Robert Burns

This issue presents the initial reports of two randomized, controlled trials that compared endoscopic sclerotherapy with transjugular intrahepatic portosystemic shunts (TIPS) in the treatment of patients who survived hemorrhage from esophageal varices [1, 2]. Only one previous such study has been published as a peer-reviewed article [3]. Although the goals and the experimental designs of these trials are similar, they are not identical. Furthermore, the results of the three studies are quite different.

In all three studies, endoscopic sclerotherapy, which is widely held to be the standard therapy for bleeding varices, can be considered to be the control treatment and TIPS to be the investigational treatment. All three studies demonstrate that both therapies can be instituted promptly and effectively. However, several inherent features of these two procedures prevent the performance of an ideal randomized trial. First, the two therapies are intrinsically so different that a truly double-blind randomized trial is virtually impossible. Bias and perceptions of bias cannot be completely eliminated in unblinded investigations. Second, in the temporal sense, the two procedures cannot be exactly compared. An experienced radiologist can create TIPS in an hour or so, after which the portosystemic pressure gradient is reduced by half. Endoscopic sclerotherapy for varices, however, is a much more deliberate, labor-intensive procedure. In Sanyal and colleagues' study [1], sclerotherapy was performed in 6 to 10 sessions during a period of 3 to 4 months. Thus, the sclerotherapy group endured an appreciably longer period at risk for portal hypertension than did the TIPS group. It is unlikely that the two treatments can be initiated with the same promptness in all patients, and such differences must be taken into account when the results of these studies are analyzed. Third, patients with TIPS frequently experience shunt stenosis caused by ongoing endothelial hyperplasia of unknown cause, which may lead to shunt malfunction in more than half of the patients over a period of 6 to 8 months [4]. Endothelialization of the stent, which occurs uniformly and which is the essence of the TIPS procedure, creates a protective endothelial coating of the naked steel stent that converts it into a vascular channel. This endothelial hyperplasia frequently continues and progresses until it compromises the patency of the shunt.

In analyses of such studies, these possible discrepancies can be minimized by the use of intention-to-treat principles, the selection of appropriate outcomes, and the measurement of these outcomes at appropriate times.

The three studies are similar in size, ranging from 49 to 80 patients; about half of these patients received each treatment. In the studies by Cello and Cabrera and their colleagues [2, 3], variceal hemorrhage recurred significantly less frequently in the TIPS group than in the sclerotherapy group in terms of percentage of patients afflicted (<25% compared with approximately 50%), total number of bleeding episodes, and number of transfusions required (Table 1). In Sanyal and colleagues' study [1], hemorrhage occurred in about one fourth of patients in both groups (that is, the frequency of bleeding was not reduced by TIPS). On the other hand, the mortality rates in the sclerotherapy and TIPS groups did not statistically significantly differ in any of the three randomized trials; in all the studies, however, mortality rates were higher in the TIPS groups than in the sclerotherapy groups. The duration of hospitalization tended to be similar after both treatments, as did the costs in the one randomized trial in which cost was calculated [2].

One aspect of the results showed relative homogeneity. The incidence of portosystemic encephalopathy was higher after TIPS than after endoscopic sclerotherapy in all three studies and was statistically significantly higher in two of them [1, 3]. Portosystemic encephalopathy [5] and stent stenosis [6] are the two most common complications of TIPS. Clearly, hepatic encephalopathy is a consequence of portosystemic shunting; creation of TIPS increases the amount of shunting and, proportionately, the frequency and severity of portosystemic encephalopathy [5].

It is logical to conclude that the differences in results among these three studies reflect differences in patient age and the cause and severity of cirrhosis, as well as differences in the effects of TIPS and sclerotherapy themselves. Established indices of the prognosis of cirrhosis in the three groups of patients studied at the time of randomization are shown in Table 2. Many physicians would expect that a meta-analysis in which the results of these three studies relative to the prognostic variables would resolve these issues.

The overall mortality rates in these three trials were similar and could not be attributed to any one prognostic factor. Paradoxical results were sometimes found. Surprisingly, for example, the study [3] with the highest mean age (56 years) reported the lowest overall mortality rate (18%). Recurrent variceal hemorrhages tended to occur more frequently in the studies in which cirrhosis was predominantly alcoholic in origin than in the other investigations [2, 3] (Table 1 and Table 2). The randomized trial [1] in which the patients had the most severe cirrhosis, as determined by Child criteria (51% had class C cirrhosis), had the highest overall mortality (24%); this rate, however, was not statistically significantly higher than the rates in the trials in which cirrhosis was less severe. Unfortunately, the studies did not present data on Child criteria, which were quite heterogeneous, in the same manner (Table 2); it was neither realistic nor appropriate to obtain the missing data for the purposes of editorial commentary.

In theory, one way that the questions raised about the efficacy of these therapies can be answered would be to perform very large randomized clinical trials, similar to the three small ones discussed here. To resolve the issue of prevention of hemorrhage from varices and improvement in survival without an unacceptably high frequency of side effects would, according to my estimate, require a randomized, controlled trial involving more than 1000 patients. At best, studies of this size are very difficult and prohibitively expensive to perform. No single hospital could ever accomplish such a trial, and organizing groups of collaborating hospitals to do so is extremely difficult. Even when such studies are undertaken, there is no guarantee of success [7]. What is clear, however, is that one can greatly reduce the portal venous pressure and the risks for variceal bleeding by using a larger stent; however, the larger the stent is, the higher the incidence of encephalopathy [5]. Conceivably, such a huge randomized study might require prophylactic TIPS, that is, creation of TIPS before the initial hemorrhage. Few investigators in the world, however, have the audacity to attempt such studies, and even fewer human investigating committees would authorize them. Furthermore, data indicate that ß-adrenergic blocker drugs can reduce the risks for recurrent bleeding and virtually eliminate the risk for initial bleeding [8]. In the meantime, it is essential that more trials be performed to provide additional data for such meta-analyses. Fortunately, as noted, several such studies are in progress.

The only viable alternative, then, for answering these questions at this time is by meta-analyses that combine all available data. Such analyses must eventually be performed. Findings from only three trials comparing endoscopic sclerotherapy with TIPS have been reported in peer-reviewed journals [1-3], and the heterogeneity of the clinical material, the experimental designs, and the results is so great that firm conclusions cannot be drawn in the absence of such a formal synthesis. Indeed, heterogeneity is the Achilles' heel of meta-analysis [9]. Results of four other randomized trials that compared endoscopic sclerotherapy with TIPS and one trial that studied endoscopic variceal ligation have been reported in preliminary form, mostly as abstracts (Grace ND. Unpublished data). It is my opinion that non-peer-reviewed articles, such as abstracts, theses, chapters in books, and journal supplements, are often incomplete, unvalidated, or erroneous ([7, 10]; McGinnis J. Unpublished data) and should not be included in meta-analyses. The advantages of including all data retrieved from such sources are outweighed by the inadvertent introduction of misleading information, unless all data and those data that conform to the recommended guidelines for analysis [10] are analyzed and presented separately. Even peer-reviewed articles published in premier medical journals require that authors of meta-analyses directly request detailed, supplementary data from the investigators to avoid inaccuracy and ambiguity [11, 12]. The performance of meta-analyses based on data from individual patients (such as the analysis by Poynard and colleagues [13]), which are considered to be less biased than meta-analyses based on data summarized by the investigators, require close cooperation and the exchange of information between the meta-analysts and the original investigators.

In the midst of all these diverse data on endoscopic, angiographic, surgical, and pharmacologic therapy, the question remains: What should be done now for a patient who needs treatment? I believe that an individualized, rational, effective therapeutic program should be devised for each patient who bleeds from varices. In addition, I believe that the patient should participate in these therapeutic decisions whenever possible. If, for example, I had bled from esophageal varices, I would like to be admitted to an academic hospital at which a clinical trial comparing TIPS with endoscopic sclerotherapy was in progress under the direction of an experienced "TIPSologist," ideally one of the coauthors of my book on TIPS [14]. I hope that I could satisfy the criteria for inclusion because the prognosis of patients who do so seems to be better than that of those who do not [15]. I would give permission to be randomly assigned to help pay my debt to society as a clinical investigator. I would fervently pray not to be randomly selected to be included in the endoscopic sclerotherapy group because the idea of having a long series of endoscopic ligations or injections of any of the nasty sclerosants available is intrinsically repugnant to me. If selected to undergo TIPS creation, I would request that a stent 10 mm in diameter be implanted and expanded to its full diameter by inflating a balloon within it. It is my opinion that a stent with a diameter of 10 to 10.5 mm is ideal for a person of my size and anatomy. It is large enough to reduce the portal venous pressure gradient to a level below the bleeding threshold (12 mm Hg) but small enough to maintain portosystemic shunting at a level below the threshold of encephalopathy. After all, I hope to continue writing editorials about portal hypertension for years to come. If the randomization scheme had gone agley and I was selected to have endoscopic therapy, I would refuse it and would request that one of my pharmacologically oriented colleagues devise a pharmacologic program using a ß-adrenergic blocker and a nitrate, which is a very effective therapeutic combination [16].

Obviously, the definitive answers are still unknown and will not be known for the foreseeable future. In the meantime, it behooves physicians to know the literature intimately and to continue to accumulate objective data to improve these answers and to permit future investigators to achieve more precise treatment.

The results of this trial do not appreciably differ from those of the three studies discussed in this editorial, but differences in the therapies administered and the less severe cirrhosis in Rossle and colleagues' trial make further comparisons difficult.