional year in age, point decrease in hematocrit and point decrease in albumin the length of stay lengthened by a factor of 0.7%, 2.5% and 17.7%, respectively. If an open approach was performed, length of stay increased by 61%. The R2 value was 0.256. The model was validated in a 427 patient external cohort, which yielded an R2 value of 0.214. Conclusions: Age, preoperative hematocrit, preoperative albumin and approach have significant effects on length of stay for patients undergoing nephrectomy. Similar predictive models could prove useful in patient education as well as quality assessment. Key Words: kidney, nephrectomy, length of stay, hypoalbuminemia, risk
IN efforts to curb the rising cost of health care in the United States payers have developed systems such as managed care and accountable care organizations. This highlights the longtime focus on quality as a measure to drive resource conservation. LoS is a common indicator in quality assessment and improvement constructs.1e4 LoS can also contribute to decisions on pre-authorization and reimbursement
Abbreviations and Acronyms ACS ¼ American College of Surgeons ASA ¼ American Society of AnesthesiologistsÒ CHF ¼ congestive heart failure DRG ¼ Diagnosis-Related Group HCT ¼ hematocrit LoS ¼ length of stay NSQIP ¼ National Surgical Quality Improvement Program SIRS ¼ systemic inflammatory response syndrome Accepted for publication March 27, 2015. Study received institutional review board approval. ACS NSQIP and participating hospitals have not verified and are not responsible for the statistical validity of the data analysis or the conclusions derived by the authors. * Correspondence: Department of Urology, Emory University School of Medicine, 1365 Clifton Rd., B1400, Atlanta, Georgia 30322 (FAX: 404-778-4006; e-mail: [email protected]).
as it is one of the primary output elements of the various DRG classification systems and used widely by payers to stratify diagnoses and procedures for reimbursement. Prolonged LoS has been associated with increased resource consumption and complications.5,6 The cost of an inpatient stay from 2009 to 2010 grew by 5.1%, nearly twice the rate of inflation.7 The cost of an inpatient
0022-5347/15/1944-0923/0 THE JOURNAL OF UROLOGY® Ó 2015 by AMERICAN UROLOGICAL ASSOCIATION EDUCATION AND RESEARCH, INC.
http://dx.doi.org/10.1016/j.juro.2015.04.112 Vol. 194, 923-928, October 2015 Printed in U.S.A.
www.jurology.com
j
923
924
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
surgical admission grew by 6.4% during the same period.7 Inpatient services account for 21% of national health care expenditures.7 As spending continues to increase, LoS bears greater and greater importance. Furthermore, increased LoS places patients at greater risk for complications, leading to worse outcomes and increased system costs.8,9 Preoperative risk factors have an important role in predicting LoS and have been shown to be meaningful for overall hospital costs.6,10 Elucidating the most significant factors presents an opportunity to mitigate modifiable variables, and improve clinical decision making and quality assessment. Previous studies have explored variables associated with prolonged LoS.11e15 Such factors in common urological cases such as nephrectomy, prostatectomy and cystectomy include advanced patient age, dependent functional status, low serum albumin, low HCT and high creatinine.12,15 Some of these prior studies used the NSQIP database to study LoS determinants.11,15 The NSQIP is a robust multi-institutional database including hundreds of variables and a variety of procedures that was designed and validated for use in quality improvement and outcomes data analysis.16 The primary objective of this study was to identify preoperative risk factors that significantly contribute to prolonged LoS after radical nephrectomy. Our aim was to examine the predictive ability of preoperative factors, assess use of the NSQIP database for urological procedures and create a predictive model to elucidate the intrinsic variability in the LoS measure. Such models have become important in the shift toward a pay for performance reimbursement structure in which physicians will be expected to meet benchmark quality measurements.
METHODS American College of Surgeons NSQIP The ACS NSQIP and participating hospitals are the source of the data used in this study. The NSQIP is a prospective, multi-institutional database with cases representing a sample of operations at each institution.5 For each patient more than 150 variables are captured, including comorbidities, laboratory values and LoS. Data are audited annually for accuracy and completeness.
Training Set Patient Population The primary cohort consisted of 1,527 patients identified from the NSQIP PUF (Participant Use Data File) who were older than 18 years and underwent radical nephrectomy between 2007 and 2011 via a laparoscopic approach (CPT code 50545/50546) or an open approach (CPT code 50220/50225). Patients treated with regional lymphadenectomy and/or vena caval thrombectomy were not included in study. Subsequent analysis and modeling
were performed on 864 patients after excluding those who underwent concomitant procedures or had incomplete data.
Variables Measured This study assessed more than 50 variables regarding patient demographics, medical comorbidities and preoperative laboratory variables collected within 90 days of surgery. The complete list includes gender, age, body mass index, ASAÒ score, various comorbidities and preoperative laboratory values (table 1).
Outcome Measures The primary outcome measure was absolute LoS among patients discharged from the hospital. The predictive models used the log-transformed value of LoS due to the large amount of skew in the nontransformed LoS variable.
External Validation Set Patient Population Results were validated in an external cohort of 427 patients at Emory University Hospital, Atlanta, Georgia, who underwent laparoscopic or open nephrectomy between November 2006 and January 2012. All participants provided written informed consent. Patients with nodal or metastatic disease were excluded from study as were those with concomitant procedures. Notably, these patients were not in the NSQIP cohort since urological patients are not generally abstracted at our institution.
Statistical Analysis All available variables were evaluated using univariate analysis (linear regression models) to assess the correlation with LoS. Variables found to be significantly associated with LoS using an a level of 0.05 on univariate analysis were included in an initial multivariable linear model. Backward selection was then used to sequentially exclude nonsignificant variables from the model when controlling for other variables. At each step the variable with the largest p value of the variables with p >0.05 was excluded. All 2-way interactions between variables in the final model were tested. However, because none had a substantial influence (more than 1%) on R2, none of them were kept in the model. This model is presented as model 1. A smaller model constructed using a subset of variables from model 1 is presented as model 2. The validation aspect of this study is an important element to demonstrate the applicability of NSQIP derived data to individual institutions. Elements in model 1 that were not sufficiently common or adequately documented in institutional records were eliminated to create model 2. The validity of the final models was checked using model diagnostics, including the distribution of residuals, the Cook distance and leverage values. The final linear regression model was then assessed in the external validation cohort. Using the parameter estimates from the final predictive model the predicted LoS was calculated for each patient in the validation cohort with data in all relevant fields. The correlation between this predicted value and the actual patient LoS was assessed using the Pearson correlation coefficient and R2 was calculated. All statistical analyses were performed with SASÒ, version 9.2.
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
925
Table 1. Patient comorbidities and covariates evaluated in NSQIP cohort and univariate effects on LoS Value Age/yr No. pts (%): Smoking Previous cerebrovascular accident Hemodialysis Infected wound Steroid use Functional dependence Sepsis or SIRS Hemiplegia CHF Bleeding disorder Chronic obstructive pulmonary disease Dyspnea ASA class 3, 4 or 5 Wound class 3 or 4 Prior operation (less than 30 days) Emergency Open vs laparoscopic approach Median (IQR)/No. missing (%): HCT (ml/dl) Albumin (gm/dl) Prothrombin time (secs) Partial thromboplastin time (secs) White blood count ( 1,000/ml) Alkaline phosphatase (IU/l) Serum glutamic-oxaloacetic transaminase (U/l) Sodium (mEq/l) International normalized ratio Estimated glomerular filtration rate (ml/min)
e 294 65 123 23 91 50 32 15 15 41 90 139 945 45 25 13 429 39.6 4.0 11.7 29.0 7.1 79.0 21.0 139.0 1.0 77.2
(6.6)/ 69 (0.7)/655 (2.8)/650 (5.1)/672 (2.8)/ 62 (33.0)/639 (10.0)/632 (4.0)/ 80 (0.1)/594 (41.7)/ 94
Estimate (SE)
p Value
0.0066 (0.0011)
<0.0001
(19.3) (4.3) (8.1) (1.5) (6.0) (3.3) (2.1) (1.0) (1.0) (2.7) (5.9) (9.1) (61.9)* (3.0) (1.6) (0.9) (28.1)
0.1155 0.3402 0.3250 0.3929 0.1792 0.8352 1.0167 0.6620 0.8125 0.3988 0.3053 0.3317 0.2872 0.4578 0.6438 0.8790 0.5128
(0.0438) (0.0859) (0.0630) (0.1448) (0.0729) (0.0947) (0.1179) (0.1806) (0.1741) (0.1078) (0.0730) (0.0597) (0.0349) (0.1016) (0.1352) (0.1945) (0.0363)
0.0085 <0.0001 <0.0001 0.0067 0.0141 <0.0001 <0.0001 0.0003 <0.0001 0.0002 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001
(4.5) (42.9) (42.6) (44.0) (4.1) (41.8) (41.4) (5.2) (38.9) (6.2)
0.036 0.3792 0.02125 0.0206 0.0305 0.0020 0.0020 0.0268 0.7330 0.0049
(0.0032) (0.0359) (0.0102) (0.0045) (0.0063) (0.0005) (0.0013) (0.0060) (0.1522) (0.0006)
<0.0001 <0.0001 0.0369 <0.0001 <0.0001 <0.0001 0.1262 <0.0001 <0.0001 <0.0001
* Data missing on 2 patients (0.1%).
RESULTS Patient Population There were 1,527 patients in the ACS NSQIP between 2007 and 2011 who underwent total nephrectomy, including 864 with complete data. The population was 54.2% male with a median SD age of 61 19 years (table 2). Of the patients 70.0% were white and a laparoscopic approach was used in 71.9%. Median albumin was 4.0 gm/dl (IQR 0.7). Median LoS was 3 days (IQR 3). Table 1 lists further details on examined variables. Because of the large sample sizes, all differences in demographics between the training and validation sets were statistically significant (p <0.05) but some differences were too small to be meaningful. The median LoS and IQR were identical in the 2 sets. However, the validation population included more males (61% vs 54%), a higher proportion of patients of black race (20% vs 11%) and fewer Hispanic patients (1% vs 10%). The validation set contained 2% greater ASA class 4 patients and 11% more ASA class 3 patients, noting a deficit of 77 patients in the validation cohort with missing data. The validation set also contained a similar percent of open surgeries (137 or 32%) as the training set. LoS Linear Regression Analysis Univariate linear regression analysis identified significant relationships between LoS and 27 preoperative
variables (table 1). After backward elimination 9 significant variables remained in the model. This model had a coefficient of determination (R2) value of 0.309. It created an equation to predict LoS, that is for model 1 LoS in days ¼ e1.98 þ 0.006 age e 0.020 HCT e 0.111 albumin þ 0.092 (if ASA class 3, 4 or 5) þ 0.352 (if dependent) þ 0.553 (if sepsis or
SIRS) þ 0.508 (if hemiplegic) þ 0.701 (if CHF) þ 0.468 (if open surgery)
. From this model 4 variables were selected to be included in reduced regression, including age,
Table 2. Demographics NSQIP Training Set No. pts No. male (%) No. female (%) No. race (%): White Black Other NonHispanic Hispanic No. ASA class (%):† 1 2 3 4 5 Median kg/m2 body mass index (IQR) Median age at surgery (IQR)
Emory Validation Set*
1,527 822 696
(54.2) (45.9)
427 260 167
(60.9) (39.1)
1,055 171 281 1,232 142
(70.0) (11.4) (18.7) (89.7) (10.3)
311 86 30 421 6
(72.8) (20.1) (7.0) (98.6) (1.4)
(2.5) (35.5) (53.2) (8.7) (0.1) (8.1)
5 83 225 37
38 542 811 133 1 28.9
61.0 (19.0)
(1.4) (23.7) (64.3) (10.6) e 27.8 (7.7) 61.0 (20.0)
* All variables significantly different vs training set (p <0.05). † In validation set 77 patients lacked documented ASA class on chart review.
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
926
HCT, albumin and laparoscopic vs open approach (table 3). The remaining variables were removed because of the lack of broad clinical applicability for this procedure, including functional dependence, sepsis/SIRS, hemiplegia and CHF. ASA class was not included due to missing data in the validation set, which prevented accurate assessment of the model. Model 2 had an R2 value of 0.256. The modeled equation that describes the relationship between these variables and LoS is shown as model 2, that is LoS in days ¼ e2.556 þ 0.007 age e 0.025 HCT e 0.195 albumin þ 0.476 (if open surgery) . From this equation one can conclude in more simple terms that for every year increase in age, LoS is prolonged by a factor of 0.7% or e0.007. For each point decrease in HCT LoS is prolonged by a factor of 2.5%. For every 1-point decrease in albumin LoS increases by a factor of 17.7%. If nephrectomy is performed in open fashion, one can expect LoS to increase by 61%. External Validation The multivariate linear regression model (model 2) was applied to the external validation cohort of 427 patients. Table 2 lists the demographics of this cohort. Actual LoS was compared to the LoS predicted by our model. The 4 variables demonstrated an R2 of 0.214 (p <0.001).
DISCUSSION LoS depends on many variables, of which some are immeasurable and impossible to predict. However, these results demonstrate that a few discrete preoperative factors can predict a considerable portion of variability in LoS. With refinements and expansions in data collection systems such as the NSQIP health care outcome prediction is becoming more feasible. This could lead to improved quality of care and accuracy of quality assessments. Table 3. Variables and coefficients of models 1 and 2 Estimate (95% CI)
Intercept Age HCT Albumin ASA 3, 4 or 5 Functional dependence Sepsis/SIRS Hemiplegia CHF Open surgery Set R2: Training Validation
Model 1
Model 2
1.980 (1.574e2.386) 0.006 (0.003e0.009) 0.020 (0.029e0.012) 0.111 (0.185e0.037) 0.092 (0.004e0.180) 0.352 (0.151e0.554) 0.553 (0.298e0.808) 0.508 (0.159e0.857) 0.701 (0.316e1.086) 0.468 (0.378e0.559)
2.556 (2.167e2.945) 0.007 (0.004e0.009) 0.025 (0.033e0.016) 0.195 (0.268e0.121) e e e e e 0.476 (0.383e0.569)
0.309
e
0.256 0.214
The CMMS (Centers for Medicare and Medicaid Services) provides a geometric mean LoS of 3.0 days for nephrectomy without comorbidities (DRG 658, fiscal year 2011). Using this as a reference case every year increase in age, every point decrease in HCT, every point decrease in albumin and performing the procedure in open fashion would increase LoS by 0.045, 0.075, 0.53 and 1.8 days, respectively. However, this model shows relative changes that are not constant across patients. For instance, these factors would cause a larger absolute increase in LoS in a patient with significant comorbidities. The R2 value of the final model is 0.256. Said another way, 25.6% of the variability that exists in LoS can be explained by patient age, preoperative HCT and albumin, and surgical approach. Advanced age and decreased hematocrit predict increased LoS, confirming established associations.8,15,17 An open over a laparoscopic approach increases predicted LoS by 61%, confirming prior findings.18,19 While this supports such an approach, it should be noted that laparoscopy is not always possible or indicated. Any quality assessment tool should account for surgical approaches since they represent relatively disparate entities. Albumin has been controversial in its use as an indicator of nutritional status. However, considerable evidence points to its clinical significance. Albumin is prognostic in several cancers, including cancer of the lung, pancreas, breast and colorectum.20 Existing literature shows an association between preoperative albumin and surgical complications, including measures such as LoS, intensive care unit stay, hospital readmission and mortality.21e23 Our data are consistent with these findings, suggesting that albumin may be an effective tool for surgical planning that lends substantial value to quality assessment tools as well as systems such as the DRG. The initial model with an R2 of 0.309 included several factors that were rare in the validation population. These factors were excluded to focus the model on variables that are widely applicable to individual practices and institutions. Nevertheless, after excluding more than half of the variables the model 2 R2 value decreased a mere 0.05, further illustrating the impact of the 4 included variables. Understanding the importance of these patient factors is clinically relevant for patient education and decision making. A physician may be better equipped to identify patients at risk for prolonged LoS and counsel them appropriately regarding the expected postoperative course. Discharge planning may then be coordinated to a greater degree of accuracy, facilitating more efficient patient flow and operations management.
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
Currently institutional reviews and quality assessments are built on relatively generalized classification systems such as the DRG. While some level of specificity exists, including the presence or absence of comorbidities, individualization is limited. Several European studies have previously demonstrated the insufficiency of DRG systems in accounting for variability in LoS.24,25 A predictive model incorporating individualized patient variables could serve as a better performance assessment tool. Single institution validation confirmed that ACS NSQIP data are an applicable and useful resource for outcome based studies in urology. The difference in R2 values between the NSQIP and validation data sets was 0.042, demonstrating the accuracy of model 2 in external populations. This robust database has been used extensively in other disciplines.6,10,11,26 It has also been used in prior studies of common urological procedures,15 although none of the studies validated their models using data from a single institution. This step is important to determine the usefulness of models derived from national data for use in local clinical care and institutional quality assessments. Continued incorporation of urological data into the NSQIP may allow for improved critical review of outcomes and quality of care. This study demonstrates the feasibility of predictive models in health care using the specific case of nephrectomy. Several prior groups have developed similar models to predict LoS.27e30 Preoperative factors such as age and comorbidities were often found to be predictive.28,30 However, existing literature on predictive models in urology is limited and should be further investigated. Grady et al created predictive models for LoS after heart transplantation and found that the R2 value increased from 0.36 to 0.71 after adding perioperative and postoperative factors to the
927
original model with only preoperative factors.29 This highlights one of the limitations of the current study. Our model highlights the importance of preoperative patient factors while excluding perioperative and postoperative variables, thus accounting for only 25.6% of variability in LoS after nephrectomy. Another limitation is the inability to sufficiently examine ASA class due to inadequate documentation in local records. The analysis was completed separately including ASA and validated using patients with available data. The R2 value of the validation was 0.179. We hypothesize that the discrepancy was likely due to reporting bias toward patients who were more ill in the local cohort. Thus, the variable was not included in the final model, although it appears to be meaningful and should be examined in future studies. Lastly, the NSQIP database does not include tumor size or location. These 2 variables may impact LoS, surgical approach and operative time. It would be prudent to investigate them in the future.
CONCLUSIONS LoS has become a critical metric in quality improvement in the changing health care landscape with clinical and financial implications to the patient and the public. Using the NSQIP database we created and validated a model that contains only 4 variables and yet accounts for 25.6% of variability in patient LoS following nephrectomy. As data collection becomes increasingly robust through more developed and comprehensive electronic records, this type of individualized preoperative analysis grows more feasible. Moving beyond generalized measures toward more tailored predictive models could improve patient care and the ability to assess quality and resource use.
REFERENCES 1. O’Keefe GE, Jurkovich GJ and Maier RV: Defining excess resource utilization and identifying associated factors for trauma victims. J Trauma 1999; 46: 473.
4. Clark DE and Ryan LM: Concurrent prediction of hospital mortality and length of stay from risk factors on admission. Health Serv Res 2002; 37: 631.
2. Guru V, Anderson GM, Fremes SE et al: The identification and development of Canadian coronary artery bypass graft surgery quality indicators. J Thorac Cardiovasc Surg 2005; 130: 1257.
5. Borja-Cacho D, Parsons HM, Habermann EB et al: Assessment of ACS NSQIP’s predictive ability for adverse events after major cancer surgery. Ann Surg Oncol 2010; 17: 2274.
3. Rotter T, Kinsman L, James E et al: Clinical pathways: effects on professional practice, patient outcomes, length of stay and hospital costs. Cochrane Database Syst Rev 2010; 3: CD006632.
6. Ingraham AM, Cohen ME, Bilimoria KY et al: Association of surgical care improvement project infection-related process measure compliance with risk-adjusted outcomes: implications for quality measurement. J Am Coll Surg 2010; 211: 705.
7. 2010 Health Care Cost and Utilization Report. Washington, D.C.: Health Care Cost Institute 2012. 8. Collins TC, Daley J, Henderson WH et al: Risk factors for prolonged length of stay after major elective surgery. Ann Surg 1999; 230: 251. 9. Kalish RL, Daley J, Duncan CC et al: Costs of potential complications of care for major surgery patients. Am J Med Qual 1995; 10: 48. 10. Davenport DL, Henderson WG, Khuri SF et al: Preoperative risk factors and surgical complexity are more predictive of costs than postoperative complications: a case study using the National
928
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
Surgical Quality Improvement Program (NSQIP) database. Ann Surg 2005; 242: 463. 11. Polverejan E, Gardiner JC, Bradley CJ et al: Estimating mean hospital cost as a function of length of stay and patient characteristics. Health Econ 2003; 12: 935. 12. Hollenbeck BK, Miller DC, Taub D et al: Risk factors for adverse outcomes after transurethral resection of bladder tumors. Cancer 2006; 106: 1527. 13. Aronow HD, Peyser PA, Eagle KA et al: Predictors of length of stay after coronary stenting. Am Heart J 2001; 142: 799. 14. Ghali WA, Hall RE, Ash AS et al: Identifying preand postoperative predictors of cost and length of stay for coronary artery bypass surgery. Am J Med Qual 1999; 14: 248. 15. Wallner LP, Dunn RL, Sarma AV et al: Risk factors for prolonged length of stay after urologic surgery: the National Surgical Quality Improvement Program. J Am Coll Surg 2008; 207: 904. 16. Khuri SF, Daley J, Henderson W et al: The Department of Veterans Affairs’ NSQIP: the first national, validated, outcome-based, riskadjusted, and peer-controlled program for the measurement and enhancement of the quality of surgical care. National VA Surgical Quality Improvement Program. Ann Surg 1998; 228: 491.
17. Leichtle SW, Mouawad NJ, Lampman R et al: Does preoperative anemia adversely affect colon and rectal surgery outcomes? J Am Coll Surg 2011; 212: 187. 18. Wadstrom J, Martin AL, Estok R et al: Comparison of hand-assisted laparoscopy versus open and laparoscopic techniques in urology procedures: a systematic review and meta-analysis. J Endourol 2011; 25: 1095.
24. Geissler A, Scheller-Kreinsen D and Quentin W: Do diagnosis-related groups appropriately explain variations in costs and length of stay of hip replacement? A comparative assessment of DRG systems across 10 European countries. Health Econ, suppl., 2012; 21: 103. 25. Scheller-Kreinsen D: How well do diagnosisrelated group systems group breast cancer surgery patients? Evidence from 10 European countries. Health Econ, suppl., 2012; 21: 41.
19. Yu HY, Hevelone ND, Lipsitz SR et al: Use, costs and comparative effectiveness of robotic assisted, laparoscopic and open urological surgery. J Urol 2012; 187: 1392.
26. Taheri PA, Butz DA and Greenfield LJ: Length of stay has minimal impact on the cost of hospital admission. J Am Coll Surg 2000; 191: 123.
20. Fanali G, di Masi A, Trezza V et al: Human serum albumin: from bench to bedside. Mol Aspects Med 2012; 33: 209.
27. Li CH, Bair MJ, Chang WH et al: Predictive model for length of hospital stay of patients surviving surgery for perforated peptic ulcer. J Formos Med Assoc 2009; 108: 644.
21. Kudsk KA, Tolley EA, DeWitt RC et al: Preoperative albumin and surgical site identify surgical risk for major postoperative complications. JPEN J Parenter Enteral Nutr 2003; 27: 1. 22. Nisar PJ, Appau KA, Remzi FH et al: Preoperative hypoalbuminemia is associated with adverse outcomes after ileoanal pouch surgery. Inflamm Bowel Dis 2012; 18: 1034. 23. Herrmann FR, Safran C, Levkoff SE et al: Serum albumin level on admission as a predictor of death, length of stay, and readmission. Arch Intern Med 1992; 152: 125.
28. Kelly M, Sharp L, Dwane F et al: Factors predicting hospital length-of-stay and readmission after colorectal resection: a population-based study of elective and emergency admissions. BMC Health Serv Res 2012; 12: 77. 29. Grady KL, Haller KB, Grusk BB et al: Predictors of hospital length of stay after heart transplantation. J Heart Transplant 1990; 9: 92. 30. Ansari MZ, MacIntyre CR, Ackland MJ et al: Predictors of length of stay for transurethral prostatectomy in Victoria. Aust N Z J Surg 1998; 68: 837.
Purpose: Length of stay is frequently used to measure the quality of health care, although its predictors are not well studied in urology. We created a predictive model of length of stay after nephrectomy, focusing on preoperative variables. Materials and Methods: We used the NSQIP database to evaluate patients older than 18 years who underwent nephrectomy without concomitant procedures from 2007 to 2011. Preoperative factors analyzed for univariate significance in relation to actual length of stay were then included in a multivariable linear regression model. Backward elimination of nonsignificant variables resulted in a final model that was validated in an institutional external patient cohort. Results: Of the 1,527 patients in the NSQIP database 864 were included in the training cohort after exclusions for concomitant procedures or lack of data. Median length of stay was 3 days in the training and validation sets. Univariate analysis revealed 27 significant variables. Backward selection left a final model including the variables age, laparoscopic vs open approach, and preoperative hematocrit and albumin. For every additional year in age, point decrease in hematocrit and point decrease in albumin the length of stay lengthened by a factor of 0.7%, 2.5% and 17.7%, respectively. If an open approach was performed, length of stay increased by 61%. The R2 value was 0.256. The model was validated in a 427 patient external cohort, which yielded an R2 value of 0.214. Conclusions: Age, preoperative hematocrit, preoperative albumin and approach have significant effects on length of stay for patients undergoing nephrectomy. Similar predictive models could prove useful in patient education as well as quality assessment. Key Words: kidney, nephrectomy, length of stay, hypoalbuminemia, risk
IN efforts to curb the rising cost of health care in the United States payers have developed systems such as managed care and accountable care organizations. This highlights the longtime focus on quality as a measure to drive resource conservation. LoS is a common indicator in quality assessment and improvement constructs.1e4 LoS can also contribute to decisions on pre-authorization and reimbursement
Abbreviations and Acronyms ACS ¼ American College of Surgeons ASA ¼ American Society of AnesthesiologistsÒ CHF ¼ congestive heart failure DRG ¼ Diagnosis-Related Group HCT ¼ hematocrit LoS ¼ length of stay NSQIP ¼ National Surgical Quality Improvement Program SIRS ¼ systemic inflammatory response syndrome Accepted for publication March 27, 2015. Study received institutional review board approval. ACS NSQIP and participating hospitals have not verified and are not responsible for the statistical validity of the data analysis or the conclusions derived by the authors. * Correspondence: Department of Urology, Emory University School of Medicine, 1365 Clifton Rd., B1400, Atlanta, Georgia 30322 (FAX: 404-778-4006; e-mail: [email protected]).
as it is one of the primary output elements of the various DRG classification systems and used widely by payers to stratify diagnoses and procedures for reimbursement. Prolonged LoS has been associated with increased resource consumption and complications.5,6 The cost of an inpatient stay from 2009 to 2010 grew by 5.1%, nearly twice the rate of inflation.7 The cost of an inpatient
0022-5347/15/1944-0923/0 THE JOURNAL OF UROLOGY® Ó 2015 by AMERICAN UROLOGICAL ASSOCIATION EDUCATION AND RESEARCH, INC.
http://dx.doi.org/10.1016/j.juro.2015.04.112 Vol. 194, 923-928, October 2015 Printed in U.S.A.
www.jurology.com
j
923
924
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
surgical admission grew by 6.4% during the same period.7 Inpatient services account for 21% of national health care expenditures.7 As spending continues to increase, LoS bears greater and greater importance. Furthermore, increased LoS places patients at greater risk for complications, leading to worse outcomes and increased system costs.8,9 Preoperative risk factors have an important role in predicting LoS and have been shown to be meaningful for overall hospital costs.6,10 Elucidating the most significant factors presents an opportunity to mitigate modifiable variables, and improve clinical decision making and quality assessment. Previous studies have explored variables associated with prolonged LoS.11e15 Such factors in common urological cases such as nephrectomy, prostatectomy and cystectomy include advanced patient age, dependent functional status, low serum albumin, low HCT and high creatinine.12,15 Some of these prior studies used the NSQIP database to study LoS determinants.11,15 The NSQIP is a robust multi-institutional database including hundreds of variables and a variety of procedures that was designed and validated for use in quality improvement and outcomes data analysis.16 The primary objective of this study was to identify preoperative risk factors that significantly contribute to prolonged LoS after radical nephrectomy. Our aim was to examine the predictive ability of preoperative factors, assess use of the NSQIP database for urological procedures and create a predictive model to elucidate the intrinsic variability in the LoS measure. Such models have become important in the shift toward a pay for performance reimbursement structure in which physicians will be expected to meet benchmark quality measurements.
METHODS American College of Surgeons NSQIP The ACS NSQIP and participating hospitals are the source of the data used in this study. The NSQIP is a prospective, multi-institutional database with cases representing a sample of operations at each institution.5 For each patient more than 150 variables are captured, including comorbidities, laboratory values and LoS. Data are audited annually for accuracy and completeness.
Training Set Patient Population The primary cohort consisted of 1,527 patients identified from the NSQIP PUF (Participant Use Data File) who were older than 18 years and underwent radical nephrectomy between 2007 and 2011 via a laparoscopic approach (CPT code 50545/50546) or an open approach (CPT code 50220/50225). Patients treated with regional lymphadenectomy and/or vena caval thrombectomy were not included in study. Subsequent analysis and modeling
were performed on 864 patients after excluding those who underwent concomitant procedures or had incomplete data.
Variables Measured This study assessed more than 50 variables regarding patient demographics, medical comorbidities and preoperative laboratory variables collected within 90 days of surgery. The complete list includes gender, age, body mass index, ASAÒ score, various comorbidities and preoperative laboratory values (table 1).
Outcome Measures The primary outcome measure was absolute LoS among patients discharged from the hospital. The predictive models used the log-transformed value of LoS due to the large amount of skew in the nontransformed LoS variable.
External Validation Set Patient Population Results were validated in an external cohort of 427 patients at Emory University Hospital, Atlanta, Georgia, who underwent laparoscopic or open nephrectomy between November 2006 and January 2012. All participants provided written informed consent. Patients with nodal or metastatic disease were excluded from study as were those with concomitant procedures. Notably, these patients were not in the NSQIP cohort since urological patients are not generally abstracted at our institution.
Statistical Analysis All available variables were evaluated using univariate analysis (linear regression models) to assess the correlation with LoS. Variables found to be significantly associated with LoS using an a level of 0.05 on univariate analysis were included in an initial multivariable linear model. Backward selection was then used to sequentially exclude nonsignificant variables from the model when controlling for other variables. At each step the variable with the largest p value of the variables with p >0.05 was excluded. All 2-way interactions between variables in the final model were tested. However, because none had a substantial influence (more than 1%) on R2, none of them were kept in the model. This model is presented as model 1. A smaller model constructed using a subset of variables from model 1 is presented as model 2. The validation aspect of this study is an important element to demonstrate the applicability of NSQIP derived data to individual institutions. Elements in model 1 that were not sufficiently common or adequately documented in institutional records were eliminated to create model 2. The validity of the final models was checked using model diagnostics, including the distribution of residuals, the Cook distance and leverage values. The final linear regression model was then assessed in the external validation cohort. Using the parameter estimates from the final predictive model the predicted LoS was calculated for each patient in the validation cohort with data in all relevant fields. The correlation between this predicted value and the actual patient LoS was assessed using the Pearson correlation coefficient and R2 was calculated. All statistical analyses were performed with SASÒ, version 9.2.
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
925
Table 1. Patient comorbidities and covariates evaluated in NSQIP cohort and univariate effects on LoS Value Age/yr No. pts (%): Smoking Previous cerebrovascular accident Hemodialysis Infected wound Steroid use Functional dependence Sepsis or SIRS Hemiplegia CHF Bleeding disorder Chronic obstructive pulmonary disease Dyspnea ASA class 3, 4 or 5 Wound class 3 or 4 Prior operation (less than 30 days) Emergency Open vs laparoscopic approach Median (IQR)/No. missing (%): HCT (ml/dl) Albumin (gm/dl) Prothrombin time (secs) Partial thromboplastin time (secs) White blood count ( 1,000/ml) Alkaline phosphatase (IU/l) Serum glutamic-oxaloacetic transaminase (U/l) Sodium (mEq/l) International normalized ratio Estimated glomerular filtration rate (ml/min)
e 294 65 123 23 91 50 32 15 15 41 90 139 945 45 25 13 429 39.6 4.0 11.7 29.0 7.1 79.0 21.0 139.0 1.0 77.2
(6.6)/ 69 (0.7)/655 (2.8)/650 (5.1)/672 (2.8)/ 62 (33.0)/639 (10.0)/632 (4.0)/ 80 (0.1)/594 (41.7)/ 94
Estimate (SE)
p Value
0.0066 (0.0011)
<0.0001
(19.3) (4.3) (8.1) (1.5) (6.0) (3.3) (2.1) (1.0) (1.0) (2.7) (5.9) (9.1) (61.9)* (3.0) (1.6) (0.9) (28.1)
0.1155 0.3402 0.3250 0.3929 0.1792 0.8352 1.0167 0.6620 0.8125 0.3988 0.3053 0.3317 0.2872 0.4578 0.6438 0.8790 0.5128
(0.0438) (0.0859) (0.0630) (0.1448) (0.0729) (0.0947) (0.1179) (0.1806) (0.1741) (0.1078) (0.0730) (0.0597) (0.0349) (0.1016) (0.1352) (0.1945) (0.0363)
0.0085 <0.0001 <0.0001 0.0067 0.0141 <0.0001 <0.0001 0.0003 <0.0001 0.0002 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001
(4.5) (42.9) (42.6) (44.0) (4.1) (41.8) (41.4) (5.2) (38.9) (6.2)
0.036 0.3792 0.02125 0.0206 0.0305 0.0020 0.0020 0.0268 0.7330 0.0049
(0.0032) (0.0359) (0.0102) (0.0045) (0.0063) (0.0005) (0.0013) (0.0060) (0.1522) (0.0006)
<0.0001 <0.0001 0.0369 <0.0001 <0.0001 <0.0001 0.1262 <0.0001 <0.0001 <0.0001
* Data missing on 2 patients (0.1%).
RESULTS Patient Population There were 1,527 patients in the ACS NSQIP between 2007 and 2011 who underwent total nephrectomy, including 864 with complete data. The population was 54.2% male with a median SD age of 61 19 years (table 2). Of the patients 70.0% were white and a laparoscopic approach was used in 71.9%. Median albumin was 4.0 gm/dl (IQR 0.7). Median LoS was 3 days (IQR 3). Table 1 lists further details on examined variables. Because of the large sample sizes, all differences in demographics between the training and validation sets were statistically significant (p <0.05) but some differences were too small to be meaningful. The median LoS and IQR were identical in the 2 sets. However, the validation population included more males (61% vs 54%), a higher proportion of patients of black race (20% vs 11%) and fewer Hispanic patients (1% vs 10%). The validation set contained 2% greater ASA class 4 patients and 11% more ASA class 3 patients, noting a deficit of 77 patients in the validation cohort with missing data. The validation set also contained a similar percent of open surgeries (137 or 32%) as the training set. LoS Linear Regression Analysis Univariate linear regression analysis identified significant relationships between LoS and 27 preoperative
variables (table 1). After backward elimination 9 significant variables remained in the model. This model had a coefficient of determination (R2) value of 0.309. It created an equation to predict LoS, that is for model 1 LoS in days ¼ e1.98 þ 0.006 age e 0.020 HCT e 0.111 albumin þ 0.092 (if ASA class 3, 4 or 5) þ 0.352 (if dependent) þ 0.553 (if sepsis or
SIRS) þ 0.508 (if hemiplegic) þ 0.701 (if CHF) þ 0.468 (if open surgery)
. From this model 4 variables were selected to be included in reduced regression, including age,
Table 2. Demographics NSQIP Training Set No. pts No. male (%) No. female (%) No. race (%): White Black Other NonHispanic Hispanic No. ASA class (%):† 1 2 3 4 5 Median kg/m2 body mass index (IQR) Median age at surgery (IQR)
Emory Validation Set*
1,527 822 696
(54.2) (45.9)
427 260 167
(60.9) (39.1)
1,055 171 281 1,232 142
(70.0) (11.4) (18.7) (89.7) (10.3)
311 86 30 421 6
(72.8) (20.1) (7.0) (98.6) (1.4)
(2.5) (35.5) (53.2) (8.7) (0.1) (8.1)
5 83 225 37
38 542 811 133 1 28.9
61.0 (19.0)
(1.4) (23.7) (64.3) (10.6) e 27.8 (7.7) 61.0 (20.0)
* All variables significantly different vs training set (p <0.05). † In validation set 77 patients lacked documented ASA class on chart review.
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
926
HCT, albumin and laparoscopic vs open approach (table 3). The remaining variables were removed because of the lack of broad clinical applicability for this procedure, including functional dependence, sepsis/SIRS, hemiplegia and CHF. ASA class was not included due to missing data in the validation set, which prevented accurate assessment of the model. Model 2 had an R2 value of 0.256. The modeled equation that describes the relationship between these variables and LoS is shown as model 2, that is LoS in days ¼ e2.556 þ 0.007 age e 0.025 HCT e 0.195 albumin þ 0.476 (if open surgery) . From this equation one can conclude in more simple terms that for every year increase in age, LoS is prolonged by a factor of 0.7% or e0.007. For each point decrease in HCT LoS is prolonged by a factor of 2.5%. For every 1-point decrease in albumin LoS increases by a factor of 17.7%. If nephrectomy is performed in open fashion, one can expect LoS to increase by 61%. External Validation The multivariate linear regression model (model 2) was applied to the external validation cohort of 427 patients. Table 2 lists the demographics of this cohort. Actual LoS was compared to the LoS predicted by our model. The 4 variables demonstrated an R2 of 0.214 (p <0.001).
DISCUSSION LoS depends on many variables, of which some are immeasurable and impossible to predict. However, these results demonstrate that a few discrete preoperative factors can predict a considerable portion of variability in LoS. With refinements and expansions in data collection systems such as the NSQIP health care outcome prediction is becoming more feasible. This could lead to improved quality of care and accuracy of quality assessments. Table 3. Variables and coefficients of models 1 and 2 Estimate (95% CI)
Intercept Age HCT Albumin ASA 3, 4 or 5 Functional dependence Sepsis/SIRS Hemiplegia CHF Open surgery Set R2: Training Validation
Model 1
Model 2
1.980 (1.574e2.386) 0.006 (0.003e0.009) 0.020 (0.029e0.012) 0.111 (0.185e0.037) 0.092 (0.004e0.180) 0.352 (0.151e0.554) 0.553 (0.298e0.808) 0.508 (0.159e0.857) 0.701 (0.316e1.086) 0.468 (0.378e0.559)
2.556 (2.167e2.945) 0.007 (0.004e0.009) 0.025 (0.033e0.016) 0.195 (0.268e0.121) e e e e e 0.476 (0.383e0.569)
0.309
e
0.256 0.214
The CMMS (Centers for Medicare and Medicaid Services) provides a geometric mean LoS of 3.0 days for nephrectomy without comorbidities (DRG 658, fiscal year 2011). Using this as a reference case every year increase in age, every point decrease in HCT, every point decrease in albumin and performing the procedure in open fashion would increase LoS by 0.045, 0.075, 0.53 and 1.8 days, respectively. However, this model shows relative changes that are not constant across patients. For instance, these factors would cause a larger absolute increase in LoS in a patient with significant comorbidities. The R2 value of the final model is 0.256. Said another way, 25.6% of the variability that exists in LoS can be explained by patient age, preoperative HCT and albumin, and surgical approach. Advanced age and decreased hematocrit predict increased LoS, confirming established associations.8,15,17 An open over a laparoscopic approach increases predicted LoS by 61%, confirming prior findings.18,19 While this supports such an approach, it should be noted that laparoscopy is not always possible or indicated. Any quality assessment tool should account for surgical approaches since they represent relatively disparate entities. Albumin has been controversial in its use as an indicator of nutritional status. However, considerable evidence points to its clinical significance. Albumin is prognostic in several cancers, including cancer of the lung, pancreas, breast and colorectum.20 Existing literature shows an association between preoperative albumin and surgical complications, including measures such as LoS, intensive care unit stay, hospital readmission and mortality.21e23 Our data are consistent with these findings, suggesting that albumin may be an effective tool for surgical planning that lends substantial value to quality assessment tools as well as systems such as the DRG. The initial model with an R2 of 0.309 included several factors that were rare in the validation population. These factors were excluded to focus the model on variables that are widely applicable to individual practices and institutions. Nevertheless, after excluding more than half of the variables the model 2 R2 value decreased a mere 0.05, further illustrating the impact of the 4 included variables. Understanding the importance of these patient factors is clinically relevant for patient education and decision making. A physician may be better equipped to identify patients at risk for prolonged LoS and counsel them appropriately regarding the expected postoperative course. Discharge planning may then be coordinated to a greater degree of accuracy, facilitating more efficient patient flow and operations management.
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
Currently institutional reviews and quality assessments are built on relatively generalized classification systems such as the DRG. While some level of specificity exists, including the presence or absence of comorbidities, individualization is limited. Several European studies have previously demonstrated the insufficiency of DRG systems in accounting for variability in LoS.24,25 A predictive model incorporating individualized patient variables could serve as a better performance assessment tool. Single institution validation confirmed that ACS NSQIP data are an applicable and useful resource for outcome based studies in urology. The difference in R2 values between the NSQIP and validation data sets was 0.042, demonstrating the accuracy of model 2 in external populations. This robust database has been used extensively in other disciplines.6,10,11,26 It has also been used in prior studies of common urological procedures,15 although none of the studies validated their models using data from a single institution. This step is important to determine the usefulness of models derived from national data for use in local clinical care and institutional quality assessments. Continued incorporation of urological data into the NSQIP may allow for improved critical review of outcomes and quality of care. This study demonstrates the feasibility of predictive models in health care using the specific case of nephrectomy. Several prior groups have developed similar models to predict LoS.27e30 Preoperative factors such as age and comorbidities were often found to be predictive.28,30 However, existing literature on predictive models in urology is limited and should be further investigated. Grady et al created predictive models for LoS after heart transplantation and found that the R2 value increased from 0.36 to 0.71 after adding perioperative and postoperative factors to the
927
original model with only preoperative factors.29 This highlights one of the limitations of the current study. Our model highlights the importance of preoperative patient factors while excluding perioperative and postoperative variables, thus accounting for only 25.6% of variability in LoS after nephrectomy. Another limitation is the inability to sufficiently examine ASA class due to inadequate documentation in local records. The analysis was completed separately including ASA and validated using patients with available data. The R2 value of the validation was 0.179. We hypothesize that the discrepancy was likely due to reporting bias toward patients who were more ill in the local cohort. Thus, the variable was not included in the final model, although it appears to be meaningful and should be examined in future studies. Lastly, the NSQIP database does not include tumor size or location. These 2 variables may impact LoS, surgical approach and operative time. It would be prudent to investigate them in the future.
CONCLUSIONS LoS has become a critical metric in quality improvement in the changing health care landscape with clinical and financial implications to the patient and the public. Using the NSQIP database we created and validated a model that contains only 4 variables and yet accounts for 25.6% of variability in patient LoS following nephrectomy. As data collection becomes increasingly robust through more developed and comprehensive electronic records, this type of individualized preoperative analysis grows more feasible. Moving beyond generalized measures toward more tailored predictive models could improve patient care and the ability to assess quality and resource use.
REFERENCES 1. O’Keefe GE, Jurkovich GJ and Maier RV: Defining excess resource utilization and identifying associated factors for trauma victims. J Trauma 1999; 46: 473.
4. Clark DE and Ryan LM: Concurrent prediction of hospital mortality and length of stay from risk factors on admission. Health Serv Res 2002; 37: 631.
2. Guru V, Anderson GM, Fremes SE et al: The identification and development of Canadian coronary artery bypass graft surgery quality indicators. J Thorac Cardiovasc Surg 2005; 130: 1257.
5. Borja-Cacho D, Parsons HM, Habermann EB et al: Assessment of ACS NSQIP’s predictive ability for adverse events after major cancer surgery. Ann Surg Oncol 2010; 17: 2274.
3. Rotter T, Kinsman L, James E et al: Clinical pathways: effects on professional practice, patient outcomes, length of stay and hospital costs. Cochrane Database Syst Rev 2010; 3: CD006632.
6. Ingraham AM, Cohen ME, Bilimoria KY et al: Association of surgical care improvement project infection-related process measure compliance with risk-adjusted outcomes: implications for quality measurement. J Am Coll Surg 2010; 211: 705.
7. 2010 Health Care Cost and Utilization Report. Washington, D.C.: Health Care Cost Institute 2012. 8. Collins TC, Daley J, Henderson WH et al: Risk factors for prolonged length of stay after major elective surgery. Ann Surg 1999; 230: 251. 9. Kalish RL, Daley J, Duncan CC et al: Costs of potential complications of care for major surgery patients. Am J Med Qual 1995; 10: 48. 10. Davenport DL, Henderson WG, Khuri SF et al: Preoperative risk factors and surgical complexity are more predictive of costs than postoperative complications: a case study using the National
928
PREDICTING LENGTH OF STAY FOLLOWING NEPHRECTOMY
Surgical Quality Improvement Program (NSQIP) database. Ann Surg 2005; 242: 463. 11. Polverejan E, Gardiner JC, Bradley CJ et al: Estimating mean hospital cost as a function of length of stay and patient characteristics. Health Econ 2003; 12: 935. 12. Hollenbeck BK, Miller DC, Taub D et al: Risk factors for adverse outcomes after transurethral resection of bladder tumors. Cancer 2006; 106: 1527. 13. Aronow HD, Peyser PA, Eagle KA et al: Predictors of length of stay after coronary stenting. Am Heart J 2001; 142: 799. 14. Ghali WA, Hall RE, Ash AS et al: Identifying preand postoperative predictors of cost and length of stay for coronary artery bypass surgery. Am J Med Qual 1999; 14: 248. 15. Wallner LP, Dunn RL, Sarma AV et al: Risk factors for prolonged length of stay after urologic surgery: the National Surgical Quality Improvement Program. J Am Coll Surg 2008; 207: 904. 16. Khuri SF, Daley J, Henderson W et al: The Department of Veterans Affairs’ NSQIP: the first national, validated, outcome-based, riskadjusted, and peer-controlled program for the measurement and enhancement of the quality of surgical care. National VA Surgical Quality Improvement Program. Ann Surg 1998; 228: 491.
17. Leichtle SW, Mouawad NJ, Lampman R et al: Does preoperative anemia adversely affect colon and rectal surgery outcomes? J Am Coll Surg 2011; 212: 187. 18. Wadstrom J, Martin AL, Estok R et al: Comparison of hand-assisted laparoscopy versus open and laparoscopic techniques in urology procedures: a systematic review and meta-analysis. J Endourol 2011; 25: 1095.
24. Geissler A, Scheller-Kreinsen D and Quentin W: Do diagnosis-related groups appropriately explain variations in costs and length of stay of hip replacement? A comparative assessment of DRG systems across 10 European countries. Health Econ, suppl., 2012; 21: 103. 25. Scheller-Kreinsen D: How well do diagnosisrelated group systems group breast cancer surgery patients? Evidence from 10 European countries. Health Econ, suppl., 2012; 21: 41.
19. Yu HY, Hevelone ND, Lipsitz SR et al: Use, costs and comparative effectiveness of robotic assisted, laparoscopic and open urological surgery. J Urol 2012; 187: 1392.
26. Taheri PA, Butz DA and Greenfield LJ: Length of stay has minimal impact on the cost of hospital admission. J Am Coll Surg 2000; 191: 123.
20. Fanali G, di Masi A, Trezza V et al: Human serum albumin: from bench to bedside. Mol Aspects Med 2012; 33: 209.
27. Li CH, Bair MJ, Chang WH et al: Predictive model for length of hospital stay of patients surviving surgery for perforated peptic ulcer. J Formos Med Assoc 2009; 108: 644.
21. Kudsk KA, Tolley EA, DeWitt RC et al: Preoperative albumin and surgical site identify surgical risk for major postoperative complications. JPEN J Parenter Enteral Nutr 2003; 27: 1. 22. Nisar PJ, Appau KA, Remzi FH et al: Preoperative hypoalbuminemia is associated with adverse outcomes after ileoanal pouch surgery. Inflamm Bowel Dis 2012; 18: 1034. 23. Herrmann FR, Safran C, Levkoff SE et al: Serum albumin level on admission as a predictor of death, length of stay, and readmission. Arch Intern Med 1992; 152: 125.
28. Kelly M, Sharp L, Dwane F et al: Factors predicting hospital length-of-stay and readmission after colorectal resection: a population-based study of elective and emergency admissions. BMC Health Serv Res 2012; 12: 77. 29. Grady KL, Haller KB, Grusk BB et al: Predictors of hospital length of stay after heart transplantation. J Heart Transplant 1990; 9: 92. 30. Ansari MZ, MacIntyre CR, Ackland MJ et al: Predictors of length of stay for transurethral prostatectomy in Victoria. Aust N Z J Surg 1998; 68: 837.