CLINICAL LIVER, PANCREAS, AND BILIARY TRACT

Transcription

1 GASTROENTEROLOGY 2003;124: CLINICAL LIVER, PANCREAS, AN BILIARY TRACT Emergency Sclerotherapy Versus Vasoactive rugs for Variceal Bleeding in Cirrhosis: A Cochrane Meta-Analysis GENNARO AMICO,* GIAA PIETROSI,* ILARIA TARANTINO,* and LUIGI PAGLIARO *epartment of Medicine, Ospedale V Cervello, Palermo; and Institute of Medicina Generale e Pneumologia, University of Palermo, Palermo, Italy Background & Aims: Emergency sclerotherapy is used as a first-line therapy for variceal bleeding in cirrhosis, although pharmacologic treatment stops bleeding in most patients. We performed a meta-analysis comparing emergency sclerotherapy with pharmacologic treatment. Methods: MELINE ( ),EMBASE ( ),and the Cochrane Library (2002;4) were searched to retrieve randomized led trials comparing sclerotherapy with vasopressin ( nitroglycerin), terlipressin,somatostatin,or octreotide for variceal bleeding in cirrhosis. Outcome measures were failure to bleeding,rebleeding,blood transfusions,adverse events,and Results: Fifteen trials were identified. Sclerotherapy was not superior to terlipressin, somatostatin,or octreotide for any outcome and to vasopressin for rebleeding,blood transfusions,death, and adverse events; it was superior to vasopressin for the of bleeding in a single trial flawed by a potential detection bias. Sclerotherapy was associated with significantly more adverse events than somatostatin. In a predefined sensitivity analysis,combining all of the trials irrespective of the treatment,risk differences (sclerotherapy minus ) and confidence intervals (CIs) were as follows: failure to bleeding, 0.03 ( 0.06 to 0.01); mortality, ( 0.07 to 0.008); adverse events,0.08 (0.02 to 0.14). Mortality risk difference was ( 0.07 to 0.04) in goodquality trials and 0.08 ( 0.14 to 0.02) in poorquality trials. Conclusions: Available evidence does not support emergency sclerotherapy as the first-line treatment of variceal bleeding in cirrhosis when compared with vasoactive drugs,which bleeding in 83% of patients. Therefore,endoscopic therapy might be added only in pharmacologic treatment failures. Sclerotherapy of esophageal varices is widely used and has been recommended 1,2 as the first-choice treatment of variceal bleeding in cirrhosis, although vasoactive drugs may stop bleeding in most patients 3,4 and several randomized led trials (RCTs) comparing emergency sclerotherapy with vasoactive drugs have yielded inconsistent results. 2 4 A recent meta-analysis showed that the efficacy of emergency endoscopic sclerotherapy or banding ligation of varices is significantly improved when associated with somatostatin or its derivatives, although the association does not improve survival. 5 However, emergency endoscopic therapy requires a skilled endoscopist, and sclerotherapy is associated with adverse events (AEs) in 10% 20% of patients and with serious AEs in 7%. 6 The efficacy of emergency banding ligation and its complications are poorly defined because they were assessed only in a few RCTs for the long-term prevention of variceal 4 By adding pharmacologic to endoscopic therapy, clearly more AEs should be expected than with either treatment alone. Because either endoscopic or pharmacologic therapy have been reported to bleeding in 75% 90% of patients, 1 4 initial treatment with the more effective therapy and their combination only in failures of initial treatment might be a more logical approach than their immediate combination. No RCTs comparing emergency banding ligation with vasoactive drugs have been reported, and it is still unclear whether emergency sclerotherapy is superior to pharmacologic therapy. Therefore, we performed a systematic review to assess the efficacy and AEs of emergency sclerotherapy compared with vasoactive drugs for acute variceal bleeding in cirrhotic patients. We report here the updating of a meta-analysis previously published in the Cochrane Library 6 after the inclusion of 3 more RCTs. Abbreviations used in this paper: AE,adverse events; AR,absolute risk difference; CI,confidence interval; RCT,randomized led trial; SAE,serious adverse events by the American Gastroenterological Association /03/$30.00 doi: /s (03)

2 1278 AMICO ET AL. GASTROENTEROLOGY Vol. 124,No. 5 Table 1. Criteria Used to Assess the Methodological Quality of the Included Studies Available as Full Reports Adequate generation of the randomization sequence Yes: random sequence generated by computer, table of random numbers, drawing of lots or envelopes, or similar methods suitable to prevent selection bias Not described. Other methods allowing potential selection bias such as case record number, date of birth date, date of admission, and so on were considered as quasi-randomization and caused the exclusion of the trial from the meta-analysis Adequate allocation concealment Yes: central randomization performed at a site remote from the location of the trial; sequentially numbered, sealed, opaque envelopes; similar methods by which investigators could not foresee treatment assignment No: procedures based on inadequate generation of randomization sequence, open allocation schedule, alternation, unsealed or nonopaque envelopes, permuted blocks with size known at the trial site, method not reported Blinding Yes: when both of the following criteria were fulfilled: additional treatments decided by physicians blinded to the trial treatment and outcome assessment performed by physicians blinded to the trial treatment No: when one or both of the above criteria were not fulfilled Intention-to-treat analysis Yes: when all of the randomized patients were included in the analysis and kept in their original groups regardless of their adherence to the study protocol No: intention-to-treat analysis, according to the above criterion, not reported Same treatment in the trial groups for failure to bleeding or rebleeding Yes No or not reported Same treatment in the trial groups for the prevention of rebleeding Yes No or not reported NOTE. etailed trial characteristics, definitions of the assessed outcomes, and treatment for failure to bleeding and to prevent rebleeding are reported in Tables 3 and 4. etailed quality assessment for each of the considered quality items for each trial is shown in Table 5. Materials and Methods Eligible RCTs were those comparing sclerotherapy with vasopressin ( nitroglycerin), terlipressin, somatostatin, or octreotide for acute variceal bleeding in patients with liver cirrhosis regardless of the etiology and severity of the liver disease, published either as full reports or abstracts, irrespective of the language. Trials using quasi-randomization methods and those lacking a direct comparison of emergency sclerotherapy with a vasoactive drug were excluded. Measures of treatment efficacy were as follows: (1) failure to bleeding (number of patients in which the interventions failed to the acute bleeding), (2) rebleeding (number of patients rebleeding before elective treatments for the prevention of rebleeding and number of patients rebleeding within 42 days or while in the hospital), (3) mortality (number of patients who died before receiving elective treatments for the prevention of rebleeding and number of patients who died within 42 days or while in the hospital), (4) transfusions (number of blood units transfused while in the hospital), and (5) AEs (number of patients with AEs and number of patients with serious adverse events [SAEs]). 7 Studies assessing at least one of these outcomes were considered for inclusion in the meta-analysis. When 42-day rebleeding or mortality was not reported, the figure at the longest available time interval, up to 42 days, was used. Retrieval of RCTs was based on the Cochrane Controlled Trials Register (The Cochrane Library, until July 2002), MELINE (1968 to July 2002), and EMBASE (1986 to July 2002). The following search terms were used: emergency sclerotherapy, vasopressin, terlipressin (or triglycyl lysine, va- Figure 1. Flow of RCTs through the process of retrieval and inclusion in the meta-analysis of RCTs comparing sclerotherapy with pharmacologic treatment for variceal bleeding in liver cirrhosis.

3 May 2003 SCLEROTHERAPY VS. RUGS FOR VARICEAL BLEEING 1279 Table 2. Characteristics of Participants in the Included Studies /C Child-Pugh class 61 A/B/C (%) Study (yr) No. of patients Mean age (yr) Males (%) Alcoholic etiology Ascites (%) Active bleeding on endoscopy (%) C Mean interval from bleeding to randomization (h) Sclerotherapy vs. vasopressin Westaby et al. 47 (1989) 33/31 55/54 52/62 39/71 100/100 12/51/36 23/45/32 Sclerotherapy vs. terlipressin Escorsell et al. 48 (2000) 114/105 55/56 68/76 41/39 43/35 17/52/31 23/45/32 6 Sclerotherapy vs. somatostatin i Febo et al. 49 (1990) 24/23 57/56 71/83 62/ / 56/ / Shields et al. 50 (1992) 41/39 59/57 63/72 63/72 61/69 18/41/41 15/20/ Planas et al. 51 (1994) 35/35 57/57 68/61 80/63 60/66 49/51 20/46/34 14/51/35 17 Escorsell et al. 29 (1998) 79/90 59/57 73/73 46/53 33/ a 8.2 a 56 Ramires et al. 52 (2000) 19/21 56/47 68/57 42/43 53/24 10/68/22 5/57/38 Sclerotherapy vs. octreotide Sung et al. 53 (1993) 49/49 57/55 79/90 35/43 59/53 37/51 14/43/43 12/43/45 Poo et al. 54 (1996) 21/22 76/77 38/23 Jenkins et al. 55 (1997) 77/73 52/57 57/57 65/63 63/49 15/32/53 16/30/53 21 Lopez et al. 56 (1999) 33/31 Bildozola et al. 57 (2000) 37/39 52/52 73/84 73/72 24/33 49/39 54/38/8 41/46/13 Sivri et al. 58 (2000) 36/30 48/46 32/41 25/21 87/91 100/100 11/36/53 8/37/55 Freitas et al. 59 (2000) 53/58 56/55 67/69 92/94 Stigmata of 24/41/35 20/45/35 12 bleeding in all Yousuf et al. 60 (2000) 48/48 45/46 75/81 0/0 100/100 46/31/23 40/42/18, endoscopic variceal sclerotherapy; C, ;, not reported. a Mean Child-Pugh score. 61 sopressin), somatostatin, octreotide. A manual search was also performed using the reference lists from articles, reviews, editorials, and the proceedings of international congresses. ecisions on which trials to include were unblindly taken independently by 2 of the reviewers (I. T. and G. P.). isagreements were resolved by discussion. Excluded trials were identified with the reason for exclusion. The methodological quality of the trials published as full reports was assessed by 4 major previously validated criteria 8 14 : adequate generation of the randomization sequence, adequate treatment allocation concealment, blinding, and intention-to-treat analysis. We also used 2 internal validity criteria that we considered relevant to the specific assessed treatments based on our clinical experience and previous consensus conferences : use of the same treatments in the 2 study groups for failure to bleeding and use of the same treatment for the prevention of recurrent bleeding. Each quality component was rated as yes or no (Table 1). The quality of trials was reported according to each separate component. 9,10,18 ata were abstracted by 2 independent reviewers (I. T. and G. P.). Agreement for each item averaged 82%. iscrepancies were solved by discussion. We performed separate meta-analyses according to the treatment because there are no convincing proofs that the vasoactive drugs we assessed as a treatment for emergency sclerotherapy are equivalent, even when the native hormones vasopressin and somatostatin are compared with their analogues terlipressin and octreotide, respectively However, trials were combined irrespectively of the sclerosant and modalities of performing sclerotherapy because there is no evidence that they may have different clinical efficacy. 2 Risk differences (sclerotherapy minus ) were pooled by a random effects model according to er Simonian and Laird 26 ; 95% confidence intervals (CI) are reported. A 2 test for heterogeneity was performed for each assessed outcome, and potential sources of heterogeneity were discussed where appropriate. For statistically significant risk differences (i.e., when the CI of the summary risk difference excludes 0), the number needed to treat for benefit or for harm and the corresponding CIs were calculated. 27,28 Because one of the included trials comparing sclerotherapy with somatostatin 29 admitted patients only after a 24-hour bleeding-free period, we performed a sensitivity analysis by excluding it when assessing the treatment effect on We also a priori decided to perform the following sensitivity analyses by combining all of the included trials, irrespective of the pharmacologic treatment, to increase power by including the largest number of available patients: (1) random versus fixed effects model, (2) full reports versus abstracts and for full reports only, (3) adequate versus not adequate or not reported generation of the randomization list and treatment allocation concealment, and (4) reported versus not reported intention-to-treat analysis. All of the results are reported as absolute risk difference (AR) such that a risk difference 0 favors sclerotherapy.

4 1280 AMICO ET AL. GASTROENTEROLOGY Vol. 124,No. 5 Table 3. Characteristics of Included Studies Study (yr), country Methods Participants Compared interventions Outcomes Notes Westaby et al. 47 (1989), United Kingdom Escorsell et al. 48 (2000), Spain i Febo et al. 49 (1990), Italy Shields et al. 50 (1992), United Kingdom Planas et al. 51 (1994), Spain Escorsell et al. 29 (1998), Spain Ramires et al. 52 (2000), Portugal Sung et al. 53 (1993), China Generation of the randomization list by random numbers table. Allocation concealment by method not described. Treatment completed as per protocol in all patients. None lost to follow-up. Blocked computer-generated randomization list. Allocation concealment by sealed, opaque envelopes. Treatment completed as per protocol in all patients. Two patients excluded because of serious protocol violation. None lost to follow-up. Randomization and allocation concealment by method not reported. No. of patients completing the assigned treatment not reported. None lost to follow-up. Blocked randomization list generated by a not-described method. Allocation concealment by sealed, opaque envelopes. Treatment completed as per protocol in all patients. None lost to follow-up. Computer-generated randomization list. Allocation concealment by sealed, opaque envelopes. Treatment completed as per protocol in all patients. None lost to follow-up. Computer-generated randomization list. Allocation concealment by opaque, sealed envelopes. Treatment performed as per protocol in all patients. None lost to follow-up. Method of generation of the randomization list unclear. Allocation concealment by opaque, sealed envelopes. Two patients were excluded after randomization because of incomplete treatment. None lost to follow-up. Computer-generated randomization list. Allocation concealment by opaque, sealed envelopes. Treatment completed in 98 of 100 randomized patients. None lost to follow-up. Cirrhotic patients with active variceal bleeding on endoscopy. Exclusion criteria not reported. Alcoholic cirrhosis significantly more frequent in the vasopressin group. Cirrhotic patients with endoscopy-proven variceal bleeding. Exclusion: recent sclerotherapy or band ligation, bleeding from fundal varices, other concomitant bleeding sources. Patients with endoscopy-proven variceal bleeding. No other inclusion/exclusion criteria reported. Etiology of varices not reported. Patients with endoscopy-proven variceal bleeding and either extrahepatic or intrahepatic portal hypertension. Exclusion: previous vasoactive therapy or. Patients with endoscopy-proven variceal bleeding. It is unclear whether only cirrhotic patients were admitted. Exclusion: bleeding from gastric varices. Cirrhotic patients achieving a 24-hour bleeding-free period after an endoscopy-proven variceal bleed and admitted 48 hours before. Exclusion: in the past week. Cirrhotic patients with endoscopy-proven variceal bleeding. Two of 40 included patients had an extrahepatic portal hypertension (portal vein thrombosis). Exclusion: hepatocellular carcinoma, decompensated diabetes mellitus, serum creatinine 2 mg/dl. Patients with endoscopy-proven variceal bleeding. : up to 5 ml intravariceal injections of 5% ethanolamine oleate, maximally 20 ml. Vasopressin: 20 IU IV bolus followed by 12-hour continuous infusion of 0.4 IU/min associated with g/min nitroglycerin. : intraparavariceal injection of either 5% ethanolamine or 1% polidocanol in a single session. Terlipressin: IV injections of 2 mg/4 h until of bleeding (48 hours maximum) and 1 mg/ 4 h thereafter for 5 more days : method not reported. Somatostatin: 48-hour continuous IV infusion of 250 g/h after an initial 250- g bolus. : 2 3 ml intravariceal injections of ethanolamine oleate in a single session. Somatostatin: Continuous IV infusion of 250 g/h for 5 days after initial IV bolus of 250 g. : intraparavariceal injection of 1% polidocanol (mean maximum volume, 48 ml) in a single session. Somatostatin: 48-hour continuous IV infusion of 250 g/h after an initial IV bolus of 250 g. : intraparavariceal injections of 5% ethanolamine or 1% polidocanol (maximally 20 ml). Somatostatin: 5-day continuous IV infusion of 250 g/h after an initial bolus of 250 g. : intravariceal injections of 5% ethanolamine (maximally 10 ml per varix). Somatostatin: 48-hour continuous IV infusion of 250- g/h plus 6 hourly 250- g IV boluses during the first 24-hour infusion. : intravariceal injections of 2 ml 3% sodium tetradecyl sulfate (maximal volume, 20 ml). Octreotide: 48-hour continuous IV infusion of 50 g/h after an initial IV bolus of 50 g. 12-hour failure to bleeding. Inhospital rebleeding and death. 48-hour failure to bleeding. Rebleeding within 5 days after initial. 42-day rebleeding and death. 48-hour failure to bleeding. Rebleeding within days day Control of bleeding on endoscopy. 5-day 28-day 48-hour failure to bleeding or Rebleeding within days day 5-day 42-day rebleeding and death. 48-hour failure to bleeding. 7-day 7-day 48-hour failure to bleeding. 48- hour 48-hour inhospital and 30-day Possible difference in the assessment of 12-hour failure between the 2 study groups. No adjusted analysis for unbalanced etiology was reported. RCT based on the Baveno Consensus Workshop guidelines. 17 Only abstract available. Need for transfusions was an entry criterion, but criteria for need were not reported. Study was based on the Baveno Consensus Workshopguidelines. 17 RCT aimed at the prevention of early rebleeding and based on the Baveno Consensus Workshop guidelines. 17 (continued on following page)

5 May 2003 SCLEROTHERAPY VS. RUGS FOR VARICEAL BLEEING 1281 Table 3. (continued) Study (yr), country Methods Participants Compared interventions Outcomes Notes Poo et al. 54 (1996), Mexico Jenkins et al. 55 (1997), United Kingdom Lopez et al. 56 (1999), Mexico Bildozola et al. 57 (2000), Argentina Sivri et al. 58 (2000), Turkey Freitas et al. 59 (2000), Portugal Yousuf et al. 60 (2000), Pakistan Randomization and allocation concealment by method not reported. No. of patients lost to follow-up or withdrawn from therapy not reported. Computer-generated randomization list in permuted blocks of 4. Allocation concealment by opaque, sealed envelopes. Treatment completed as per protocol in all patients. None lost to follow-up. Randomization and allocation concealment by method not reported. No. of patients completing the treatment not reported. No. lost to follow-up not reported. Randomization list generated by random numbers table. Allocation concealment by method not reported. Eight patients excluded after randomization because of protocol violations did not complete the assigned treatment. None lost to follow-up. Computer-generated randomization list. Allocation concealment by sealed, opaque envelopes. Treatment completed as per protocol in all patients. None lost to follow-up. Generation of the randomization list and allocation concealment by method not reported. No. lost to follow-up not reported. Complete results reported for only 104 of 111 patients. Randomization and allocation concealment by method not reported. Treatment completed as per protocol in all patients. None lost to follow-up. Cirrhotic patients with endoscopy-proven variceal bleeding. Exclusion: other sources of bleeding, previous vasoactive therapy or. Patients with endoscopy-proven variceal bleeding, heart rate 100 bpm, systolic blood pressure 100 mm Hg, or requiring 2 units of blood. Exclusion: or vasoactive drugs in the past 7 days. Child-Pugh class 61 BorC cirrhotic patients with endoscopy-proven variceal bleeding. Exclusion criteria not reported. Cirrhotic patients with endoscopy-proven variceal bleeding. Exclusion: referred patients already treated with vasoactive drugs or Sengstaken-Blakemore balloon tamponade, in the past 2 months. Cirrhotic patients with endoscopy-proven variceal bleeding. Exclusion: hepatocellular carcinoma. Cirrhotic patients with endoscopy-proven nonspurting variceal bleeding. Cirrhotic patients with endoscopy-proven variceal bleeding. Exclusions: coronary artery disease, cerebrovascular accident, pregnancy, respiratory failure. : intravariceal injection of 1% polidocanol. No other details available. Octreotide: 48-hour continuous IV infusion of 50 g/h. : intravariceal 2-3 ml injections of ethanolamine oleate. Maximal volume injected not reported. Octreotide: 48-hour continuous IV infusion of 50 g/h : polidocanol injections. etails not reported. Octreotide: 24-hour continuous IV infusion of 50 g/h. : intraparavariceal injections of 2% polidocanol up to a maximum of 25 ml. Octreotide: initial 100- g IV bolus and then 48-hour continuous IV infusion of 50 g/h followed by subcutaneous injections of 100 g every 8 hours for 72 hours. : intraparavariceal 2 3 ml injections of 1% polidocanol (maximal, 20 ml). Octreotide: 12-hour continuous IV infusion of 50 g/h after initial 50- g bolus. : intraparavariceal ml of absolute ethanol (maximum, 12 ml). Octreotide: 48-hour continuous IV infusion of 25 g/h. : intravariceal injections of sodium tetradecyl sulfate (volume not reported). Octreotide: 50- g subcutaneous injections every 6 hours. uration of treatment not reported. 48-hour 30-day rebleeding and 48-hour failure to bleeding. 48- hour rebleeding and death. 60-day 24-hour failure to bleeding. Mortality (time of assessment not reported). Failure to bleeding. 5-day 5-day 6-hour failure to bleeding. 72- hour In-hospital 48-hour failure to bleeding. 7-day 30-day 12-hour failure to bleeding. 36- hour treatment failure. 72-hour 72-hour Only abstract available. Sclerotherapy after the 48-hour trial period also in the octreotide group. Only abstract available. Time frame for assessing of bleeding unclear. All patients surviving the initial bleeding episode entered a weekly elective program. Patients with spurting variceal bleeding were included in a parallel trial of sclerotherapy plus octreotide vs. sclerotherapy alone., endoscopic variceal sclerotherapy; IV, intravenous. Publication bias was assessed by the Begg and Mazumdar adjusted rank correlation test for publication bias 30 and by the Egger et al. regression asymmetry test for publication bias. 31 All of the statistical analyses were performed by Review Manager 4.01 (The Cochrane Collaboration, Oxford, England), Metaview 4 (Update Software Ltd., Oxford, England), and STATA 6 (Stata Corp., College Station, TX). Results A total of 359 articles or abstracts were retrieved; 329 were excluded because of the reasons shown in Figure 1. Among 30 eligible studies, were excluded because of lack of a direct comparison of emergency sclerotherapy with a vasoactive drug. Fifteen RCTs

6 1282 AMICO ET AL. GASTROENTEROLOGY Vol. 124,No. 5 Table 4. efinition of Relevant Outcomes and Other Major Treatments Used in the Included Studies Study (yr) efinition of of bleeding efinition of rebleeding efinition of need for blood transfusion Treatment for failure to initial bleeding Long-term prevention of rebleeding Westaby et al. 47 (1989) Escorsell et al. 48 (2000) i Febo et al. 49 (1990) Shields et al. 50 (1992) Planas et al. 51 (1994) Escorsell et al. 29 (1998) Ramires et al. 52 (2000) Sung et al. 53 (1993) Absence of active variceal bleeding either by repeat endoscopy or nasogastric aspirate. 24-hour bleeding-free period within 48 hours after randomization. Bleeding was hematemesis, blood in gastric aspirate, hypovolemia with melena, or bleeding at endoscopy. No hematemesis and/or melena, Hb stable or decrease 2 g/dl, stable vital signs. Cessation of bleeding assessed endoscopically minutes after starting somatostatin or 10 minutes after. No hematemesis or melena with stable systolic blood pressure and heart rate during the 48-hour trial period. 24-hour bleeding-free period within 48 hours of admission. Bleeding was hematemesis, systolic blood pressure 80 mm Hg, and heart rate 120 bpm, hematocrit decrease 10 in 6 hours, blood in gastric aspirate. Absence of hematemesis or melena accompanied by systolic blood pressure 100 mm Hg or heart rate 100 bpm within 48 hours. Absence of bleeding at the 48-hour endoscopy. Absence of recurrent hematemesis or melena of 500 ml, systolic blood pressure 90 mm Hg, heart rate 110 bpm, or 6 units of blood or plasma within 12 hours. Endoscopy-proven variceal bleeding after stools had cleared of melena. New evidence of bleeding after a 24-hour bleeding-free period. Hematemesis and/or melena or decrease in Hb 2g/dL or decrease of vital signs in absence of hematemesis or melena. Further hematemesis or melena accompanied by either systemic disturbance or decrease in Hb. Recurrence of hemorrhage within 5 days after the 48-hour trial period: hematemesis or melena associated with decrease of systolic blood pressure 30 mm Hg, heart rate 120 bpm, or need for 2 units of blood to maintain vital signs. Hematemesis, hypovolemia (systolic blood pressure 80 mm Hg and heart rate 120 bpm), decrease in hematocrit 10 in 6 hours, blood in gastric aspirate. Hematemesis or melena accompanied by systolic blood pressure 100 mm Hg or heart rate 100 bpm. Rebleeding was always endoscopically confirmed. Recurrent hematemesis or melena of 500 ml, systolic blood pressure 90 mm Hg, heart rate 110 bpm, or 6 units of blood or plasma within 12 hours. To maintain hematocrit between 0.28 and To maintain vital signs. To maintain hemodynamic stability. To sustain blood pressure. Additional therapy if 2 or more blood transfusions were needed over 2 hours. In both groups: vasopressin nitroglycerin, balloon tamponade or sclerotherapy. Both groups: vasoactive therapy (except terlipressin), endoscopic therapy, balloon tamponade, surgical shunt, or TIPS. Both groups: sclerotherapy. Both groups: balloon tamponade. : somatostatin. Control:, Sengstaken-Blackmore tube. Treatment for rebleeding: alternative therapy, using either Sengstaken-Blackmore tube, terlipressin, TIPS, shunt surgery, or a combination of the above. Both groups: sclerotherapy. Both groups: balloon tamponade (Minnesota tube). Both groups: weekly sclerotherapy. Both groups, after the initial 6 7 day study period; endoscopic therapy, or pharmacologic therapy, or combination or surgical shunt, or TIPS. Both groups: sclerotherapy after discharge until day 60 from randomization. Treatment between days 5 and 28 unclear. Both groups: RCT of propranolol plus isosorbide 5-mononitrate vs. long-term sclerotherapy vs. shunt surgery. Both groups, after the 5- day study period:, pharmacologic therapy, or shunt surgery. Both groups: weekly sclerotherapy. (continued on next page)

7 May 2003 SCLEROTHERAPY VS. RUGS FOR VARICEAL BLEEING 1283 Table 4. (continued) Study (yr) efinition of of bleeding efinition of rebleeding efinition of need for blood transfusion Treatment for failure to initial bleeding Long-term prevention of rebleeding Poo et al. 54 (1996) Jenkins et al. 55 (1997) Lopez et al. 56 (1999) Bildozola et al. 57 (2000) Sivri et al. 58 (2000) Freitas et al. 59 (2000) Yousuf et al. 60 (2000) Absence of recurrent bleeding within 48 hours, defined as hematemesis or melena or 3 blood units to maintain systolic blood pressure 90 mm Hg or heart rate 100 bpm. Absence of hematemesis and/or melena with instability of vital signs or decrease of Hb requiring blood transfusion over the 48 hours following randomization. Based on endoscopy 24 hours after starting therapy. Criteria were not reported. Absence of fresh blood in the hourly gastric aspiration for 12 consecutive hours together with stability of hematocrit and vital signs. Stable blood pressure (no reduction 20 mm Hg after reaching stable values); stable Hb ( 9 g/dl); hematocrit 30% and blood transfusion requirement 2/2 hours or 4/24 hours. Absence of hematemesis or 100 ml of bright blood in nasogastric aspirate; no bright-red blood per rectum; no decrease in Hb 4 g/ dl in 48 hours; absence of shock and melena. Absence of blood in gastric aspirate for at least 1 hour and stable vital signs 12 hours after randomization. Recurrence of hematemesis or melena or 3 units of blood to maintain systolic blood pressure 90 mm Hg or heart rate 100 bpm. Hematemesis and/or melena with instability of vital signs or decrease in Hb requiring blood transfusion. To maintain systolic blood pressure 90 mm Hg or heart rate 100 bpm. Both groups: balloon tamponade. Both groups: balloon tamponade. Not assessed. Both groups: balloon tamponade. Recurrence of hematemesis or melena, hemodynamic instability (blood pressure 90 mm Hg and heart rate 100 bpm) and 5% decrease in hematocrit after initial of bleeding. Overt hemorrhage or aspiration of 100 ml of fresh blood; fresh melena; decrease of Hb 4 g/72 h; heart rate 100 bpm and blood pressure 100 mm Hg in the presence of continuing melena. Hematemesis or 100 ml of bright-red blood in nasogastric aspirate; bright-red blood per rectum; Hb decrease of 4 g/dl in 48 hours; shock and melena. Fresh blood in gastric aspirate at any time following initial of bleeding and unstable vital signs. To maintain constant hematocrit and vital signs. To maintain hematocrit 30%. Both groups: endoscopic therapy and/or balloon tamponade or surgery. Alternative therapy or balloon tamponade. Both groups: emergency sclerotherapy. When this failed, balloon tamponade, surgery, or other. Both groups: sclerotherapy. Treatment not reported, but all patients surviving the 5-day study period entered a RCT of treatment for long-term prevention of recurrent bleeding. Both groups:. Both groups: or band ligation at the end of the 48-hour trial period., not reported; TIPS, transjugular intrahepatic portosystemic shunt; Hb, hemoglobin;, endoscopic variceal sclerotherapy. were included, 29, of which were still available only as abstracts. 49,54,56 Among the included studies, one compared sclerotherapy with vasopressin plus nitroglycerin, 47 one with terlipressin, 48 5 with somatostatin, 29,49 52 and 8 with octreotide The principal characteristics of included patients are reported in Table 2 (for characteristics and methodological assessment of each included study, see Tables 3 5). Quantitative data synthesis and sensitivity analyses are reported in Tables 6 and 7. Sclerotherapy Compared With Vasopressin A total of 64 patients were included in the trial comparing sclerotherapy with vasopressin plus nitroglyc-

8 1284 AMICO ET AL. GASTROENTEROLOGY Vol. 124,No. 5 Table 5. Quality Assessment of the Fully Published Studies Study (yr) Adequate generation of the randomization sequence Adequate allocation concealment Blinding Intention-to-treat analysis Same treatment in the trial groups for failure to bleeding or rebleeding Same treatment in the trial groups for the prevention of rebleeding Westaby et al. 47 (1989) Yes No No Yes Yes Yes Escorsell et al. 48 (2000) Yes Yes No No Yes Yes Shields et al. 50 (1992) No Yes No Yes Yes Yes Planas et al. 51 (1994) Yes Yes No Yes Yes Yes Escorsell et al. 29 (1998) Yes Yes No Yes Yes Yes Ramires et al. 52 (2000) No Yes No No Yes Sung et al. 53 (1993) Yes Yes No No Yes Yes Jenkins et al. 55 (1997) Yes No No Yes Yes Yes Bildozola et al. 57 (2000) Yes No No No Yes Yes Sivri et al. 58 (2000) Yes Yes No Yes Yes Yes Freitas et al. 59 (2000) No No No No Yes Yes Yousuf et al. 60 (2000) No No No Yes No, not reported. erin. 47 Failure to bleeding was significantly reduced by sclerotherapy (AR, 0.23; CI, 0.44 to 0.03; number needed to treat for benefit, 4; CI, 2 33) but rebleeding and mortality were not. However, failure to bleeding was assessed 12 hours after starting treatment and, at this time, patients in the vasopressin group underwent variceal sclerotherapy as part of the elective sclerotherapy program for the long-term prevention of Therefore, at the time of assessment of failure to bleeding, patients in the vasopressin group but not those in the sclerotherapy group underwent endoscopy. This provided the medical group a different and more precise assessment of bleeding than the sclerotherapy group, which underwent repeat endoscopy only in the case of suspected persistence of bleeding, resulting in a potential although inadvertent detection bias. No significant differences were found between the 2 groups in the number of transfusions and AEs. Sclerotherapy Compared With Terlipressin The only trial comparing sclerotherapy with terlipressin 48 included 219 patients and showed no significant difference in failure to bleeding, rebleeding, number of blood transfusions, mortality, and AEs (Tables 6 and 7). Eight patients in the sclerotherapy group experienced SAEs (3 with aspiration pneumonia, 4 with a bleeding esophageal ulcer, and one with sepsis from pleural empyema resulting in death) compared with 4 in the terlipressin group (2 with lower limb ischemia, one with severe hyponatremia, and one with seizures) (AR, 0.03; CI, 0.03 to 0.09). Sclerotherapy Compared With Somatostatin A total of 406 patients were included in the 5 RCTs comparing sclerotherapy with somatostatin. 29,49 52 No significant differences were found in failure to bleeding as well as The number of transfused blood units was similar in the treatment groups in all of the studies. The weighted mean difference was assessable only in 2 studies, 29,51 which were heterogeneous in this respect ( 2 for heterogeneity, 3.73; df 1; P 0.054; weighted mean difference, 0.48; CI, 1.8 to 2.7) (Table 6). The mortality rate did not differ significantly between the 2 interventions. AEs and SAEs were significantly more frequent with sclerotherapy, with an AR of 0.14 (CI, ; number needed to treat for harm, 7; CI, 4 14) and 0.07 (CI, ; number needed to treat for harm, 14; CI, 8 100), respectively. SAEs were reported in 12 of 155 patients treated with sclerotherapy and included 4 deaths (2 from esophageal perforation, one from sepsis, and one from pleural empyema), 5 with clinically significant bleeding from esophageal ulcers, one with a cerebrovascular accident and pulmonary embolism, one with a pulmonary embolism, and one with bacterial peritonitis. No significant heterogeneity (P 0.10) was found in any of the pooled estimates other than the weighted mean difference for the number of transfused blood units. By excluding the study by Escorsell et al., 29 which included patients only after a 24-hour bleeding-free period, the risk difference for overall mortality was 0.06 (CI, 0.17 to 0.05).

9 May 2003 SCLEROTHERAPY VS. RUGS FOR VARICEAL BLEEING 1285 Table 6. Results of Each Study, Pooled Estimates, and Sensitivity Analyses for Failure to Control Bleeding, Rebleeding, and Transfusion Rebleeding Failure to bleeding Before other treatments Within 42 days a No. of blood units transfused RCT n/n AR (CI) n/n AR (CI) n/n AR (CI) Mean S WM (CI) Sclerotherapy vs. vasopressin Westaby et al. 47 4/33 11/ ( 0.44 to 0.03) Sclerotherapy vs. terlipressin Escorsell et al /114 20/ ( 0.12 to 0.09) Sclerotherapy vs. somatostatin i Febo et al. 49b 2/24 5/ ( 0.34 to 0.07) Shields et al. 50 1/41 1/39 0 ( 0.07 to 0.07) Planas et al. 51 6/35 7/ ( 0.21 to 0.15) 10/33 8/ ( 0.17 to 0.26) 16/114 15/105 0 ( 0.09 to 0.09) 1/24 5/ ( 0.36 to 0.01) 6/41 8/ ( 0.23 to 0.11) 5/35 7/ ( 0.23 to 0.12) Escorsell et al /79 14/90 0 ( 0.11 to 0.11) Ramires et al. 52 5/19 5/ ( 0.24 to 0.29) Subtotal 14/119 18/118 ( 0.07 to 0.05) Sclerotherapy vs. octreotide Sung et al. 53 5/49 8/ ( 0.19 to 0.07) Poo et al. 54b 1/21 2/ ( 0.19 to 0.11) Jenkins et al /77 13/ ( 0.05 to 0.21) Lopez et al. 56b 2/37 6/ ( 0.21 to 0.13) Bildozola et al. 57 4/33 5/ ( 0.23 to 0.03) Sivri et al. 58 9/36 8/ ( 0.23 to 0.20) Freitas et al /53 12/ ( 0.13 to 0.17) Yousuf et al. 60 2/48 4/ ( 0.14 to 0.05) Subtotal 55/354 58/ ( 0.08 to 0.02) 29/114 26/ ( 0.11 to 0.12) 1/24 5/ ( 0.36 to 0.01) 6/41 8/ ( 0.23 to 0.11) 5/35 7/ ( 0.23 to 0.12) 24/79 27/90 0 ( 0.13 to 0.14) 5/19 4/21 7 ( 0.19 to 0.33) 24/179 34/ ( 0.13 to 0.02) 41/198 51/ ( 0.13 to 0.03) ( 1.01 to 1.41) ( 0.5 to 0.3) ( 0.4 to 4.4) 114 patients 125 patients 0.48 ( 1.8 to 2.7) c 8/49 7/ ( 0.12 to 0.16) 8/49 7/ ( 0.12 to 0.16) 3/21 2/ ( 0.14 to 0.24) ( 0.6 to 1.6) 14/77 11/ /77 11/ ( 0.09 to 0.15) ( 0.09 to 0.15) ( 0.9 to 2.1) 8/37 11/ ( 0.26 to 0.13) 8/37 11/ ( 0.26 to 0.13) 6/36 5/30 0 6/36 5/30 0 ( 0.18 to 0.18) ( 0.18 to 0.18) 15/53 16/ ( 0.16 to 0.17) 2/48 4/ /48 4/ ( 0.14 to 0.05) ( 0.14 to 0.05) 36/247 34/ ( 0.07 to 0.05) 56/321 56/ ( 0.06 to 0.05) ( 0.8 to 0.2) ( 1.8 to 0.6) 171 patients 164 patients 0.12 ( 0.58 to 0.35) All of the above trials irrespective of the pharmacologic treatment Random effects 93/ / /540 87/ / / patients 394 patients 0.2 (14 RCTs) ( 0.06 to 0.01) (10 RCTs) ( 0.06 to 0.02) (14 RCTs) ( 0.05 to 0.03) (7 RCTs) ( 0.5 to 0.2) Fixed effects 93/ / /540 87/ / / patients 394 patients 0.4 (14 RCTs) ( 0.07 to 0.01) (10 RCTs) ( 0.06 to 0.02) (14 RCTs) ( 0.05 to 0.04) (7 RCTs) ( 0.5 to 0.03) Full reports 86/542 95/ /516 82/ / / patients 372 patients 0.3 (11 RCTs) ( 0.06 to 0.01) (9 RCTs) ( 0.06 to 0.03) (12 RCTs) ( 0.05 to 0.04) (6 RCTs) ( 0.6 to 0.2) Abstracts 7/78 12/ /24 5/ /45 7/ patients 22 patients 0.5 (3 RCTs) ( 0.16 to 0.03) (1 RCT) ( 0.36 to 0.01) (2 RCTs) ( 0.28 to 0.16) (1 RCT) ( 0.6 to 1.6) Adequate generation of randomization list and allocation concealment Inadequate or unclear generation of randomization list and allocation concealment Intention-to-treat analysis Intention-to-treat analysis not reported or unclear 40/214 43/ /313 48/ /313 72/309 0 (4 RCTs) ( 0.10 to 0.04) (5 RCTs) ( 0.06 to 0.06) (5 RCTs) ( 0.07 to 0.06) (4 RCTs) 46/308 53/ /203 34/ /308 62/309 (7 RCTs) ( 0.08 to 0.03) (4 RCTs) ( 0.09 to 0.04) (7 RCTs) ( 0.07 to 0.04) (2 RCTs) 264 patients 260 patients 0.2 ( 0.8 to 0.4) 110 patients 112 patients 0.15 ( 0.8 to 0.5) 38/237 33/ /316 49/ /316 62/ patients 228 patients 0.1 (5 RCTs) ( 0.05 to 0.05) (6 RCTs) ( 0.07 to 0.03) (6 RCTs) ( 0.07 to 0.04) (4 RCTs) ( 0.8 to 0.6) 48/305 62/ /200 33/193 75/306 72/ patients 144 patients 0.2 (6 RCTs) ( 0.11 to 0.008) (3 RCTs) ( 0.08 to 0.07) (6 RCTs) ( 0.06 to 0.08) (2 RCTs) ( 0.7 to 0.3) NOTE. No statistically significant heterogeneity was found if not otherwise specified., endoscopic variceal sclerotherapy;, vasoactive drug; n/n, number of patients with the outcome/total number of patients; WM, weighted mean difference. a When a study reported rebleeding at a different time, the figure at the longest available time interval, up to 42 days, was used. b Abstract. c 2 for heterogeneity 3.73, df 1, P

10 1286 AMICO ET AL. GASTROENTEROLOGY Vol. 124,No. 5 Table 7. Results of Each Study, Pooled Estimates, and Sensitivity Analyses for Mortality and Adverse Events Mortality AEs Before other treatments Within 42 days a Total Serious RCT n/n AR (CI) n/n AR (CI) n/n AR (CI) n/n AR (CI) Sclerotherapy vs. vasopressin Westaby et al. 47 9/33 12/31 Sclerotherapy vs. terlipressin Escorsell et al /114 26/105 Sclerotherapy vs. somatostatin i Febo et al. 49b 5/24 6/ Shields et al. 50 3/41 5/39 ( 0.19 to 0.08) 8/41 12/39 0 Planas et al. 51 4/35 4/35 ( 0.15 to 0.15) 8/35 10/35 Escorsell et al /79 13/90 Ramires et al. 52 6/19 6/ Subtotal 7/76 9/74 ( 0.13 to 0.07) 39/198 47/208 Sclerotherapy vs. octreotide 0.02 Sung et al. 53 4/49 3/49 ( 0.08 to 0.12) 20/49 14/49 Poo et al. 54b 5/21 3/ Jenkins et al. 55 9/77 8/73 ( 0.09 to 0.11) 13/77 22/73 Lopez et al. 56 7/33 6/ Bildozola et al. 57 3/37 8/39 ( 0.28 to 0.03) 3/37 8/39 Sivri et al. 58 1/36 1/30 0 Freitas et al. 59 2/53 2/58 ( 0.07 to 0.07) 8/53 13/58 0 Yousuf et al. 60 5/48 5/48 ( 0.12 to 0.12) 5/48 5/48 0 Subtotal 23/264 26/267 ( 0.05 to 0.05) 62/354 72/350 All of the above trials irrespective of the pharmacologic treatment Random effects 30/340 35/341 ( 0.05 to 0.03) 129/ / ( 0.34 to 0.11) 3/33 3/ ( 0.19 to 0.03) 34/114 21/105 ( 0.15 to 0.14) 1/33 0/ ( 0.02 to 0.21) 8/14 4/ ( 0.05 to 0.11) 0.03 ( 0.03 to 0.09) 0.05 ( 0.29 to 0.19) 2/24 1/ ( to 0.18) ( 0.30 to 0.08) 12/41 5/39 ( to 0.34) 1/41 0/39 ( 0.04 to 0.09) ( 0.26 to 0.15) 10/35 5/35 ( 0.05 to 0.33) 5/35 0/35 (0.02 to 0.27) ( 0.10 to 0.11) 19/79 4/90 (0.09 to 0.30) 6/79 0/90 (0.01 to 0.14) 0.03 ( 0.25 to 0.31) ( 0.10 to 0.05) 43/179 15/187 (0.07 to 0.22) 12/155 0/164 (0.01 to 0.13) 0.12 ( 0.06 to 0.31) 18/49 5/ (0.11 to 0.42) ( 0.13 to 0.23) 1/21 1/22 (0.12 to 0.13) ( 0.27 to 0) 15/77 19/73 ( 0.20 to 0.07) 0.12 ( 0.28 to 0.03) ( 0.18 to 0.22) 4/37 0/39 (0.00 to 0.17) 0.11 ( 0.09 to 0.08) 5/36 1/30 ( 0.02 to 0.24) 0.07 ( 0.22 to 0.07) ( 0.12 to 0.12) 6/48 8/48 ( 0.18 to 0.10) ( 0.08 to 0.03) 43/220 26/213 ( 0.03 to 0.15) c ( 0.07 to 0.008) 130/594 73/584 (7 RCTs) (15 RCTs) (12 RCTs) Fixed effects 30/340 35/341 ( 0.06 to 0.03) 129/ /694 ( 0.08 to 0.003) 129/594 73/584 (7 RCTs) (15 RCTs) (12 RCTs) 0.04 Full reports 30/340 35/341 ( 0.05 to 0.03) 112/ /618 ( 0.08 to 0.004) 127/549 71/539 (7 RCTs) (12 RCTs) (10 RCTs) 0.02 Abstracts 17/78 19/76 ( 0.10 to 0.15) 3/45 2/45 (3 RCTs) (2 RCTs) Adequate generation of randomization list and allocation 0.01 concealment 8/84 7/84 ( 0.07 to 0.09) 60/313 64/309 ( 0.07 to 0.04) 86/313 36/309 Inadequate or unclear generation of randomization list and allocation concealment 22/256 28/257 Intention-to-treat analysis 21/201 22/195 (2 RCTs) (5 RCTs) (5 RCTs) ( 0.06 to 0.03) 52/308 78/309 (5 RCTs) (7 RCTs) ( 0.07 to 0.05) 47/316 63/ (0.02 to 0.08) NNH, 20 (13 to 50) 0.06 (0.02 to 0.09) NNH, 17 (11 to 50) 0.05 (0.02 to 0.08) NNH, 20 (13 to 50) 0.08 (0.03 to 0.14) d 21/302 4/300 NNH, 13 (7 to 33) d (5 RCTs) 0.09 (0.05 to 0.13) d 21/302 4/300 NNH, 11 (8 to 20) (5 RCTs) 0.10 (0.04 to 0.16) e 21/302 4/300 NNH, 10 (6 to 25) (5 RCTs) 0.02 ( 0.07 to 0.11) 0.16 (0.10 to 0.21) 19/228 4/230 NNH, 6 (5 to 10) (3 RCTs) 0.08 ( 0.14 to 0.02) 41/236 35/ ( 0.05 to 0.12) 2/74 0/70 NNT, 13 (7 to 50) (5 RCTs) (2 RCTs) ( 0.08 to 0.02) 67/316 42/315 ( to 0.18) f 12/155 0/164 (4 RCTs) (6 RCTs) (6 RCTs) (3 RCTs) Intention-to-treat analysis not reported or unclear 9/139 13/146 ( 0.08 to 0.06) 65/305 79/303 ( 0.13 to 0.01) 60/233 29/224 (0.03 to 0.20) 9/147 4/136 NNH, 8 (3 RCTs) (6 RCTs) (4 RCTs) (5 to 33) (2 RCTs) 0.07 (0.02 to 0.12) NNH, 14 (8 to 100) 0.03 ( 0.02 to 0.08) 0.07 (0.01 to 0.13) NNH, 14 (8 to 100) 0.03 ( 0.02 to 0.08) NOTE. No statistically significant heterogeneity was found if not otherwise specified., endoscopic variceal sclerotherapy;, vasoactive drug; n/n, number of patients with the outcome/total number of patients; NNH, number needed to treat for harm (CI); NNT, number needed to treat for benefit (CI). a When a study reported mortality at a different time, the figure at the longest available time interval, up to 42 days, was used. b Abstract. c 2 for heterogeneity 13.81, df 5, P d 2 for heterogeneity 21.53, df 11, P e 2 for heterogeneity 18.89, df 9, P f 2 for heterogeneity 13.74, df 5, P

11 May 2003 SCLEROTHERAPY VS. RUGS FOR VARICEAL BLEEING 1287 Sclerotherapy Compared With Octreotide A total of 704 patients were included in the 8 RCTs comparing sclerotherapy with octreotide There were no differences between the 2 treatments for failure to bleeding as well as rebleeding (Table 6). Transfusion requirement was not reported by Lopez et al. 56 and Yousuf et al. 60 ; no significant differences were found by Sung et al. 53 and Freitas et al., 59 who did not report means with standard deviations. The mean weighted difference of the number of transfused blood units, estimable in 4 trials 54,55,58,59 including 335 patients, was not significant ( 0.12; CI, 0.58 to 0.35). No significant differences were found in mortality (Table 7). AEs were reported in 7 trials (Table 7). Lopez et al. 56 reported thoracic pain in 9 patients and esophageal ulcers in 12 in the sclerotherapy group, whereas only transient hyperglycemia was observed in the octreotide group (number of patients not reported). In the other trials, all of the observed AEs were reported without qualification of their severity. When the 6 trials reporting complete data are combined, the pooled risk difference was 0.06 (CI, 0.03 to 0.15) with significant heterogeneity ( 2, 13.81; df 5; P 0.017). Sensitivity Analyses by Combining All of the Studies Irrespective of the Control Pharmacologic Therapy The analyses according to the random versus fixed effects models and according to the type of publication confirmed that sclerotherapy and medical interventions were not significantly different for failure to bleeding, rebleeding, and mortality (either before or after starting elective treatments). However, a type II error may not be ruled out for overall mortality, which was nearly significantly reduced by sclerotherapy (AR, 0.035; CI, 0.07 to 0.008) (Table 7). Only the number of transfused blood units was marginally but significantly reduced by sclerotherapy by the fixed effects model (AR, 0.4; CI, 0.5 to 0.3); the difference between the 2 models was mostly explained by the study by Planas et al., 51 with a relative weight of 46% in the random effects model and 95% in the fixed effects model ( 2 for heterogeneity, 9.23; df 6; P 0.16). Sensitivity Analyses According to the Quality of RCTs Irrespective of the Control Pharmacologic Therapy In general, RCTs of low quality exaggerated benefits from sclerotherapy (Tables 6 and 7). Failure to bleeding was significantly reduced in RCTs without and equivalent in those with intention-to-treat analysis. Overall mortality was significantly reduced in RCTs with generation of the randomization list and allocation concealment not adequate or unclear, whereas it was equivalent in those with adequate methods (Figure 2). AEs were not significantly different in RCTs with generation of the randomization list and allocation concealment not adequate or unclear and significantly increased with sclerotherapy in those with adequate methods (Figure 2). Publication Bias Egger and Begg tests for publication bias showed that the risk of having missed trials was acceptably low; for failure to bleeding, rebleeding, mortality, and AEs, P values ranged from 0.23 to 0.95 by the Egger s test and from 0.59 to 1.00 by the Begg s test. The Egger s publication bias plot for the risk difference for 42-day mortality is reported in Figure 3. iscussion This systematic review showed that emergency sclerotherapy is not significantly superior to treatment with terlipressin, somatostatin, or octreotide with regard to any of the assessed efficacy outcomes and may even carry a higher burden of AEs. It was superior to vasopressin plus nitroglycerin for the of bleeding but not for the other outcomes in one trial 47 possibly flawed by a detection bias. There were appreciable differences across trials in the definitions of the major outcomes, particularly for failure to bleeding. In several studies, the 42-day or in-hospital rebleeding or mortality were reported as major treatment outcome measures, even when elective treatments for the prevention of rebleeding were used between the end of the experimental treatment and the time when the outcome was assessed. To overcome this problem, we also abstracted the rebleeding and mortality rates before elective treatments were given. Criteria for blood transfusions were not a priori stated in almost all trials. Therefore, it is difficult to draw conclusions using this outcome measure, considering the many variables included in the decision to give a transfusion and unblindness of these studies. In addition, criteria to relate AEs to the treatment were never defined a priori and only a few trials reported a distinction between nonserious AEs or SAEs. However, it is of interest that 6 deaths were attributed to sclerotherapy compared with none in the vasoactive groups. The methodological quality was satisfactory only in 3 trials 29,51,58 fulfilling all of the quality criteria we con-

12 1288 AMICO ET AL. GASTROENTEROLOGY Vol. 124,No. 5 Figure 2. Forty-two-day mortality (top) and AEs (bottom) irrespective of the vasoactive treatment in full reports, according to whether adequate generation of the randomization list and adequate treatment allocation concealment were used or not. RCTs are identified by the first author and the year of publication. R, risk difference;, endoscopic variceal sclerotherapy. sidered except blinding. Although a double-blind design is almost impossible for RCTs of endoscopic variceal sclerotherapy, a blind assessment of some important outcomes such as of bleeding, rebleeding, need for transfusions, and AEs, all of which bear a potential for subjective judgment, might have resulted in more stable estimates. 13 Major methodological quality problems were the lack of adequate generation of the randomization list, treatment allocation concealment, and lack of intention-to-treat analysis in several studies (Table 5). It is of interest that RCTs lacking these quality criteria exaggerated the beneficial effects of sclerotherapy, particularly on survival, and minimized AEs, whereas the opposite was found in good-quality studies. Only the trial comparing sclerotherapy with vasopressin showed a significant reduction of failure to bleeding with sclerotherapy. 47 In this trial, however, all of the patients in the vasopressin group underwent sclerotherapy at the end of the 12-hour study period. Thus, endoscopy provided a more precise method of evaluating the persistency of bleeding in this group compared with the sclerotherapy group. Although it is difficult to evaluate how this potential bias actually affected the study conclusions, it is of interest that this was the only one of the 15 included studies showing superiority of sclerotherapy in ling bleeding. No other significant differences were found between emergency sclerotherapy and each of the vasoactive drugs. The sensitivity analyses performed, including all of the studies irrespective of the treatment, confirmed that emergency sclerotherapy is not superior to vasoactive drugs for any of the assessed outcomes, although a type II error may not be ruled out for overall mortality with an AR of (CI, 0.07 to 0.008) (random effects); more than 4000 patients would be needed to show this difference to be statistically significant. However, it is of interest that when the studies available as full reports were combined according to whether the generation of the randomization list and allocation concealment were adequate, the difference was virtually abolished (AR, ; CI, 0.07 to 0.04) in the 5 studies (622 patients) with adequate methods and was remarkably increased and statistically significant in the 7 studies (617 patients) with nonadequate methods (AR, 0.08; CI, 0.14 to 0.02). These findings strongly suggest that the nonsignificant trend toward a death risk reduction with sclerotherapy found by combining all of the 15 studies is more likely a distorted estimate resulting from some potential bias in several low-quality studies rather than a falsely negative result due to insufficient power. Similar trends toward benefit from sclerotherapy were found in most of the sensitivity analyses according to the methodological criteria; almost Figure 3. Egger s publication bias plot. The standardized risk difference ( minus pharmacologic interventions) for 42-day mortality is plotted against the precision of the risk difference along with the regression line and the CI (arrows) about the intercept. Failure of this CI to include 0 would indicate asymmetry in the funnel plot and might give evidence of publication bias. Standardized risk difference risk difference/standard error. Precision of risk difference 1/standard error.