Predicting Hospital Mortality in Critically Ill Cancer Patients according to Acute Kidney Injury Severity

Background: Acute kidney injury (AKI) is a frequent complication in hospitalized patients, especially in those in intensive care units (ICU). The RIFLE classification might be a valid prognostic factor for critically ill cancer patients. The present study aims to evaluate the discriminatory capacity of RIFLE versus other general prognostic scores in predicting hospital mortality in critically ill cancer patients. Methods: This is a single-center study conducted in a cancer-specialized ICU in Brazil. All of the 288 patients hospitalized from May 2006 to June 2008 were included. RIFLE classification, APACHE II, SOFA, and SAPS II scores were calculated and the area under receiver operating characteristic (AROC) curves and logistic multiple regression were performed using hospital mortality as the outcome. Results: AKI, defined by RIFLE criteria, was observed in 156 (54.2%) patients. The distribution of patients with any degree of AKI was: risk, n = 96 (33.3%); injury, n = 30 (10.4%), and failure, n = 30 (10.4%). Mortality was 13.6% for non-AKI patients, 49% for RIFLE `R` patients, 62.3% for RIFLE `I` patients, and 86.8% for RIFLE `F` patients (p = 0.0006). Logistic regression analysis showed that RIFLE criteria, APACHE II, SOFA, and SAPS II were independent factors for mortality in this population. The discrimination of RIFLE was good (AROC 0.801, 95% CI 0.748-0.854) but inferior compared to those of APACHE II (AROC 0.940, 95% CI 0.915-0.966), SOFA (AROC 0.910, 95% CI 0.876-0.943), and SAPS II (AROC 0.869, 95% CI 0.827-0.912). Conclusion: AKI is a frequent complication in ICU patients with cancer. RIFLE was inferior to commonly used prognostic scores for predicting mortality in this cohort of patients. Copyright (C) 2011 S. Karger AG, Basel

Doença de Kawasaki: experiência clínica em hospital universitário

OBJETIVO: A doença de Kawasaki é uma vasculite sistêmica aguda de etiologia desconhecida. Seu diagnóstico baseia-se em critérios clínicos. O objetivo deste estudo foi descrever os casos de pacientes com doença de Kawasaki internados no Hospital Universitário da Universidade de São Paulo entre janeiro/2000 e junho/2008. MÉTODOS: Dentre todos os pacientes internados na Enfermaria de Pediatria no período acima, foram selecionados aqueles cujo CID de alta foi doença de Kawasaki. Realizou-se estudo descritivo por meio da análise dos prontuários dessas crianças. RESULTADOS: Foram encontrados 18 casos. A média de internações foi de 2,1 casos/ano. A idade variou de três meses a nove anos. A proporção meninos:meninas foi 1:1,25. Receberam outros diagnósticos prévios 17 pacientes, sendo escarlatina em 2/3 dos casos. O tempo de febre antes do diagnóstico variou de cinco a 11 dias. Nove crianças apresentaram quatro sinais sugestivos de doença de Kawasaki; oito apresentaram cinco sinais e uma apresentou dois sinais, o que foi considerado doença de Kawasaki incompleta. Receberam gamaglobulina 15 crianças (entre o sexto e o décimo dias de evolução) e 11 (73%) ficaram afebris após infusão da medicação. Os demais tiveram febre até 24 horas após a administração. Todos os pacientes realizaram ecocardiograma e três apresentaram aneurisma leve da coronária. CONCLUSÕES: A doença de Kawasaki é habitualmente confundida com outras doenças, o que causa retardo no tratamento e aumento no risco de complicações cardíacas.

Adverse drug events in an intensive care unit of a university hospital

Purpose Adverse drug events (ADEs) are harmful and occur with alarming frequency in critically ill patients. Complex pharmacotherapy with multiple medications increases the probability of a drug interaction (DI) and ADEs in patients in intensive care units (ICUs). The objective of the study is to determine the frequency of ADEs among patients in the ICU of a university hospital and the drugs implicated. Also, factors associated with ADEs are investigated. Methods This cross-sectional study investigated 299 medical records of patients hospitalized for 5 or more days in an ICU. ADEs were identified through intensive monitoring adopted in hospital pharmacovigilance and also ADE triggers. Adverse drug reactions (ADR) causality was classified using the Naranjo algorithm. Data were analyzed through descriptive analysis, and through univariate and multiple logistic regression. Results The most frequent ADEs were ADRs type A, of possible causality and moderate severity. The most frequent ADR was drug-induced acute kidney injury. Patients with ADEs related to DIs corresponded to 7% of the sample. The multiple logistic regression showed that length of hospitalization (OR = 1.06) and administration of cardiovascular drugs (OR = 2.2) were associated with the occurrence of ADEs. Conclusion Adverse drug reactions of clinical significance were the most frequent ADEs in the ICU studied, which reduces patient safety. The number of ADEs related to drug interactions was small, suggesting that clinical manifestations of drug interactions that harm patients are not frequent in ICUs.

A comparison of observed versus estimated baseline creatinine for determination of RIFLE class in patients with acute kidney injury

Methods. Data from the Beginning and Ending Supportive Therapy for the Kidney (BEST Kidney) study, a prospective observational study from 54 ICUs in 23 countries of critically ill patients with severe AKI, were analysed. The RIFLE class was determined by using observed (o) pre-morbid and estimated (e) baseline SCr values. Agreement was evaluated by correlation coefficients and Bland-Altman plots. Sensitivity analysis by chronic kidney disease (CKD) status was performed. Results. Seventy-six percent of patients (n = 1327) had a pre-morbid baseline SCr, and 1314 had complete data for evaluation. Forty-six percent had CKD. The median (IQR) values were 97 mu mol/L (79-150) for oSCr and 88 mu mol/L (71-97) for eSCr. The oSCr and eSCr determined at ICU admission and at study enrolment showed only a modest correlation (r = 0.49, r = 0.39). At ICU admission and study enrolment, eSCr misclassified 18.8% and 11.7% of patients as having AKI compared with oSCr. Exclusion of CKD patients improved the correlation between oSCr and eSCr at ICU admission and study enrolment (r = 0.90, r = 0.84) resulting in 6.6% and 4.0% being misclassified, respectively. Conclusions. While limited, estimating baseline SCr by the MDRD equation when pre-morbid SCr is unavailable would appear to perform reasonably well for determining the RIFLE categories only if and when pre-morbid GFR was near normal. However, in patients with suspected CKD, the use of MDRD to estimate baseline SCr overestimates the incidence of AKI and should not likely be used. Improved methods to estimate baseline SCr are needed.

Timing of renal replacement therapy and clinical outcomes in critically ill patients with severe acute kidney injury

Purpose: The aim of this study is to evaluate the relationship between timing of renal replacement therapy (RRT) in severe acute kidney injury and clinical outcomes. Methods: This was a prospective multicenter observational study conducted at 54 intensive care units (ICUs) in 23 countries enrolling 1238 patients. Results: Timing of RRT was stratified into ""early"" and ""late"" by median urea and creatinine at the time RRT was started. Timing was also categorized temporally from ICU admission into early (<2 days), delayed (2-5 days), and late (>5 days). Renal replacement therapy timing by serum urea showed no significant difference in crude (63.4% for urea <= 24.2 mmol/L vs 61.4% for urea >24.2 mmol/L; odds ratio [OR], 0.92; 95% confidence interval [CI], 0.73-1.15; P = .48) or covariate-adjusted mortality (OR, 1.25; 95% CI, 0.91-1.70; P = .16). When stratified by creatinine, late RRT was associated with lower crude (53.4% for creatinine >309 mu mol/L vs 71.4% for creatinine <= 309 mu mol/L; OR, 0.46; 95% CI, 0.36-0.58; P < .0001) and covariate-adjusted mortality (OR, 0.51; 95% CI, 0.37-0.69; P < .001).However, for timing relative to ICU admission, late RRT was associated with greater crude (72.8% vs 62.3% vs 59%, P < .001) and covariate-adjusted mortality (OR, 1.95; 95% CI, 1.30-2.92; P = .001). Overall, late RRT was associated with a longer duration of RRT and stay in hospital and greater dialysis dependence. Conclusion: Timing of RRT, a potentially modifiable factor, might exert an important influence on patient survival. However, this largely depended on its definition. Late RRT (days from admission) was associated with a longer duration of RRT, longer hospital stay, and higher dialysis dependence. (C) 2009 Elsevier Inc. All rights reserved.

Impact of mild acute kidney injury (AKI) on outcome after open repair of aortic aneurysms

Recently, mild AKI has been considered as a risk factor for mortality in different scenarios. We conducted a retrospective analysis of the risk factors for two distinct definitions of AKI after elective repair of aortic aneurysms. Logistic regression was carried out to identify independent risk factors for AKI ( defined as >= 25% or >= 50% increase in baseline SCr within 48 h after surgery, AKI 25% and AKI 50%, respectively) and for mortality. Of 77 patients studied ( mean age 68 +/- 10, 83% male), 57% developed AKI 25% and 33.7% AKI 50%. There were no differences between AKI and control groups regarding comorbidities and diameter of aneurysms. However, AKI patients needed a supra-renal aortic cross-clamping more frequently and were more severely ill. Overall in-hospital mortality was 27.3%, which was markedly higher in those requiring a supra-renal aortic cross-clamping. The risk factors for AKI 25% were suprarenal aortic cross-clamping ( odds ratio 5.51, 95% CI 1.05-36.12, p = 0.04) and duration of operation for AKI 25% ( OR 6.67, 95% CI 2.23-19.9, p < 0.001). For AKI 50%, in addition to those factors, post-operative use of vasoactive drugs remained as an independent factor ( OR 6.13, 95% CI 1.64-22.8, p = 0.005). The risk factors associated with mortality were need of supra-renal aortic cross-clamping ( OR 9.6, 95% CI 1.37-67.88, p = 0.02), development of AKI 50% ( OR 8.84, 95% CI 1.31-59.39, p = 0.02), baseline GFR lower than 49 mL/min ( OR 17.07, 95% CI 2.00 145.23, p = 0.009), and serum glucose > 118 mg/dL in the post-operative period ( OR 19.99, 95% CI 2.32-172.28, p = 0.006). An increase of at least 50% in baseline SCr is a common event after surgical repair of aortic aneurysms, particularly when a supra-renal aortic cross-clamping is needed. Along with baseline moderate chronic renal failure, AKI is an independent factor contributing to the high mortality found in this scenario.

Small airway remodeling in acute respiratory distress syndrome: a study in autopsy lung tissue

Introduction: Airway dysfunction in patients with the Acute Respiratory Distress Syndrome (ARDS) is evidenced by expiratory flow limitation and dynamic hyperinflation. These functional alterations have been attributed to closure/obstruction of small airways. Airway morphological changes have been reported in experimental models of acute lung injury, characterized by epithelial necrosis and denudation in distal airways. To date, however, no study has focused on the morphological airway changes in lungs from human subjects with ARDS. The aim of this study is to evaluate structural and inflammatory changes in distal airways in ARDS patients. Methods: We retrospectively studied autopsy lung tissue from subjects who died with ARDS and from control subjects who died of non pulmonary causes. Using image analysis, we quantified the extension of epithelial changes (normal, abnormal and denudated epithelium expressed as percentages of the total epithelium length), bronchiolar inflammation, airway wall thickness, and extracellular matrix (ECM) protein content in distal airways. The Student`s t test or the Mann-Whitney test was used to compare data between the ARDS and control groups. Bonferroni adjustments were used for multiple tests. The association between morphological and clinical data was analyzed by Pearson rank test. Results: Thirty-one ARDS patients (A: PaO(2)/FiO(2) <= 200, 45 +/- 14 years, 16 males) and 11 controls (C:52 +/- 16 years, 7 males) were included in the study. ARDS airways showed a shorter extension of normal epithelium (A:32.9 +/- 27.2%, C:76.7 +/- 32.7%, P < 0.001), a larger extension of epithelium denudation (A:52.6 +/- 35.2%, C:21.8 +/- 32.1%, P < 0.01), increased airway inflammation (A:1(3), C:0(1), P = 0.03), higher airway wall thickness (A:138.7 +/- 54.3 mu m, C:86.4 +/- 33.3 mu m, P < 0.01), and higher airway content of collagen I, fibronectin, versican and matrix metalloproteinase-9 (MMP-9) compared to controls (P = 0.03). The extension of normal epithelium showed a positive correlation with PaO(2)/FiO(2) (r(2) = 0.34; P = 0.02) and a negative correlation with plateau pressure (r(2) = 0.27; P = 0.04). The extension of denuded epithelium showed a negative correlation with PaO(2)/FiO(2) (r(2) = 0.27; P = 0.04). Conclusions: Structural changes in small airways of patients with ARDS were characterized by epithelial denudation, inflammation and airway wall thickening with ECM remodeling. These changes are likely to contribute to functional airway changes in patients with ARDS.

A quantitative coronary angiography-matched comparison between a prospective randomised multicentre cutting balloon angioplasty and bare metal stent trial (REDUCE III) and the Rapamycin-Eluting Stent Evaluation At Rotterdam Cardiology Hospital (RESEARCH) study

Aims: There remains significant concern about the long-term safety of drug-eluting stents (DES). However, bare metal stents (BMS) have been used safely for over two decades. There is therefore a pressing need to explore alternative strategies for reducing restenosis with BMS. This study was designed to examine whether IVUS-guided cutting balloon angioplasty (CBA) with BMS could convey similar restenosis rates to DES. Methods and results: In the randomised REstenosis reDUction by Cutting balloon angioplasty Evaluation (REDUCE III) study, 521 patients were divided into four groups based on device and IVUS use before BMS (IVUS-CBA-BMS: 137 patients; Angio-CBA-BMS: 123; IVUS-BA-BMS: 142; and Angio-BA-BMS: 119). At follow-up, the IVUS-CBA-BMS group had a significantly lower restenosis rate (6.6%) than the other groups (p=0.016). We performed a quantitative coronary angiography (QCA) based matched comparison between an IVUS-guided CBA-BMS strategy (REDUCE III) and a DES strategy (Rapamycin-Eluting-Stent Evaluation At Rotterdam Cardiology. Hospital, the RESEARCH study). We matched the presence of diabetes, vessel size, and lesion severity by QCA. Restenosis (>50% diameter stenosis at follow-up) and target vessel revascularisation (TVR) were examined. QCA-matched comparison resulted in 120-paired lesions. While acute gain was significantly greater in IVUS-CBA-BMS than DES (1.65 +/- 0.41 mm vs. 1.28 +/- 0.57 mm, p=0.001), late loss was significantly less with DES than with IVUS-CBA-BMS (0.03 +/- 0.42 mm vs. 0.80 +/- 0.47 mm, p=0.001). However, no difference was found in restenosis rates (IVUS-CBA-BMS: 6.6% vs. DES: 5.0%, p=0.582) and TVR (6.6% and 6.6%, respectively). Conclusions: An IVUS-guided CBA-BMS strategy yielded restenosis rates similar to those achieved by DES and provided an effective alternative to the use of DES.

Depression: a predictor of smoking relapse in a 6-month follow-up after hospitalization for acute coronary syndrome

Objective The objective of the study was to investigate whether depression is a predictor of postdischarge smoking relapse among patients hospitalized for myocardial infarction (MI) or unstable angina (ILIA), in a smoke-free hospital. Methods Current smokers with MI or UA were interviewed while hospitalized; patients classified with major depression (MD) or no humor disorder were reinterviewed 6 months post discharge to ascertain smoking status. Potential predictors of relapse (depression; stress; anxiety; heart disease risk perception; coffee and alcohol consumption; sociodemographic, clinical, and smoking habit characteristics) were compared between those with MD (n = 268) and no humor disorder (n = 135). Results Relapsers (40.4%) were more frequently and more severely depressed, had higher anxiety and lower self-efficacy scale scores, diagnosis of UA, shorter hospitalizations, started smoking younger, made fewer attempts to quit, had a consort less often, and were more frequently at the `precontemplation` stage of change. Multivariate analysis showed relapse-positive predictors to be MD [odds ratio (OR): 2.549; 95% confidence interval (CI): 1.519-4.275] (P<0.001); `precontemplation` stage of change (OR: 7.798; 95% CI: 2.442-24.898) (P<0.001); previous coronary bypass graft surgery (OR: 4.062; 95% CI: 1.356-12.169) (P=0.012); and previous anxiolytic use (OR: 2.365; 95% CI: 1.095-5.107) (P=0.028). Negative predictors were diagnosis of MI (OR: 0.575; 95% CI: 0.361-0.916) (P=0.019); duration of hospitalization (OR: 0.935; 95% CI: 0.898-0.973) (P=0.001); smoking onset age (OR: 0.952; 95% CI: 0.910-0.994) (P=0.028); number of attempts to quit smoking (OR: 0.808; 95% CI: 0.678-0.964) (P=0.018); and `action` stage of change (OR: 0.065; 95% CI: 0.008-0.532) (P= 0.010). Conclusion Depression, no motivation, shorter hospitalization, and severity of illness contributed to postdischarge resumption of smoking by patients with acute coronary syndrome, who underwent hospital-initiated smoking cessation.

Lithium increases plasma brain-derived neurotrophic factor in acute bipolar mania: A preliminary 4-week study

Several studies have suggested an important role for brain-derived neurotrophic factor (BDNF) in the pathophysiology and therapeutics of bipolar disorder (BPD). The mechanisms underlying the therapeutic effects of lithium in BPD seem to involve a direct regulation of neurotrophic cascades. However, no clinical study evaluated the specific effects of lithium on BDNF levels in subjects with BPD. This study aims to investigate the effects of lithium monotherapy on BDNF levels in acute mania. Ten subjects with bipolar I disorder in a manic episode were evaluated at baseline and after 28 days of lithium therapy. Changes in plasma BDNF levels and Young Mania Rating Scale (YMRS) scores were analyzed. A significant increase in plasma BDNF levels was observed after 28 days of therapy with lithium monotherapy (510.9 +/- 127.1 pg/mL) compared to pre-treatment (406.3 +/- 69.5 pg/mL) (p = 0.03). Although it was not found a significant association between BDNF levels and clinical improvement (YMRS), 87% of responders presented an increase in BDNF levels after treatment with lithium. These preliminary data showed lithium`s direct effects on BDNF levels in bipolar mania, suggesting that short-term lithium treatment may activate neurotrophic cascades. Further studies with larger samples and longer period may confirm whether this biological effect is involved in the therapeutic efficacy of lithium in BPD. (C) 2011 Elsevier Ireland Ltd. All rights reserved.

Heart-lung interactions with different ventilatory settings during acute lung injury and hypovolaemia: an experimental study

Background. The functional haemodynamic variables pulse pressure variation (PPV), stroke volume variation (SVV), and systolic pressure variation (SPV) are widely used to assess haemodynamic status. However, it is not known how these perform during acute lung injury (ALI). This study evaluated the effects of different ventilatory strategies on haemodynamic parameters in pigs with ALI during normovolaemia and hypovolaemia. Methods. Eight anaesthetized Agroceres pigs [40 (1.9) kg] were instrumented with pulmonary artery, PiCCO, and arterial catheters and ventilated. Three ventilatory settings were randomly assigned for 10 min each: tidal volume (VT) 15 ml kg(-1) and PEEP 5 cm H(2)O, VT 8 ml kg(-1) and PEEP 13 cm H(2)O, or VT 6 ml kg(-1) and PEEP 13 cm H(2)O. Data were collected at each setting at baseline, after ALI (lung lavage+Tween 1.5%), and ALI with hypovolaemia (haemorrhage to 30% of estimated blood volume). Results. At baseline, high VT increased PPV, SVV, and SPV (P < 0.05 for all). During ALI, high VT significantly increased PPV and SVV [(P = 0.002 and P = 0.008) respectively.]. After ALI with hypovolaemia, ventilation at VT 6 ml kg(-1) and PEEP 13 cm H(2)O decreased the accuracy of functional haemodynamic variables to predict hypovolaemia, with the exception of PPV (area under the curve 0.875). The parameters obtained by PiCCO were less influenced by ventilatory changes. Conclusions. VT is the ventilatory parameter which influences functional haemodynamics the most. During ventilation with low VT and high PEEP, most functional variables are less able to accurately predict hypovolaemia secondary to haemorrhage, with the exception of PPV.

Cardiopulmonary Effects of Matching Positive End-Expiratory Pressure to Abdominal Pressure in Concomitant Abdominal Hypertension and Acute Lung Injury

Background: To evaluate the cardiopulmonary effects of positive end-expiratory pressure (PEEP) equalization to intra-abdominal pressure (IAP) in an experimental model of intra-abdominal hypertension (IAH) and acute lung injury (ALI). Methods: Eight anesthetized pigs were submitted to IAH of 20 mm Hg with a carbon dioxide insufflator for 30 minutes and then submitted to lung lavage with saline and Tween (2.5%). Pressure x volume curves of the respiratory system were performed by a low flow method during IAH and ALI, and PEEP was subsequently adjusted to 27 cm center dot H(2)O for 30 minutes. Results: IAH decreases pulmonary and respiratory system static compliances and increases airway resistance, alveolar-arterial oxygen gradient, and respiratory dead space. The presence of concomitant ALI exacerbates these findings. PEEP identical to AP moderately improved oxygenation and respiratory mechanics; however, an important decline in stroke index and right ventricle ejection fraction was observed. Conclusions: Simultaneous IAH and ALI produce important impairments in the respiratory physiology. PEEP equalization to AP may improve the respiratory performance, nevertheless with a secondary hemodynamic derangement.

Time course of haemodynamic, respiratory and inflammatory disturbances induced by experimental acute pulmonary polystyrene microembolism

Background and objective The time course of cardiopulmonary alterations after pulmonary embolism has not been clearly demonstrated and nor has the role of systemic inflammation on the pathogenesis of the disease. This study aimed to evaluate over 12 h the effects of pulmonary embolism caused by polystyrene microspheres on the haemodynamics, lung mechanics and gas exchange and on interleukin-6 production. Methods Ten large white pigs (weight 35-42 kg) had arterial and pulmonary catheters inserted and pulmonary embolism was induced in five pigs by injection of polystyrene microspheres (diameter similar to 300 mu mol l(-1)) until a value of pulmonary mean arterial pressure of twice the baseline was obtained. Five other animals received only saline. Haemodynamic and respiratory data and pressure-volume curves of the respiratory system were collected. A bronchoscopy was performed before and 12 h after embolism, when the animals were euthanized. Results The embolism group developed hypoxaemia that was not corrected with high oxygen fractions, as well as higher values of dead space, airway resistance and lower respiratory compliance levels. Acute haemodynamic alterations included pulmonary arterial hypertension with preserved systemic arterial pressure and cardiac index. These derangements persisted until the end of the experiments. The plasma interleukin-6 concentrations were similar in both groups; however, an increase in core temperature and a nonsignificant higher concentration of bronchoalveolar lavage proteins were found in the embolism group. Conclusion Acute pulmonary embolism induced by polystyrene microspheres in pigs produces a 12-h lasting hypoxaemia and a high dead space associated with high airway resistance and low compliance. There were no plasma systemic markers of inflammation, but a higher central temperature and a trend towards higher bronchoalveolar lavage proteins were found. Eur J Anaesthesiol 27:67-76 (C) 2010 European Society of Anaesthesiology.

The epidemiology of acute respiratory failure in hospitalized patients: A Brazilian prospective cohort study

Purpose: The purpose of this study was to assess risk factors associated with the development of acute respiratory failure (ARF) and death in a general intensive care unit (ICU). Materials and Methods: Adults who were hospitalized at 12 surgical and nonsurgical ICUs were prospectively followed up. Multivariable analyses were realized to determine the risk factors for ARF and point out the prognostic factors for mortality in these patients. Results: A total of 1732 patients were evaluated, with an ARF prevalence of 57%. Of the 889 patients who were admitted without ARF, 141 (16%) developed this syndrome in the ICU. The independent risk factors for developing ARF were 64 years of age or older, longer time between hospital and ICU admission, unscheduled surgical or clinical reason for ICU admission, and severity of illness. Of the 984 patients with ARF, 475 (48%) died during the ICU stay. Independent prognostic factors for death were age older than 64 years, time between hospital and ICU admission of more than 4 days, history of hematologic malignancy or AIDS, the development of ARF in ICU, acute lung injury, and severity of illness. Conclusions: Acute respiratory failure represents a large percentage of all ICU patients, and the high mortality is related to some preventable factors such as the time to ICU admission. (C) 2011 Elsevier Inc. All rights reserved.

A decision support system to facilitate management of patients with acute gastrointestinal bleeding

Objective: To develop a model to predict the bleeding source and identify the cohort amongst patients with acute gastrointestinal bleeding (GIB) who require urgent intervention, including endoscopy. Patients with acute GIB, an unpredictable event, are most commonly evaluated and managed by non-gastroenterologists. Rapid and consistently reliable risk stratification of patients with acute GIB for urgent endoscopy may potentially improve outcomes amongst such patients by targeting scarce health-care resources to those who need it the most. Design and methods: Using ICD-9 codes for acute GIB, 189 patients with acute GIB and all. available data variables required to develop and test models were identified from a hospital medical records database. Data on 122 patients was utilized for development of the model and on 67 patients utilized to perform comparative analysis of the models. Clinical data such as presenting signs and symptoms, demographic data, presence of co-morbidities, laboratory data and corresponding endoscopic diagnosis and outcomes were collected. Clinical data and endoscopic diagnosis collected for each patient was utilized to retrospectively ascertain optimal management for each patient. Clinical presentations and corresponding treatment was utilized as training examples. Eight mathematical models including artificial neural network (ANN), support vector machine (SVM), k-nearest neighbor, linear discriminant analysis (LDA), shrunken centroid (SC), random forest (RF), logistic regression, and boosting were trained and tested. The performance of these models was compared using standard statistical analysis and ROC curves. Results: Overall the random forest model best predicted the source, need for resuscitation, and disposition with accuracies of approximately 80% or higher (accuracy for endoscopy was greater than 75%). The area under ROC curve for RF was greater than 0.85, indicating excellent performance by the random forest model Conclusion: While most mathematical models are effective as a decision support system for evaluation and management of patients with acute GIB, in our testing, the RF model consistently demonstrated the best performance. Amongst patients presenting with acute GIB, mathematical models may facilitate the identification of the source of GIB, need for intervention and allow optimization of care and healthcare resource allocation; these however require further validation. (c) 2007 Elsevier B.V. All rights reserved.