Msandri.it

Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH of Rebleeding in Bleeding Gastroduodenal Ulcer Background and Study Aims: The aims of this study were to accuracy of the predictive model was 71 % (95 % CI = 63 – 79 %).
identify risk factors for recurrence of hemorrhage in bleeding The model showed a better sensitivity of 90 % for early rebleed- gastroduodenal ulcers after endoscopic injection therapy, and to ing (< 48 hours) than for late rebleeding (‡ 48 hours) where the develop a simple and relevant prognostic score which could be sensitivity was 65 %. A prognostic score was obtained and pa- used to assess the early risk of recurrence and the residual risk tients were classified into four risk classes: very low (VL), low (L), high (H), and very high (VH). The rebleeding rates for the Patients and Methods: A prospective study was conducted from four classes were 0 %, 7.9 %, 31.8 % and 67.9 %, and the mortality January 1995 to December 1998, in 738 patients who were ad- rates were 5.9 %, 8.6 %, 13.9 % and 35.7 %, respectively. The resid- mitted to our department for acute bleeding peptic ulcer and ual risk of rebleeding after 48 hours was 0 %, 3.3 %, 10.4 %, and who underwent endoscopic examination. Ulcers with active 14.3 % in the VL, L, H and VH classes, respectively. After 5 days bleeding or signs of recent bleeding were treated with injection the residual risk was under 4 % in all classes.
therapy using epinephrine (1/10 000) and 1 % polidocanol.
Conclusions: This study demonstrates that the proposed prog- Results: Multivariate analysis revealed that liver cirrhosis, recent nostic score, which is easily obtained after emergency endosco- surgery, systolic blood pressure below 100 mmHg, hematemesis, py, is useful in clinical practice because it can identify patients Forrest classification, and ulcer size and site were significantly with different levels of rebleeding risk. It can be helpful in pa- predictive variables for the recurrence of hemorrhage. Among tient management and decision making for discharge.
these, Forrest classification was the most important. The overall treatment of rebleeding would improve the outcome in such pa-tients.
Gastroduodenal peptic ulcer is the most frequent cause of acutehemorrhage of the upper digestive tract, being responsible in Several clinical factors and endoscopic signs have been found to about 50 % of cases, with an overall mortality rate of 10 to 14 % be associated with further hemorrhage [8]. However, as individ- [1 – 4]. Endoscopic therapy represents the treatment of choice ual factors, they lack predictive accuracy. Scoring systems and and provides effective control of bleeding peptic ulcers. The rate mathematical models have been proposed, to stratify patients of initial hemostasis provided by injection therapy is greater into different outcome categories and improve the prediction of than 90 %, but the incidence of rebleeding remains high, from 10 rebleeding, but usually their complexity has limited their appli- to 30 % [5 – 8]. Recurrence of hemorrhage is one of the most im- cation in routine clinical situations.
portant factors affecting the prognosis, and early prediction and 1 First Department of General Surgery, Verona University Medical School, Ospedale Maggiore Borgo Trento, Verona, Italy 2 Calcolo Scientifico, Servizi Informatici di Ateneo University of Verona, Verona, Italy A. Guglielmi, M.D. · First Department of General Surgery · Verona University Medical School · Ospedale Maggiore Borgo Trento · Piazzale Stefani 1 · 37126 Verona · Italy Fax: + 39-45-8345355 · E-mail: [email protected] Submitted 24 September 2001 · Accepted after Revision 16 May 2002 Endoscopy 2002; 34 (10): 771–779  Georg Thieme Verlag Stuttgart · New York · ISSN 0013-726X Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH While the stigmata that predict rebleeding in the absence of en- after ulcer healing. For this reason H. pylori test results were not doscopic therapy have been studied extensively, few studies included in the data analysis of the present study.
have specifically investigated stigmata which are predictive ofrebleeding after endoscopic therapy [9,10].
We considered three groups of variables, related to patient his-tory, magnitude of bleeding, and endoscopic findings. The clini- The aim of this study was to identify the clinical and endoscopic cal variables related to patient history were: gender; age; bleed- risk factors for rebleeding after endoscopic treatment, and to de- ing at home or during hospitalization; previous peptic ulcer dis- vise a mathematical model for assessing the risk of early rebleed- ease; previous gastrointestinal hemorrhage; intake of nonsteroi- ing. In addition, we classified patients into four categories asso- dal anti-inflammatory drugs (NSAIDs), and/or anticoagulant ciated with an increasing risk of rebleeding.
drugs; associated diseases; recent surgical operations (within30 days prior to the hemorrhage), or previous operations (morethan 30 days). With regard to operations, we considered only major cardiovascular, thoracic, orthopedic, abdominal and neu- rosurgical procedures. Concomitant diseases were classified All the patients with acute gastric and/or duodenal ulcer bleed- into seven groups: malignancies, liver cirrhosis, chronic renal ing, who presented consecutively to our endoscopy service be- failure, arterial hypertension, diabetes mellitus, rheumatic dis- tween January 1 1995 and December 31 1998 were included in eases, and peripheral vasculopathy (ischemic heart disease, cere- this study. Our endoscopy service is the 24-hour referral center brovascular accidents, peripheral arteriopathy).
for emergency endoscopies in the city area. Previous gastroduo-denal surgery or the presence of malignant ulcers were criteria The variables related to the magnitude of bleeding were: hema- temesis; coffee-ground vomit; melena; anemia; blood pressure;heart rate; hypovolemic shock; hematocrit and hemoglobin level Patients admitted to the study presented clinical signs and/or at admission; and number of units of blood transfused before en- symptoms of recent or active gastroduodenal bleeding, i. e. he- matemesis, coffee-ground vomit, melena, or anemia. After initialstabilization, all patients underwent emergency endoscopy The endoscopic variables were: the number, size and site of pep- within 2 hours of admission to our department. In all cases endo- tic ulcers; the Forrest classification; and the presence of gastritis scopic examination confirmed active or recent bleeding from or duodenitis. Peptic ulcer sites were grouped as the fundus–cor- gastroduodenal ulcer. Endoscopy was performed in all cases by pus, the antropyloric region, the duodenal bulb, and the postbul- members of the same team of four experienced physicians who bar region. In the case of multiple ulcers, we considered the le- use the same criteria of diagnosis and treatment. Stigmata of ac- sion at greatest risk of recurrence according to the Forrest classi- tive or recent bleeding were categorized according to the Forrest classification [11]. During the emergency endoscopy, all ulcers with active bleeding, a visible nonbleeding vessel, or an adherent clot (Forrest Ia, Ib, IIa, IIb) were treated by endoscopic injection Sample size was planned following the formula proposed by with epinephrine (1/10 000) and 1 % polidocanol. Lesions with a black ulcer base or a clean base (Forrest IIc, III) were treated bymedication.
Continuous variables were categorized according the systemshown in Table 1. Variables with more than two categories were Patients received blood transfusions before endoscopy if their recorded using an appropriate variable coding scheme. As sug- hemoglobin level was below 8 g/dl. Patients with severe hemor- gested by Kramer [13], cutoff points were defined according to rhagic shock, or with massive bleeding which could not be con- the existing medical literature [1,10,14].
trolled with endoscopic therapy, underwent emergency surgery.
All patients received intravenous ranitidine 50 mg three times a All the covariates of known clinical relevance, and those whose day and, after refeeding, omeprazole 20 mg twice a day. All pa- univariate test (chi-squared for categorical variables and Ken- tients underwent repeat endoscopy 48 hours after definitive dall’s tau for ordinal variables) had a P-value of less than 0.25 haemostasis, or earlier in cases of clinical suspicion of recur- were considered candidates for entry into the logistic model [15].
In the logistic regression model, we estimated the optimal cutoff Clinical suspicion of recurrence was defined as the presence of point using the maximum discrimination point criterion which hematemesis, melena, hypovolemia or a decrease in haemoglo- maximises the quantity: (sensitivity + specificity)/2 [16].
bin level by 2 g/dl after initial stabilization. In all cases the clini- cal suspicion of recurrence was confirmed by endoscopy. The Many works in the field of statistical forecasting show that, for therapy of choice for recurrence was endoscopic injection ther- obtaining an unbiased estimate of the classification accuracy of apy. Patients with rebleeding which could not be controlled by a predictive model, the classic in-sample goodness-of-fit meas- endoscopic therapy underwent emergency surgery.
ures are not adequate [17,18]. In fact, they quantify the closeness of the model predictions to the in-sample data, but they usually We analysed recurrences which happened within 30 days of the give no information regarding the true out-of-sample prediction acute bleeding episode. Biopsy-based Helicbacter pylori testing capabilities of the model. Data-splitting methods represent a was not done during emergency endoscopy, but was carried out valuable solution to the above problem. The sample is repeatedly and randomly divided in two subsets: the learning set and the Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH testing set [19]. The parameters of the model are estimated n The 30-day mortality rate was 10 % in the whole sample: 51 times from the n learning sets, while the corresponding testing deaths (8.5 %) in the group without recurrence of bleeding and sets are used to calculate the (out-of-sample) percentages of cor- 23 deaths (23.5 %) in the group with rebleeding (P = 0.001). Six rectly classified patients. The mean and 95 % confidence intervals patients died of hemorrhagic shock (8.1 %), and the remaining (95 % CI) of the n “bootstrap” estimates give a reliable description patients died from causes unrelated to bleeding.
of the true predictive performance of the model. The data-split-ting estimate gives a prospective assessment of the predictive The results of the univariate analysis of clinical and endoscopic variables are presented in Table 1. Among these, endoscopic find-ings, with the exception of the number of ulcers, showed the In the Discussion section, we use odds ratios as rough approxi- highest crude odds ratios. Gastritis or duodenitis was observed mations of risk ratios (odds ratios are good estimates of risk ra- in association with peptic ulcer in 44 % of patients (n = 325). This association was more frequent in Forrest III ulcers (116, 61.7 %)than in Forrest II and Forrest I ulcers, with 156 cases (43.6 %) Univariate and multivariate analyses were undertaken using and 53 cases (27.6 %), respectively (P < 0.01).
STATA 7.0 (chi-squared, Kendall’s tau, and multivariate logisticregression model estimations) and MATLAB 5.3, together with The multivariate logistic model estimated for the risk of rebleed- the Stixbox toolbox (data splitting, repeated logistic regression ing is shown in Table 2 (Pearson’s goodness-of-fit test, P = 0.86; model estimations, sensitivity and specificity data-splitting esti- Hosmer–Lemeshow test, P = 0.77). The significant predictive variables were: liver cirrhosis, recent surgery, systolic bloodpressure at admission, hematemesis, Forrest classification, ulcersize, and ulcer site. Although univariate analysis showed a signif- icant association between rebleeding and the covariate “bleed-ing during hospitalisation,” multivariate analysis indicated that The present study was conducted from January 1, 1995 to De- this variable had no predictive power and therefore could be ex- cember 31, 1998. During this period, 1597 patients with acute cluded from the model. The estimated cutoff point of the model nonvariceal upper gastrointestinal bleeding were seen at our en- was 0.163. If the probability of rebleeding is higher than the doscopic service. Among these, 738 patients (46.2 %) with acute threshold value, the patient is classified as being at risk, and gastroduodenal bleeding ulcer were included in this study. Ta- vice versa. The sensitivity, specificity and total (in-sample) accu- ble 1 gives the general characteristics of the study population.
racy of the model were all equal to 76 %. The mean values of sen- Among the 738 patients, bleeding recurred in 98 patients sitivity, specificity and total (out-of-sample) accuracy obtained (13.3 %) within 30 days from the first observation, with a mean by means of 10 000 data-splitting operations were all approxi- number of 1.44 rebleeding episodes. A single recurrence was ob- mately equal to 71 % (95 % CI = 63 – 79 %). The bootstrap estimate served in 67.3 % (66 patients), two recurrences in 24.4 % (24 pa- of the cutoff point was 0.148 (95 % CI = 0.11 – 0.20). The receiver tients), three in 4.1 % (four patients), four in 2.1 % (two patients) operating characteristic (ROC) curve (Figure 1) shows different and five in 2.1 % (two patients). The highest rebleeding rates levels of sensitivity and specificity for different cutoff points.
were observed between 12 and 24 hours (20.4 %) and between24 and 48 hours (32.7 %). In 18 cases (18.4 %), the rebleeding Figure 2 shows the residual percentage of rebleeding for patients time ranged from 96 hours to a maximum of 11 days after the correctly and incorrectly (false negatives) classified by the mod- first hemorrhagic episode. No rebleedings were observed after el. Within the false-negative group, only 13.4 % of recurrences took place before 24 hours, whereas in the correctly classifiedpatient group 42.5 % of recurrences happened before 24 hours.
Endoscopic injection therapy was performed in 447 patients In addition, patients in the false-negative group had a greater (60.6 %) with Forrest Ia, Ib, IIa and IIb ulcers, while those with tendency to rebleed after 48 hours. This means that the proposed Forrest IIc and III ulcers received only medical therapy. In pa- logistic model predicted short-term rebleeding with greater ac- tients with active bleeding, endoscopic injection therapy curacy, whereas most of the classification errors related to later achieved hemostasis in 95.9 % of cases. In four patients the bleed- hemorrhagic events. This fact is more clearly illustrated in Fig- ing was not controlled at first endoscopic treatment: two of ure 3, which shows the sensitivity curves of the model for ulcers these patients underwent emergency surgery and two patients for early (< 48 hours) and late (‡ 48 hours) rebleeding compared died of hemorrhagic shock before surgery could be carried out.
with the sensitivity curve for overall rebleeding (dotted line).
Taking a cutoff point of 0.163, the model had a markedly superior Endoscopic examination was carried out in all cases of recur- sensitivity for early rebleedings (89 %) compared with later ones rence. Endoscopic treatment proved effective in 86.8 % of cases of first rebleeding and in 62.5 % of patients with two or more re-bleedings. A total of 32 patients (4.3 %) underwent emergency Using the coefficients of the estimated logistic model, it is possi- surgical operations after effective first endoscopic therapy: 25 ble to calculate a prognostic score which utilizes four clinical for recurrent bleeding which was not controlled by endoscopic variables, i. e. liver cirrhosis (LV), recent surgery (RS), blood pres- treatment, and seven for endoscopic complications which in- sure (BP), and hematemesis (HM), and three endoscopic vari- cluded gastric or duodenal perforation (six patients) and necro- ables, ulcer size (S), ulcer site (L), and Forrest class (F). Values sis of the gastric wall (one patient).
for these variables can be easily collected at the time of the first endoscopic examination. The prognostic score is then given by: Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH Table 1 Variables analysed in relation to rebleeding with crude odds ratios and 95 % confidence intervals Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH Significant predictor variables for rebleeding B is the coefficient of the variable in the logistic regression model and SE its standard error. P is the statistical significance for the hypothesis that B = 0. Exp(B) is the oddsratio and 95 % CIthe 95 % confidence interval of exp(B). * With endoscopic injection therapy.
0.83 LV + 0.92 RS + 1.30 BP + 0.45 HM + 1.91 F1 + 2.42 F2 + The mortality rates for the VL, L, H, and VH groups, were 5.9 %, 2.36 F3 + 2.67 F4 + 2.59 F5 + 0.42 S1 + 1.33 S2 + 0.17 L1 + 8.6 %, 13.9 %, and 35.7 %, respectively.
In order to identify four classes of patients with an increasing risk of rebleeding, we estimated three cutoff points for the prog-nostic score: if the score is less than 1.66, the patient is classified Endoscopic therapy has proved effective in control bleeding in as being at very low risk (VL); if the score is higher than 5.75, acute peptic ulcer, but the recurrence rate is still 10 – 30 % [5 – 8].
3.91, or 1.66, the patient is classified as being at very high risk The recurrence of bleeding is one of the most important factors (VH), high risk (H), or low risk (L), respectively. Table 3 shows affecting prognosis. In our study, the observed mortality was the distribution of the patient sample in the four risk classes.
three times higher in patients with rebleeding, as is reported in The rebleeding rate in these classes consistently rose, from 0 % the literature [22]. The early identification of patients with an in- in the VL class to 7.9 %, 31.8 %, and 67.9 %, in the L, H, and VH creased risk of recurrence may improve the outcome.
classes, respectively. After 48 hours the residual risk of rebleed-ing decreased to 3.3 %, 10.4 % and 14.3 % for the L, H and VH In our study we used endoscopic injection with a combination of groups, respectively, as shown in Figure 4. After 5 days, all cat- epinephrine and polidocanol because this technique is widely egories of patients showed a residual risk lower than 4 %, which used and is one of the most effective endoscopic treatments [9, 23]. The timing of endoscopic monitoring, the usefulness ofendoscopic re-treatment, and appropriate treatment for recur- rence are still matters of debate [24, 25].
Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH ROC curve. Sensitivity and specificity for various cutoff Figure 3 Sensitivity curves for early and late bleeding, and for bleed- ing overall. The overall specificity curve is also shown. For a cutoff pointof 0.163, the model shows a sensitivity of 89 % for early rebleeding and66 % for late rebleeding Residual precentage of rebleeding patients Table 3 Distributions of patients within the four riskclasses: very low (VL), low (L), high (H) and very high (VH) Residual percentage of rebleeding after admission, for cor- rectly classified patients, patients with a false-negative result, and the We adopted endoscopic injection as the treatment of choice for rebleeding in all patients, using emergency surgery only when endoscopy was not effective. Rates for emergency surgery re- ported in the literature vary from 10 % to 20 % of patients[8, 22, 26]. In this study, using a restrictive surgical policy, wehave achieved a low emergency surgery rate of 4.6 % with a mor- tality of 10 %; this compares favorably with other studies, where before the bleeding episode, were associated with a significantly death has occurred in 5 – 15 % [9, 22, 23, 27].
increased risk of rebleeding, roughly doubling that risk. Chronic renal failure, though associated with an increased risk, did not Many patient-related variables are recognized as being associat- reach statistical significance, unlike the findings reported by ed with the occurrence of rebleeding [28, 29]. In our study, only other investigators [28]. However, it did correlate significantly the presence of liver cirrhosis, and recent surgery in the 30 days with mortality (crude odds ratio = 4.28, P < 0.01). In our sample, Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH categorization, obtained from seven variables, available after en- doscopic examination, can be used to improve patient manage- ment and selection of appropriate treatment.
The patients belonging to the lowest risk class (VL) showed a re- currence of 0 %. They should be discharged after the first endo-scopic observation, provided that they are hemodynamically stable and without severe co-morbidity.
In the low risk (L) category, the recurrence rate was 7.9 %, andmost of the rebleedings happened in the first 48 hours. Only 3.3 % of the recurrences took place after 48 hours. For this reason,clinical observation should be prolonged for at least 48 hours.
For patients belonging to the high risk (H) category, the recur- rence rate was 31.8 %. In this group of patients clinical observa- tion is necessary because of the high risk of recurrence, but pro- Residual riskof rebleeding according to different classes of grammed endoscopic examinations and endoscopic treatment of recurrence achieved satisfactory results. To arrive at a residualrisk of rebleeding of less than 4.0 %, patients should stay in hospi-tal for at least 5 days.
age was not found to be a risk factor, thus confirming the findings In the very high risk (VH) group, endoscopic injection therapy did not provide satisfactory results, with a recurrence rate of67.9 % and a mortality rate of 35.7 %. These results might be im- Amongst symptoms, the presence of hematemesis increased the proved by early elective therapy, such as surgery or second-look risk by approximately 50 %. Analysing the relationship between endoscopy with prophylactic sclerotherapy, an approach which hematemesis and ulcer site, we found that ulcers located in the has also been supported by other authors [4, 9].
duodenum re-bled in about 25.0 % of cases when associatedwith hematemesis, and in 10.0 % of cases in the presence of other This study demonstrates that the proposed prognostic score, symptoms (melena, coffee-ground vomit, anemia). Moreover, which is easily obtained after emergency endoscopy, can identify hematemesis increased the risk of rebleeding in ulcers within patients at different risk of rebleeding and can be helpful in pa- the same Forrest class. Ulcers with active bleeding in patients tient management and decisions about discharge.
with hematemesis re-bled in 28.2 % of cases, and only in 14.0 % of cases in the presence of other symptoms.
Endoscopic variables (in particular the Forrest classification)were found to be the most influential variables for the risk of re- The authors gratefully acknowledge the technical assistance of D.
bleeding, in agreement with the data reported in the literature Cristofalo and the many helpful suggestions of S. Golia and M.
[10,14, 26, 32 – 38]. Patients with active bleeding (Forrest Ia and Ib) and those with signs of recent bleeding (Forrest IIa, IIb)showed an approximately tenfold greater risk of recurrence ofhemorrhage than those with a clean ulcer base (Forrest III).
The predictive performance of our model yielded overall results 1 Rockall TA, Logan RF, Devlin HB, Northfield TC. Risk assessment after similar to those reported by other authors [8, 26]. These models acute upper gastrointestinal haemorrhage. Gut 1996; 38: 316 – 321 2 Rockall TA, Logan RF, Devlin HB, Northfield TC. Incidence of and mor- are not widely applied in routine clinical situation on account of tality from acute upper gastrointestinal haemorrhage in the United their complexity. In addition, time-dependent studies concern- Kingdom. Steering Committee and members of the National Audit of ing the residual risk of rebleeding following therapeutic endos- Acute Upper Gastrointestinal Haemorrhage. Br Med J 1995; 311: copy are rare. In our study, in order to investigate this feature, 3 Consensus conference. Therapeutic endoscopy and bleeding ulcers.
we followed up all patients for 30 days after their acute episode.
4 Imhof M, Schroders C, Ohmann C, Roher H. Impact of early operation Our model performed markedly better in predicting early re- on the mortality from bleeding peptic ulcer – ten years’ experience.
bleeding (< 48 hours) than late rebleeding, with sensitivities of 5 Sacks HS, Chalmers TC, Blum AL et al. Endoscopic hemostasis. An ef- 90 % and 70 %, respectively. We found that most recurrences fective therapy for bleeding peptic ulcers. JAMA 1990; 264: 494 – 499 were within the first 48 hours, but we observed rebleedings up 6 Cook DJ, Guyatt GH, Salena BJ, Laine LA. Endoscopic therapy for acute to 11 days from the first observation.
nonvariceal upper gastrointestinal hemorrhage: A meta-analysis. Gas-troenterology 1992; 102: 139 – 148 7 Messmann H, Schaller P, Andus T et al. Effect of programmed endo- To improve the clinical usefulness of the predictive model, we scopic follow-up examinations on the rebleeding rate of gastric or constructed a prognostic score and classified patients into four duodenal peptic ulcers treated by injection therapy: A prospective categories of increasing risk of recurrence and mortality. This randomized controlled trial. Endoscopy 1998; 30: 583 – 589 Art. "955", 16.8.02/ENDOSC 2001/00386//Mihr GmbH 8 Villanueva C, Balanzo J, Espinos JC et al. Prediction of therapeutic fail- 25 Aabakken L. Nonvariceal upper gastrointestinal bleeding. Endoscopy ure in patients with bleeding peptic ulcer treated with endoscopic in- jection. Dig Dis Sci 1993; 38: 2062 – 2070 26 Jaramillo JL, Galvez C, Carmona C et al. Prediction of further hemor- 9 Saeed ZA, Cole RA, Ramirez FC et al. Endoscopic retreatment after suc- rhage in bleeding peptic ulcer. Am J Gastroenterol 1994; 89: 2135 – cessful initial hemostasis prevents ulcer rebleeding: A prospective randomized trial. Endoscopy 1996; 28: 288 – 294 27 Brullet E, Campo R, Calvet X et al. Factors related to the failure of en- 10 Bornman PC, Theodorou NA, Shuttleworth RD et al. Importance of hy- doscopic injection therapy for bleeding gastric ulcer. Gut 1996; 39: povolaemic shock and endoscopic signs in predicting recurrent haem- orrhage from peptic ulceration: A prospective evaluation. Br Med J 28 Coleman SY, Pritchett CJ, Wong J, Branicki FJ. Risk models for rebleed- ing and postoperative mortality in bleeding gastric ulcer. Ann R Coll 11 Forrest JA, Finlayson ND, Shearman DJ. Endoscopy in gastrointestinal 29 Pundzius J. Clinical and endoscopic signs for the prediction of recur- 12 Whittemore AS. Sample size for logistic regression with small re- rent bleeding from gastroduodenal ulcers. Eur J Surg 1994; 160: sponse probability. J Am Statist Assoc 1981; 76: 27 – 32 13 Kramer MS. Clinical epidemiology and biostatistics: A primer for clin- 30 Lin H-J, Tseng G-Y, Lo W-C et al. Predictive factors for rebleeding in pa- ical investigators and decision-makers. Berlin: Springer-Verlag, 1988 tients with peptic ulcer bleeding after multipolar electrocoagulation. J 14 Chow LW, Gertsch P, Poon RT, Branicki FJ. Risk factors for rebleeding and death from peptic ulcer in the very elderly. Br J Surg 1998; 85: 31 Hsu P-I, Lin X-Z, Chan S-H. Bleeding peptic ulcer – risk factors for re- bleeding and sequential changes in endoscopic findings. Gut 1994; 35: 15 Mickey RM, Greenland S. The impact of confounder selection criteria on effect estimation. Am J Epidemiol 1989; 129: 125 – 137 32 Corley DA, Stefan AM, Wolf M et al. Early indicators of prognosis in up- 16 Cramer JS. Predictive performance of the binary logit model in unbal- per gastrointestinal hemorrhage. Am J Gastroenterol 1998; 93: 336 – anced samples. Statistician 1999; 48 Part 1: 85 – 94 17 Schwarzer G, Vach W, Schumacher M. On the misuses of artificial 33 Branicki FJ, Coleman SY, Fok PJ et al. Bleeding peptic ulcer: A prospec- neural networks for diagnostic classification in oncology. Statist Med tive evaluation of risk factors for rebleeding and mortality. World J 18 Adams NM, Hand DJ. Improving the practice of classifier performance 34 Lin HJ, Perng CL, Lee FY et al. Clinical courses and predictors for re- assessment. Neural Computat 2000; 12: 305 – 311 bleeding in patients with peptic ulcers and non-bleeding visible ves- 19 Picard RR, Berk KN. Data splitting. Am Statist 1990; 44: 140 – 147 sels: A prospective study. Gut 1994; 35: 1389 – 1393 20 Gould W. Interpreting logistic regression in all its forms. Stata Techni- 35 Hsu PI, Lai KH, Lin XZ et al. When to discharge patients with bleeding peptic ulcers: a prospective study of residual risk of rebleeding. Gas- 21 Holtsberg A. Stixbox: A toolbox for Matlab. http://www.maths.lth.se/ 36 Wara P. Endoscopic prediction of major rebleeding – a prospective 22 Bourienne A, Pagenault M, Heresbach D et al. Étude prospective multi- study of stigmata of hemorrhage in bleeding ulcer. Gastroenterology centrique des facteurs pronostiquØs des hØmorragies ulcØreuses gas- troduodØnales. Gastroenterol Clin Biol 2000; 24: 193 – 200 37 Kolkman JJ, Meuwissen SG. A review on treatment of bleeding peptic Rutgeerts P, Rauws E, Wara P et al. Randomised trial of single and re- ulcer: a collaborative task of gastroenterologist and surgeon. Scand J peated fibrin glue compared with injection of polidocanol in treat- ment of bleeding peptic ulcer. Lancet 1997; 350: 692 – 696 38 Brullet E, Campo R, Bedos G et al. Site and size of bleeding peptic ulcer.
24 Palmer KR. Ulcers and nonvariceal bleeding. Endoscopy 2000; 32: Is there any relation to the efficacy of hemostatic sclerotherapy? En-

Source: http://www.msandri.it/docs/Endoscopy_2002.pdf