Somatostatin v placebo in bleeding oesophageal varices: randomised trial and meta-analysis

PATIENT SELECTION

The patients were enrolled from April 1987 to the end of April 1992. Patients with clinical indication of bleeding oesophageal varices and with verified or suspected cirrhosis of the liver were eligible for the study. Children and pregnant or lactating women were excluded. Informed consent was obtained unless the patient's condition was so poor that it was not possible. The study was approved by the regional ethics committees and by the National Board of Health.

TREATMENT PROTOCOL

Vials containing either 3 mg somatostatin or placebo of identical appearance were delivered by DuraScan Medical Products. The contents of two vials were dissolved in 1 litre of isotonic glucose and infused at a rate of 42 ml/h, corresponding to 250 μg/h of somatostatin. Treatment was given for 24 hours for each episode of bleeding. Bleeding after at least 6 hours with clear aspirate from the nasogastric tube was regarded as a new episode.

Randomisation in blocks of 10 patients was made by DuraScan according to a table of random numbers. Sealed envelopes were available at the departments; they were not opened on any occasion. The patients were allocated as quickly as possible after the start of bleeding to somatostatin or placebo by using the medicine package with the lowest available number.

The trial was pragmatic as all other treatment followed the usual routines. Ancillary treatments were similar at the participating departments except that antiulcer drugs were more commonly used at the medical than at the surgical department. We originally planned to exclude patients in whom endoscopy after inclusion in the study did not support varices as the cause of bleeding or in whom later investigations could not verify cirrhosis of the liver. However, as such decisions could bias the trial we decided before data analysis began to keep all randomised patients in the analysis.

At entry we noted duration of bleeding; number of transfusions before randomisation; presence of encephalopathy and ascites; plasma concentrations of bilirubin, aspartate amino transferase, alkaline phosphatase, albumin, and prothrombin; and presence of diabetes.

Outcome measures after six weeks were survival, blood transfusions, episodes of bleeding, days with bleeding, use of the Sengstaken-Blakemore tube, and complications. Only observed or spontaneously reported adverse reactions were noted.

STATISTICAL METHODS

We planned to include 100 patients, assuming arrest of bleeding in 40% with placebo, a minimally relevant difference of 30%, and type I and type II errors of 0.05 and 0.10, respectively. No interim analyses were made. Fisher's exact test was used for binary data and the Mann-Whitney test for other variables. Analysis of data and writing of manuscripts was blinded; the code was not broken before both versions of the manuscript had been agreed on by all authors. We calculated 95% confidence intervals for main outcome variables.

META-ANALYSIS

The primary aim of the meta-analysis was to study whether somatostatin or its derivative octreotide improve survival. Secondary variables were number of blood transfusions and number of patients needing balloon tamponade.

All randomised trials comparing one of the drugs with placebo or no treatment in patients suspected of bleeding from oesophageal varices were eligible. An online Medline search from 1966 onwards was done in January 1995. The term “explode somatostatin/all subheadings” or the text words somatostatin or octreotide were combined with any of the following: variceal, varices, bleed#, haemorrhage, hemorrhage, oesophag#, esophag#, haematemesis, hematemesis, melaena, melena.

Furthermore, the company marketing octreotide was contacted, reference lists of relevant articles, reviews, and editorials were scanned, and authors of trials were asked whether they knew of additional studies, including unpublished ones.

Details on numbers of randomised patients, deaths, randomisation and blinding procedures, numbers of drop outs, and exclusions after randomisation were extracted. Additional information was sought by writing to the authors. The outcomes were weighted by the inverse variance10 11; for blood transfusions we used medians when available.

PRESENT TRIAL

Because of slow recruitment and as somatostatin was no longer commercially available the trial was stopped after five years, when 86 of the planned 100 patients had been entered. The two treatment groups were comparable at baseline (table I). Alcoholic liver disease was suspected or verified in almost all patients; in only six were other diagnoses definitely established. One patient on somatostatin suffered from Budd-Chiari's syndrome because of polycythaemia, another (who died during the study) from haemangioendothelioma. Two patients on placebo had oesophageal ulcers, one had cirrhosis because of reactive hepatitis, and one had liver metastases from a rectal carcinoma (died during study).

Endoscopy was performed within 24 hours after randomisation in 44 of the patients but could not always be performed during active bleeding. Thus, not all patients had visible blood, and in 13 of the 81 patients who ultimately underwent endoscopy varices were not seen. Sclerotherapy was used in 29 patients treated with somatostatin and in 31 patients treated with placebo. Antiulcer drugs were used in 10 patients on somatostatin and in 12 on placebo; tranexamic acid was used in two v one. Operations were performed in two v one; finally, vasopressin and propranolol were used in one patient each on placebo.

Sixteen patients died in each group (P=1.00). The number of days of bleeding and numbers of patients who needed balloon tamponade were also similar (table II). More patients in the somatostatin group had more than one episode of bleeding (P=0.10), and they needed more blood transfusions, eight v five (P=0.07).

In the somatostatin group, the oesophagus ruptured in two patients in whom balloon tamponade was used; one of them died. A third patient developed stenosis after sclerotherapy. No adverse effects were reported with placebo.

META-ANALYSIS

We identified only two trials in addition to our own with somatostatin and none with octreotide. In the trial by Valenzuela et al 18 of 102 randomised patients were excluded for various reasons.8 In the trial by Burroughs et al 13 of 133 randomised patients were excluded because the source of bleeding was judged to be non-variceal.9 Randomisation more than once was allowed; thus the remaining 120 admissions referred to only 92 patients. We would have preferred to include all randomised patients in the meta-analysis and to exclude any doubles. The authors were not, however, able to give us information on these patients.

The meta-analysis comprised 290 patients. Somatostatin had no effect on survival (table III); a total of 40 patients died in the somatostatin groups v 33 patients in the placebo groups (P=0.59; odds ratio 1.16; 95% confidence interval 0.67 to 2.01). There was no heterogeneity between the trials (P=0.97).

There was considerable heterogeneity for average number of blood transfusions (P=0.02). In the trial by Valenzuela et al patients on somatostatin received on average 1.1 litres of blood products while patients on placebo received 1.4 litres. This difference corresponds to one unit of blood (variance of the difference=3.46). In the trial by Burroughs et al the median number of units of blood or plasma transfused was 3 units with somatostatin and 6 units with placebo (variance=1.07). We found the reverse: 6 units of blood transfused with somatostatin and 3 units with placebo (variance=3.85). It would be problematic to take this heterogeneity into account by doing a random effects analysis as the variation between studies would be determined with great uncertainty, based on only three trials.11 If the heterogeneity is ignored patients in the somatostatin groups received on average 1.6 transfusions fewer than in the placebo groups (P=0.06; -3.2 to +0.1). (In a random effects analysis the confidence interval would have been wider). As the variance in the trial by Burroughs et al9 was surprisingly small, however, being only one third of the roughly identical variance in the other two trials, weighting by inverse variance seems to be unjustified. If this is accepted, then a simple, unweighted average of only a third of a unit of blood products is the best estimate of the effect of somatostatin over placebo.

Valenzuela et al did not report data on the use of balloon tamponade.8 There was heterogeneity for the other two trials with opposite trends (P=0.05). Balloon tamponade was used in seven patients on somatostatin and in 15 patients on placebo in the trial by Burroughs et al9 and in 16 v 13 patients in our trial.