Decisions, decisions, decisions

Decisions, decisions, decisions

GASTROENTEROLOGY 2000;118:1268–1274 SELECTED SUMMARIES Henry J. Binder, M.D. Selected Summaries Editor Yale University School of Medicine New Haven, ...

61KB Sizes 0 Downloads 152 Views

GASTROENTEROLOGY 2000;118:1268–1274

SELECTED SUMMARIES Henry J. Binder, M.D. Selected Summaries Editor Yale University School of Medicine New Haven, Connecticut 06520-8019

STAFF OF CONTRIBUTORS

Laurence Blendis, Toronto, Canada Robert Bresalier, Detroit, MI Guadalupe Garcı´a-Tsao, New Haven, CT William L. Hasler, Ann Arbor, MI Cyrus Kapadia, New Haven, CT

Ronald L. Koretz, Sylmar, CA Kris V. Kowdley, Seattle, WA Wayne Lencer, Boston, MA James D. Lewis, Philadelphia, PA Gary R. Lichtenstein, Philadelphia, PA

DECISIONS, DECISIONS, DECISIONS Younossi ZM, Singer ME, McHutchison JG, Shermock KM (Department of Gastroenterology and the I. H. Page Center for Health Outcomes Research, Cleveland Clinic Foundation, Cleveland, Ohio; Department of Epidemiology and Biostatistics, Institute for Public Health Sciences, Case Western Reserve University, Cleveland, Ohio; and Division of Gastroenterology, Scripps Clinic and Research Foundation, La Jolla, California). Costeffective ness of interferon a2b combined with ribavirin for the treatment of chronic hepatitis C. Hepatology 1999;30:1318– 1324. Younossi et al. expressed concern over the fact that there are nearly 4 million carriers of hepatitis C virus (HCV) in the United States and that 25%–30% of patients with chronic hepatitis C will progress to end-stage liver disease resulting in death or liver transplantation. They then noted that antiviral treatment can produce sustained viral responses, with rates as high as 30% for combination (interferon/ribavirin) therapy. With this in mind, they embarked on a decision analysis to compare various therapeutic regimens to no treatment. The anticipated costs and benefits were estimated from data available in the literature. The investigators stated that well-designed prospective studies of patients with chronic hepatitis C were lacking. Because of this belief, they modeled the natural history of the disease from data obtained from observations of patients followed up in tertiary referral centers. To correct the data for deaths from other causes, the age-specific death rates in the general population were factored into the model. The therapeutic responses were derived from trials in the literature; for combination therapy, a baseline sustained response rate of 31% was used. Cost data were obtained from wholesale price listings and Medicare fee schedules. The investigators also listed the following ‘‘important assumptions’’: (1) progression to hepatocellular carcinoma (HCC) only occurs after development of cirrhosis; (2) the annual incidence of HCC was the same in compensated and decompensated cirrhosis; (3) no progression to cirrhosis occurs during or after therapy in patients who respond; (4) a sustained viral response was considered to be a cure; (5) liver transplantation only is done for individuals with decompensated cirrhosis;

William Sandborn, Rochester, MN Mitchell L. Schubert, Richmond, VA Fergus Shanahan, Wilton, Cork, Ireland David C. Whitcomb, Pittsburgh, PA

(6) patients with compensated cirrhosis develop decompensation or HCC before experiencing a liver-related death; (7) all treatment responses, or lack thereof, were based on virological response; (8) side effects of the different treatments were similar, and the untreated group was spared any side effects; and (9) there is no spontaneous clearance of HCV. When the model was run on a computer, each of the therapeutic interventions resulted in costs of $35,000–40,000 and resulted in the gain of 14 to almost 16 quality-adjusted life years. The most cost-effective strategy was to begin with combination therapy, with the duration of therapy being guided by the genotype of the virus infecting any particular patient. (According to this plan, the patients were given 48 weeks of treatment if they were infected with genotype I, and only 24 weeks of such therapy if infected by any other genotype.) The cost for each additional quality-adjusted life year in this latter scenario was $7500, an amount that compared favorably with what society is willing to spend for other health care interventions. The investigators believed that all of the treatment strategies were advantageous compared with not providing any therapy. In fact, they stated that combination therapy was more cost-effective than no treatment at all. The sensitivity analysis indicated that the model was robust for all the parameters that were changed. The investigators stated that the model was valid because it predicted the 3-, 5-, and 10-year mortalities after development of cirrhosis, using data from a tertiary referral center as the gold standard. They concluded that combination therapy, adjusting the duration of treatment by the viral genotype, was the most cost-effective strategy. Comment. There are a number of ways whereby health care professionals make decisions regarding treatment. In an ideal universe, all knowledge would be available, including data from well-done prospective, randomized-controlled trials. Any given patient could then be assessed, and, based on the evidence, a decision made regarding how to best deal with his or her particular case. However, in real life, we often have to make decisions on less-solid data. It is becoming popular to resort to decision analyses such as the one described by Younossi et al. to arrive at therapeutic conclusions. However, this approach has potential problems. These analyses are most appropriately applied when we know the

June 2000

precise effect of the different therapies and their exact costs. We can then use computer models to compare the costs and effects of each intervention. Unfortunately, when hard data are not available, assumptions have to be made. Before accepting the conclusions of Younossi et al., let us review some of the considerations that were the basis of the computer model, in particular those that assessed either the natural history of the disease or the clinical effect (reduction in incidence of hepatic mortality and morbidity) of the treatment. Because Younossi et al. believed that there was an inadequate amount of information available from prospective studies, they resorted to modeling the natural history on the experiences of tertiary referral centers. This information can give an estimate of how long it takes for livers to fail in patients who are referred there, but is not necessarily extrapolatable to the entire population. Such studies cannot determine the total number of patients from which these referrals came (i.e., the denominator). Patients referred to tertiary centers will, on the average, have more severe disease than those not referred (referral bias). Finally, these calculations will not even include patients in these centers who, while destined to develop liver failure, have not done so at the time of the last observation because their disease is progressing more slowly. Because of the lack of knowledge about the denominator and both types of selection biases, any model of disease progression from such centers will overestimate the rate and incidence of that progression. There are data from several cohort studies, i.e., studies of patients who, as a population, were identified at, or near, the time of infection. These studies have the advantage that the denominator is known. (On the other hand, they may somewhat underestimate the ultimate incidence of liver failure in chronic hepatitis C because not all of the patients become chronically infected.) In Seeff’s multi-institutional study of patients who developed posttransfusion hepatitis C in the 1970s, the overall survival of these individuals is the same as those who were transfused but did not develop hepatitis (Hepatology 1998;28:272A). Only 4% of them died of liver disease in the first 2 decades. Only 2 of 17 U.S. Army recruits with serological evidence of hepatitis C infection had any manifestations of liver failure 45–50 years later (Hepatology 1998;28:360A). Liver failure is not being observed in women infected with HCV in the late 1970s by contaminated globulin (N Engl J Med 1999;340:1228–1233). Vogt et al. did not observe this problem in children infected from blood transfusions during that same era (N Engl J Med 1999;341:866–870). We also have an epidemiological perspective. There may actually only be 2.7 million hepatitis C carriers in the United States (N Engl J Med 1999;341:556–562). Even so, there are only 10,000 deaths or liver transplantations every year because of hepatitis C. The average latent phase, i.e., the average amount of time that elapses between infection and death or transplantation, is calculated by dividing the prevalence by the incidence. Using 2.7 million and 10,000/year as the prevalence and incidence, that average latent phase is 270 years. The hepatitis C death/transplantation rate has been estimated to triple over the next decade (Hepatology 1998;28:390A). Even if that is so (and the estimate is based on a natural history model derived from data from tertiary referral centers), the average latent phase (90 years) is still greater than a normal life span. It must be that the vast majority of patients who are infected with the HCV are never destined to get into trouble. This conclusion is at odds with the model of Younossi et al. I used a pencil, paper, and hand calculator to extrapolate the model for 25 years. At the end of that time, it appears that 33% of the affected population would have died of liver failure, 1%–2% would have HCC, 2%–3% would have

SELECTED SUMMARIES

1269

undergone liver transplantation, and another 9% would be alive with decompensated liver disease. This calculation probably overestimated the incidence of liver morbidity and mortality, because I was unable to correct the data for expected deaths from other causes. However, it is likely that those deaths would not be a major factor, and the numbers I derived are remarkably similar to what was predicted by a similar model, a lifetime risk of decompensated cirrhosis of 46% ( JAMA 1998;280:2088–2093). It is probably incorrect to assume that patients with chronic hepatitis C will have otherwise normal life expectancies. These individuals got infected for some reason. Whether it was due to a risky life style or to transfusions for an underlying disease, their life expectancy is likely to be lower than that of an average person. In the multi-institutional posttransfusion hepatitis study, the mortalities of both groups were greater than that expected in the general population. If these patients tend to die prematurely of other diseases, they will be even less likely to live long enough to develop end-stage liver failure. The investigators also assumed that there is no spontaneous clearance of the HCV. Such events have occurred in 26% of the patients described by Seeff et al. (Hepatology 1998;28:407A) and 45% of those reported by Vogt et al. (N Engl J Med 1999;341:866– 870). Currently, we have no direct data proving that treatment improves subsequent hepatic morbidity or mortality. Younossi et al. seem to have assumed that the reduction in the hepatitis C carrier rate induced by treatment (for combination therapy, 31%) directly translates into a similar reduction (again, for combination therapy, 31%) in the subsequent incidence of liver failure. This assumption presumes that each patient is equally likely both to develop end-stage liver disease and to have a serological response induced by therapy. This has to be incorrect, because we know that responders tend to have been infected for shorter periods of time and by genotypes 2 or 3, to be female, to have lower titers of HCV RNA, and not to have extensive fibrosis on biopsy. Given their intrinsic characteristics, responders may very well be less likely to develop end-stage liver disease if left untreated. If this is true, the subgroup of patients who do not respond will have, on average, a worse prognosis than the group as a whole. With regard to the model, there will be less than a 31% decrease in the ultimate incidence of liver failure as a consequence of combination therapy. In fact, the extreme scenario would be that the small group of patients destined to develop liver failure (perhaps 10%–20% of the entire population) is composed of completely different people than the population (for combination therapy, 31%) who demonstrate a response. If this is the case, the ultimate incidence of liver failure will be precisely the same as it would have been without treatment. In this scenario, the only savings would be in the number of years that some patients would have abnormal laboratory test findings. The sensitivity analysis did not include the possibility of the treatment having no effect on the ultimate development of end-stage liver disease. If my hypothesis is true, treatment would be a very expensive way just to normalize some laboratory test results or to reduce the number of patients who are at risk of transmitting a disease that is known to be difficult to transmit anyway. Antiviral therapy is dangerous and expensive. There is no direct evidence that any lives will be saved as a result of its use. The assumptions of these decision analyses are not reliable enough to allow us to make any firm conclusions. At this time, a decision to treat is still only the consequence of a hopeful speculation. RONALD L. KORETZ, M.D.

1270

SELECTED SUMMARIES

Reply. Decision analysis and computer modeling can be powerful tools combining data from various sources to evaluate the impact of an intervention for a particular disease or to project the future course of a slowly progressive disease (Viral Hepatitis Rev 1999;5:220–230). These methods provide a systematic analysis of the same complex set of data confronting clinicians. Dr. Koretz’s assertion that these techniques should only be used when perfect data are available is tantamount to saying that a physician should not treat a patient until all the data are known with certainty. Just as clinicians and policy makers must make decisions on the basis of the best available data, a decision model can be used to evaluate these data. In fact, decision models are most needed when the data are imprecise. The extensive use of sensitivity analysis can identify which areas of imprecision have impact on the conclusion. Thus, the results not only serve as an aid in decision making, but can also identify which data need to be known with more certainty. It should be emphasized that although these analyses can be done from different perspectives, the societal perspective is the most appropriate one. In a clinical setting, decisions should be based not only on the best evidence for the efficacy of an intervention but also on the preferences of patients and physicians (Hepatology 1999;30:829–832). In this setting, decisions based solely on cost-effectiveness analyses are flawed and not appropriate. On the other hand, cost-effectiveness analyses are also performed to inform policy makers that treatment of chronic hepatitis C compares favorably with other health care interventions (such as coronary bypass grafting or renal dialysis). Clinical investigators in hepatology should be savvy enough to understand the role of different study designs that are used in clinical research and be aware of their value in different settings. In our analysis, the perspective was from the health care system’s perspective and extensive sensitivity analyses were performed to deal with uncertainties (Hepatology 1999;30:1328–1334). Additionally, our analyses were purposefully performed independent of any influence from the pharmaceutical industry. Thus, we feel our results are free of bias in this respect and represent our opinions on the economic issues related to this form of therapy. The second issue concerns the natural history of HCV and its potential for progression. If we believe Dr. Koretz that chronic hepatitis C progresses with an average latent period of 90 years, then we all have imaginary patients requiring liver transplantation or developing HCC. We do agree that selection bias has played a major role in the literature supporting progression. However, this same selection bias can also be seen in the studies favoring no progression. In fact, the 2 VA studies quoted as supporting no progression have a selection bias (comorbidities of cardiovascular diseases) or suffer from a relatively small sample size (only 17 of the recruits were ELISA and RIBA positive). Nonetheless, these studies raise an important issue. Chronic hepatitis C is not uniformly progressive, and patients can be roughly divided into slow progressors, rapid progressors, and those in between. We agree with this. Except for excess alcohol and age, other factors influencing progression are largely unknown. Obviously, a simple test to predict those at the greatest likelihood of progression and targeting the therapy to this group would be ideal. Unfortunately, no such tests currently exist. In a decision analysis, these different rates of progression can be dealt with in-depth sensitivity analyses, which we performed as published. The nihilistic position is that hepatitis C is rarely progressive and that the current efficacy rates for combination therapy (generated from randomized controlled trial of more than 1000 patients) are incorrect. The treatment of this disease is thus not worth the toxicity and cost. However, those of us who see increasingly more patients with HCV-related cirrhosis believe otherwise. We believe that although the

GASTROENTEROLOGY Vol. 118, No. 6

number of new HCV infections is decreasing, the epidemic of chronic liver disease from HCV is on the rise ( J Clin Gastroenterol 2000;30: 125–143). In this era of efficiency and cost-consciousness and as advocates for our patients, our role is to convince policy makers that regardless of less than perfect data for the natural history of this disease, chronic hepatitis C can be progressive and that our current treatments, far from perfect, compare well with the other health care interventions. This global approach to health care policy as well as our evidence-based approach at the bedside will serve our patients well now and in the future. ZOBAIR M. YOUNOSSI, M.D., M.P.H. JOHN G. MCHUTCHISON, M.D. MENDEL E. SINGER, Ph.D.

STOPPING THE UNSTOPPABLE? Barange K, Peron J-M, Imani K, Otal P, Payen J-L, Rousseau H, Pascal J-P, Joeffre F, Vinel J-P (Service d’Hepato-GastroEnterologie, Federation Digestive, CHU Purpan and Service de radiologie, CHU Rangueil, Toulouse, France). Transjugular intrahepatic portosystemic shunt in the treatment of refractory bleeding from ruptured gastric varices. Hepatology 1999;30: 1139–1143. Barange et al. studied the effect of transjugular intrahepatic portosystemic shunting (TIPS) for refractive gastric variceal bleeding over a 5-year period. All patients gave a previous history of upper gastrointestinal bleeding. Four of 48 patients died shortly after admission to intensive care unit. Forty-four patients were available for therapy. In 10 patients (23%), the bleeding was successfully controlled and sustained by the combination of vasoactive drugs and sclerotherapy or cyanoacrylate. Two patients were successfully treated surgically. The remaining 32 patients were treated with TIPS. TIPS was inserted in 20 patients as an emergency procedure for active bleeding, and hemostasis was achieved immediately in 18 and eventually in the remaining 2 patients. In 12 patients, TIPS was successfully inserted semielectively, for early rebleeding, after hemostasis had initially been achieved. TIPS insertion resulted in a significant decrease in the mean hepatoportal gradient from 17 to 8.4 mm Hg. Eight patients bled within 1 year, and a ninth patient rebled in the second 12-month period, giving an overall failure rate of 28%. The actuarial rate of remaining free of rebleeding at 12 months was 64%. Seven of these patients died, 5 without further treatment. At 12 months, the actuarial probability rate of survival was approximately 60%. TIPS was considered responsible for the deaths of 4 patients (13%). These deaths were caused by acute liver failure, peritoneal bleeding, massive bleeding secondary to total portal vein and mesenteric vein thrombosis, and sepsis. Overall, there were 18 episodes of shunt obstruction and sepsis. Five patients (15.6%) developed de novo encephalopathy, but in 4 patients encephalopathy present before the insertion of TIPS disappeared subsequently. Comment. Only 15%–20% of patients with varices have gastric varices (Dig Dis Sci 1991;36:303–309). But for the past 40 years (N Engl J Med 1960;263:665–669), the treatment of gastric, as opposed