A Scoring System to Predict Readmission of Patients With Acute Pancreatitis to the Hospital Within Thirty Days of Discharge

Transcription

1 CLINICAL GASTROENTEROLOGY AND HEPATOLOGY 2011;9: A Scoring System to Predict Readmission of Patients With Acute Pancreatitis to the Hospital Within Thirty Days of Discharge TOM L. WHITLOCK, APRIL TIGNOR, EMILY M. WEBSTER, KATHRYN REPAS, DARWIN CONWELL, PETER A. BANKS, and BECHIEN U. WU Division of Gastroenterology, Brigham and Women s Hospital, Boston, Massachusetts This article has an accompanying continuing medical education activity on page e18. Learning Objectives At the end of this activity, the learner should identify the major factors that predict early readmission in acute pancreatitis. BACKGROUND & AIMS: Reducing rapid readmission of patients after discharge could improve quality of treatment and reduce costs. Little is known about clinical predictors of early readmission for acute pancreatitis (AP). We developed a strategy to identify and stratify patients with AP at risk for readmission within 30 days of discharge. METHODS: We derived and validated a model in a cohort of patients hospitalized with AP from June 2005 October Early readmission was defined as admission to the hospital or reevaluation in the emergency department within 30 days of discharge. The cohort was divided into a derivation cohort (admitted June 2005 December 2007, n 248) and a validation cohort (admitted January 2008 October 2009, n 198). A weighted scoring system was developed using logistic regression for the prediction of early readmission. Accuracy was assessed by area under the receiveroperator characteristic (ROC) curve analysis. RESULTS: Of the total patients, 21% (92/446) had early readmission. Multivariable analysis identified the following discharge characteristics as independent risk factors for early readmission: gastrointestinal symptoms, eating less than a solid diet, pancreatic necrosis, treatment with antibiotics, and pain (P.05). Weighted risk scores stratified patients into groups of low, moderate, and high risk for early readmission: 4%, 15%, and 87%, respectively, in the derivation cohort and 5%, 18%, and 68%, respectively, in the validation cohort. Area under the ROC curve demonstrated an accurate prediction (c-statistic 0.83). CONCLUSIONS: We created a scoring system that accurately predicts which patients with AP have high and low risk of readmission within 30 days of discharge. Keywords: Prediction Rule; Quality Control; Retrospective Cohort; Unplanned Readmission. Early, unplanned readmissions occur frequently in a wide variety of acute and chronic conditions, with rates ranging from 8% 30%. 1 9 Early readmissions contribute significantly to health care expenses and are associated with significant morbidity and mortality. 1 Identifying conditions with high rates of early readmission and reducing the need for early readmission are goals for health care quality improvement and cost reduction. We recently reported a 19% incidence of early readmission among patients hospitalized for acute pancreatitis (AP). 10 This is comparable to many other disease entities with high rates of early readmission. Because AP is a frequent and expensive health issue in the United States, high rates of early readmission are likely contributing to excess health care costs. 11,12 Identifying patients at high risk of early readmission is the first step in improving their care and reducing unnecessary expenditures. High rates of early readmission in AP and other conditions suggest physicians ability to stratify patients risk of readmission might be inadequate. This has been borne out in literature regarding other disease entities. 13,14 Clinical prediction rules have proved helpful in many diseases and have led to interventions targeting risk factors to reduce early readmissions. 15,16 We recently identified risk factors for early readmission after an episode of AP. 10 The strongest risk factors included discharge on less than a solid diet and gastrointestinal symptoms of nausea, vomiting, or diarrhea at discharge. We suspect these risk factors might prove useful in predicting early readmission and ultimately changing clinical practice. Early readmission is also an important measure of quality control and clinical success in AP in addition to morbidity and mortality, which have been the focus of most outcomes literature to date. 17 The primary purpose of this study was to develop and validate a strategy for identifying and stratifying patients at risk for early readmission in AP. Methods Study Design, Setting, and Population This was a retrospective cohort study of adult patients (age 18 years) with AP who were admitted or transferred to the Brigham and Women s Hospital in Boston, Massachusetts, between June 1, 2005, and October 31, Patients were eligible for inclusion if they had a diagnosis of AP defined by 2 or more of the following: (1) characteristic abdominal pain, (2) serum amylase or lipase level 3 times the upper limit of normal, or (3) a contrast-enhanced computed tomography (CECT) scan of the abdomen within the first 7 days of hospi- Abbreviations used in this paper: AP, acute pancreatitis; AUC, area under the curve; BMI, body mass index; CECT, contrast-enhanced computed tomography; ROC, receiver-operator characteristic by the AGA Institute /$36.00 doi: /j.cgh

2 176 WHITLOCK ET AL CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 9, No. 2 Table 1. Definitions of Potential Risk Factor Variables Variable Discharge variables Fever Tachycardia Tachypnea Hypertension Hypotension Hypoxia Leukocytosis Bandemia Gastrointestinal symptoms Pain Tolerating solid diet Antibiotics Opiates Abdominal drains Invasive devices Home disposition Demographic variables Older age Race Insured Active smoker Moderate alcohol Clinical variables Pancreatic necrosis Gallstone pancreatitis Alcohol-induced pancreatitis Transfer Definition Documented temperature 102 F within 24 h before discharge Documented heart rate 100 beats/min within 24 h before discharge Documented respiratory rate 20 breaths/min within 24 h before discharge Documented systolic blood pressure 140 mm Hg or diastolic blood pressure 90 mm Hg within 24 h before discharge Documented systolic blood pressure 100 mm Hg or diastolic blood pressure 60 mm Hg within 24 h before discharge Documented oxygen saturation 95% or need for supplemental oxygen to maintain oxygen saturation 95% White blood cell count 10,000/mm 3 before discharge Band count 3% before discharge Documented nausea, vomiting, or diarrhea within 24 h before discharge Documented pain at time of discharge Patient discharged on solid diet as opposed to full liquid, clears, sips, or nothing by mouth Antibiotics on discharge medication list Opiates on discharge medication list Patient discharged with abdominal drains in place Patient discharged with invasive devices or catheters in place (ie, gastrotomy tubes, intravenous catheters, Foley catheters, etc) Patient sent home rather than a nursing or rehabilitation facility Age 65 y at time of hospitalization White race was compared with minorities Patient with health insurance at time of hospitalization Patient s self-reported smoking status Patient s self-reported alcohol consumption of 1 drink/day for women or 2 drinks/day for men Presence of pancreatic necrosis on CECT or magnetic resonance imaging during hospitalization Index hospitalizations due to gallstone pancreatitis were compared with other etiologies Hospitalizations due to alcohol-induced pancreatitis were compared with other etiologies Patient transferred from an outside hospital vs directly admitted to our facility talization demonstrating changes consistent with AP. Only a patient s first admission for AP during the study time frame was included for analysis. It is subsequently referred to as the index admission. Patients were excluded for (1) radiographic or endoscopic evidence of chronic pancreatitis such as characteristic ductal abnormalities or calcifications, (2) lack of documented follow-up beyond 30 days of discharge, (3) death during their index hospitalization, (4) a scheduled or unrelated readmission within 30 days of discharge from an index hospitalization, or (5) discharge to hospice or palliative care after an index hospitalization. The study was approved by the Brigham and Women s Institutional Review Board. Data Collection Demographic, clinical, laboratory, and radiographic data were collected prospectively on consecutive patients enrolled in previous studies. Details of these collection methods have been described before. 18,19 Discharge data including vital signs, symptoms, medications, diet, and presence of invasive devices such as drains and feeding tubes were subsequently obtained from medical chart review. Individuals performing chart review were blinded to readmission status, which was extracted separately. Derivation and Validation Subcohorts Data from June 1, 2005 December 31, 2007, were used to develop a prediction algorithm; these patients represented the derivation cohort (the group used for deriving the model). All multivariate analysis and risk score development were performed by using the derivation cohort. Data from January 1, 2008 October 31, 2009, comprised the validation cohort. The validation cohort was only used after the risk score was created for testing purposes. Risk Factor Variables and Outcome To ensure ease of clinical application of the final model, risk factors were chosen on the basis of availability at discharge. Discharge characteristics included vital signs; laboratory data; gastrointestinal symptoms of nausea, vomiting, or diarrhea; pain; presence of abdominal drains or other invasive devices; discharge medications such as opiates and antibiotics; and diet. These variables were systematically extracted from standardized discharge forms such as nursing flow sheets or discharge summaries. Data with multiple measurements such as laboratory values and vital statistics were collected from the last measurement before discharge. Symptoms and pain were elicited every 4 hours on floor patients by clinical nursing staff and documented on standardized nursing flow sheets. These variables were considered present if they were documented on the same day as discharge. Standard diet orders were documented on discharge summaries and included solid (regular, low fat, and soft), full liquid, clear liquid, sips, and nothing by mouth diet orders. Discharge medications were extracted from discharge summary medication lists. Antibiotics included oral and intra-

3 February 2011 PREDICTING EARLY READMISSION IN AP 177 Table 2. Characteristics of Patients in the Derivation and Validation Cohorts Characteristics Derivation cohort (n 248) Validation cohort (n 198) Excluded group (n 147) Median age, y 51.6 (39.7, 63.3) 52.3 (41.9, 65.9) 52.2 (41.7, 62.5) Sex Female 135 (54.4) 106 (53.5) 65 (44.2) Race White 160 (65.0) 130 (67.0) 97 (66.0) Black 54 (22.0) 38 (19.6) 32 (21.0) Other 32 (13.0) 26 (13.4) 18 (12.2) Median BMI, kg/m (24.2, 31.8) 27.3 (22.9, 32.3) 27.0 (23.1, 33.3) Active smoker 58 (23.4) 47 (24.1) 39 (26.5) Moderate alcohol intake 57 (23.1) 43 (22.2) 29 (20.0) Insurance status Private 145 (58.5) 105 (53.6) 84 (57.1) Governmental 75 (30.4) 76 (38.8) 53 (36.1) None/free care 28 (11.3) 15 (7.6) 10 (6.8) Etiology of pancreatitis Gallstones 86 (34.7) 71 (36.9) 36 (24.5) Alcohol 49 (19.8) 37 (19.3) 39 (26.5) Post-ERCP 36 (14.5) 19 (9.9) 15 (10.2) Idiopathic 29 (11.7) 36 (18.8) 27 (18.4) Drug-induced 15 (5.2) 7 (3.7) 4 (2.7) Other 35 (14.1) 22 (11.5) 26 (17.7) Median length of stay (d) 4 (2.5, 8.0) 4 (3.0, 9.0) 4 (3.0, 9.0) NOTE. Data presented as number (percentage), median (25th percentile, 75th percentile). ERCP, endoscopic retrograde cholangiopancreatography. venous antimicrobials, and opiates included oral and intravenous opioid medications. Demographic factors including age, gender, race, insurance status, smoking status, alcohol intake, and body mass index (BMI) were included for analysis. Clinical characteristics including etiology of pancreatitis and transfer status were also evaluated. These potential risk factors are further defined in Table 1. The outcome was early readmission defined as a rehospitalization or reevaluation in the emergency department within 30 days of discharge from an index hospitalization for AP. Data Analysis All univariate relationships were examined by using a 2 or Fisher exact test, as appropriate. Covariates found to be associated with early readmission on univariate analysis at a level of significance P.1 were eligible for inclusion in a forward selection multiple logistic regression model to identify independent predictors of early readmission. For the final model, significance was defined at a level of P.05. Risk factors identified in the final model were evaluated for significant interaction at a level of P.05. The results of the multivariate analysis were then used to develop a clinical prediction model. Each beta coefficient was rounded to the nearest integer to create a scoring system. A weighted risk score for an individual patient was determined by assigning points for each risk factor present and adding the results. Risk scores in the derivation cohort were grouped together into low, moderate, and high risk on the basis of similar frequencies of early readmissions. Finally, the validation cohort was used to test the clinical prediction model. Weighted risk scores were calculated for patients in the validation cohort, and patients were grouped into low, moderate, and high risk. Readmission rates for these groups were again calculated for the validation cohort. Performance of the weighted risk scores in the derivation cohort and the validation cohort was also compared by using receiveroperating characteristic (ROC) curve analysis. Calibration of the model was assessed by plotting observed and predicted event frequencies by deciles of risk by using the Hosmer Lemeshow goodness-of-fit statistic. 20 All statistical analyses were performed by using the SAS System for Windows, version 9.2 (SAS Institute, Inc, Cary, NC). Results Study Cohort During the study period there were a total of 593 patients admitted for AP, 338 in the derivation cohort between June 1, 2005, and December 31, 2007, and 255 in the validation cohort between January 1, 2008, and October 31, After applying exclusion criteria, the cohort consisted of 248 patients in the derivation cohort and 198 in the validation cohort. Table 3. Independent Risk Factors for Early Readmission Risk factor coefficient P value Points Eating less than a solid diet Gastrointestinal symptoms (nausea, vomiting, or diarrhea) Pancreatic necrosis Antibiotics Pain NOTE. Points calculated by rounding off coefficient to nearest integer.

4 178 WHITLOCK ET AL CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 9, No. 2 Derivation Cohort Univariate correlates (P.1) of early readmission in the derivation cohort eligible for entry into the model included female gender, hypoxia, leukocytosis, gastrointestinal symptoms, pain, tolerating less than a solid diet, use of antibiotics, use of opiates, abdominal drains, invasive devices, moderate alcohol, pancreatic necrosis, alcohol-induced pancreatitis, and transfer. When univariate correlates were entered into forward stepwise logistic regression analysis, the independent correlates (P.05) of an early readmission in the derivation cohort were eating less than a solid diet; gastrointestinal symptoms of nausea, vomiting, or diarrhea; pancreatic necrosis; use of antibiotics; and pain at the time of discharge (Table 3). The Hosmer Lemeshow goodness-of-fit statistic was 0.36 for the model in the derivation cohort, and a calibration figure was constructed (Figure 1). The nonsignificant P value from the Hosmer Lemeshow measure demonstrates that the observed early readmission rates and the predicted early readmission rates were statistically similar across the risk groups defined by the test; this demonstrates that the model has good fit. Figure 1. Calibration curve for model. Observed vs predicted probability of an early readmission in the derivation and validation cohorts. The diagonal line would be a perfect fit. For early readmission the Hosmer Lemeshow goodness-of-fit test statistic is 0.36 for the derivation cohort and 0.50 for the validation cohort. Baseline characteristics among the 2 cohorts as well as the excluded group were similar with respect to age, gender, race, BMI, smoking status, alcohol intake, insurance status, etiology, and hospital length of stay (Table 2). Rates of early readmission were also similar between the derivation and validation cohorts at 19% (47/248) and 23% (45/198), respectively. Weighted Risk Score When points were assigned to each of the predictors on the basis of their beta coefficients (Table 3), the model identified groups with increasing risk of early readmission (Figure 2). For example, a patient with none of the risk factors would receive 0 points and fall into the low-risk group, a patient only with pancreatic necrosis would receive 2 points and fall into the moderate-risk group, and a patient with pancreatic necrosis, on antibiotics, and in pain on discharge would receive 5 points and fall into the high-risk group. Of 248 patients with an index admission for AP, 162 (65%) fell into the low-risk group (0 1 points), of whom only 4% (6/162) required early readmission. Forty-seven (19%) fell into the moderate-risk group (2 3 points), of whom 15% (7/47) required early readmission, and 39 (16%) fell into the high-risk group ( 4 points), of whom 87% (34/39) required early readmission (Figure 2). Validation Cohort When weighted risk scores were applied to the validation cohort, they successfully stratified patients according to risk for readmission. As demonstrated in Figure 2, the low-risk group (0 1 points) included 57% (112/198) and had an early readmission rate of 5% (6/112). The moderate-risk group (2 3 points) included 20% (39/198) of the cohort and had an early readmission rate of 18% (7/39), and the high-risk group ( 4 points) comprised 24% (47/198) and had an early readmission rate of 68% (32/47). When the ROC curves for the weighted risk score models in the derivation and validation cohorts were compared (Figures 3 Figure 2. Rates of early readmission in derivation and validation cohorts based on weighted risk score.

5 February 2011 PREDICTING EARLY READMISSION IN AP 179 Figure 3. ROC for early readmission risk score in the derivation cohort. Figure 4. ROC for early readmission risk score in the validation cohort. and 4), the area under the curve (AUC) for the derivation cohort was 0.88, and the AUC for the validation cohort was 0.83, demonstrating similar performance. Discussion We have derived and validated a straightforward risk score that allows stratification of patients at the time of discharge according to their risk for an early readmission. In this risk score, points are assigned for the presence of each of the following: (1) eating less than a solid diet at discharge; (2) gastrointestinal symptoms of nausea, vomiting, or diarrhea at discharge; (3) pancreatic necrosis; (4) use of antibiotics at discharge; and (5) pain at discharge. Patients identified as high risk in this model ( 4 points) represented one-fourth of the cohort and had a risk of early readmission of at least 68%. The model also identified a low-risk group (0 1 points) of patients representing more than half of the entire cohort. In these patients the probability of early readmission was 5%. Few reports to date have specifically examined early readmission in AP. We recently published data describing a 19% rate of early readmissions as well as the characteristics of the patients and their readmissions. 10 Exploratory analyses from this prior study identified discharge on less than a solid diet and persistent gastrointestinal symptoms at discharge as risk factors for early readmission. In the present study, we have confirmed these associations as well as identified the important role of pancreatic necrosis, use of antibiotics, and pain on discharge in determining a patient s risk of early readmission. We have extended these observations to the development of a simple weighted risk score that can assist physicians in their discharge planning efforts. Because this model was derived and validated across a wide spectrum of disease severities, the variables included in the model likely represent strong markers for an assortment of conditions in both mild and severe pancreatitis that predispose patients to early readmission. To remember these risk factors and the order of their importance we propose the pneumonic SNNAP, which stands for the following: S, gastrointestinal Symptoms of nausea, vomiting, or diarrhea at discharge; N, Nutrition for discharge on less than a solid diet; N, pancreatic Necrosis; A, use of Antibiotics at discharge; and P, Pain at discharge. This study is clinically important because it provides a simple means for identifying and stratifying patients at high and low risk for early readmission after an episode of AP. With this information, a clinician might decide to delay discharge of a patient at high risk for readmission, increase the level of outpatient care, or schedule earlier clinic followup. In addition, the weighted risk score can be used to identify a group of patients with a low likelihood of early readmission that a clinician can feel reasonably safe discharging. Reducing the rates of early readmission in AP ultimately leads to improved overall care and might reduce overall health care costs. This study has several limitations. First, some of the discharge information was extracted retrospectively via chart review. Although the sources of these data were standardized discharge forms and nursing flow sheets, the possibility of under-reporting exists. Second, 10% of our cohort was excluded as a result of lack of follow-up, and it is possible this group differed with respect to rates of readmission or other clinical factors, demographics, etiology of pancreatitis, and severity of pancreatitis. Nonetheless analysis of the entire group of excluded patients suggested that they did not differ significantly from the study population. Third, this is a single center experience and might not be generalizable to other institutions.

6 180 WHITLOCK ET AL CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 9, No. 2 In summary, we have developed and validated a simple clinical risk score for determining a patient s risk for early readmission after an episode of AP. The 5 independent predictors of early readmission in AP are (1) discharge on less than a solid diet; (2) gastrointestinal symptoms including nausea, vomiting, or diarrhea at discharge; (3) pancreatic necrosis; (4) use of antibiotics at discharge; and (5) pain at discharge. This new scoring system might aid clinicians in their decisions regarding discharge planning for patients with AP. References 1. Jencks SF, Williams MV, Coleman EA, Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med 2009;360: Krumholz HM, Chen YT, Wang Y, et al. Predictors of readmission among elderly survivors of admission with heart failure. Am Heart J 2000;139: Capelastegui A, Espana PP, Bilbao A, et al. Pneumonia: criteria for patient instability on hospital discharge. Chest 2008;134: Reznik M, Hailpern SM, Ozuah PO. Predictors of early hospital readmission for asthma among inner-city children. J Asthma 2006;43: Jasti H, Mortensen EM, Obrosky DS, et al. Causes and risk factors for rehospitalization of patients hospitalized with community acquired pneumonia. Clin Infect Dis 2008;46: Kind AJ, Smith MA, Frytak JR. Bouncing-back: patterns and predictors of complicated transitions thirty days after hospitalization for acute ischemic stroke. J Am Geriatr Soc 2007;55: Comette P, D Hoore W, Malhomme B, et al. Differential risk factors for early and later hospital readmission of older patients. Aging Clin Exp Res 2005;17: Salamzadeh J, Wong IC, Hosker HS, et al. A Cox regression analysis of covariates for asthma hospital readmissions. J Asthma 2003;40: Alonso Martinez JL, Llorente Diez B, Echegaray Agara M. Hospital readmission in internal medicine. An Med Interna 2001;18: Whitlock TL, Repas K, Tignor A. Early readmission in acute pancreatitis: incidence and risk factors. Am J Gastroenterol 2010; 105: Fagenholz PJ, Fernandez-del Castillo C, Harris NS, et al. Direct medical costs of acute pancreatitis hospitalizations in the United States. Pancreas 2007;35: Brown A, Young B, Morton J, et al. Are health related outcomes in acute pancreatitis improving? An analysis of national trends in the US from 1997 to JOP 2008;9: Stack LC, Lannon PB, Miley AD. Accuracy of clinicians expectancies for psychiatric rehospitalization. Am J Community Psych 1983;11: Adamiak GT, Karlberg I. Impact of physician training level on emergency readmission within internal medicine. Internat. Int J Technol Assess Health Care 2004;20: Hasan O, Meltzer DO, Shaykevich SA, et al. Hospital readmission in general medicine patients: a prediction model. J Gen Intern Med 2010;25: Tanaka M, Tamamoto H, Kita T, et al. Early prediction of the need for non-routine discharge planning for the elderly. Arch Gerontol Geriatr 2008;47: Spanier BW, Dijkgraaf MG, Bruno MJ. Epidemiology, aetiology and outcome of acute and chronic pancreatitis. An update. Best Pract Res Clin Gastroenterol 2008;22: Singh VK, Wu BU, Bollen TL, et al. A prospective evaluation of the bedside index for severity in acute pancreatitis score in assessing mortality and intermediate markers of severity in acute pancreatitis. Am J Gastroenterol 2009;104: Singh VK, Wu BU, Bollen TL, et al. Early systemic inflammatory response syndrome is associated with severe acute pancreatitis. Clin Gastroenterol Hepatol 2009;7: Lemeshow S, Hosmer DWH. A review of goodness of fit statistics for use in the development of logistic regression models. Am J Epidemiol 1982;115: Reprint requests Address requests for reprints to: Tom Whitlock, MD, Brigham and Women s Hospital, Division of Gastroenterology, 75 Francis Street, Boston, Massachusetts Tom_Whitlock@post.harvard. edu; fax: (617) Conflicts of interest The authors disclose no conflicts of interest.