Cost-effectiveness analysis of variceal ligation vs. beta-blockers for primary prevention of variceal bleeding


  • Thomas F. Imperiale,

    Corresponding author
    1. Division of Gastroenterology and Hepatology, Department of Medicine, Indiana University School of Medicine, IN
    2. The Regenstrief Institute Inc., Indianapolis, IN
    • The Regenstrief Institute Inc., 1050 Wishard Boulevard, Indianapolis, IN 46202
    Search for more papers by this author
    • Fax: 317-630-6611

  • Robert W. Klein,

    1. Medical Decision Modeling Inc., Indianapolis, IN
    Search for more papers by this author
  • Naga Chalasani

    1. Division of Gastroenterology and Hepatology, Department of Medicine, Indiana University School of Medicine, IN
    Search for more papers by this author

  • Potential conflict of interest: Nothing to report.


Although both β-blockade (BB) and endoscopic variceal ligation (EVL) are used for primary prevention of variceal bleeding (VB) in patients with cirrhosis with moderate to large esophageal varices (EVs), the more cost-effective option is uncertain. We created a Markov decision model to compare BB and EVL in such patients, examining both cost-effectiveness (cost per life year [LY]) and cost-utility (cost per quality-adjusted life year [QALY]). Outcomes included cost per LY, cost per QALY, proportions of persons with VB, TIPS, and all-cause mortality. EVL and BB were compared using the incremental cost-effectiveness ratio (ICER) and incremental cost-utility ratio (ICUR). When considering only LYs, initial EVL exceeds the benchmark of $50,000/LY, with an ICER of $98,407. However, when quality of life (QoL) is considered, EVL is cost-effective compared to BB (ICUR of $25,548/QALY). In sensitivity analysis, EVL is cost-effective if the yearly risk of EV bleeding is ≥ 0.26 (base case 0.15), the relative risk of bleeding on BB is ≥ 0.69 (base case 0.58), or if the relative risk of bleeding with EVL is < 0.27 (base case 0.35). The ICUR favored EVL unless the relative risk of bleeding on BB is < 0.46, the relative risk of bleeding with EVL is > 0.46, or the time horizon is ≤ 24 months. Whether EVL is “cost-effective” relative to BB therapy for primary prevention of EV bleeding depends on whether LYs or QALYs are considered. If only LYs are considered, then EVL is not cost-effective compared to BB therapy; however, if QoL is considered, then EVL is cost-effective. (HEPATOLOGY 2007;45:870–878.)

Portal hypertension is the pathophysiologic basis for the formation of esophageal varices, which are present in 30%-40% of patients with cirrhosis.1 Bleeding from varices occurs in about 30% of patients with varices,2 and carries a mortality risk of 15%-30% per episode.3, 4 Because of the high rates of bleeding and mortality, primary prevention is indicated. Nonselective β-adrenergic blockade has been the most widely investigated therapy. Several clinical trials5–8 and meta-analyses9, 10 have established the efficacy of nonselective β-blockade in reducing the risk of VB, though its effect on mortality is less certain. Despite the efficacy of β-blocker therapy, its effectiveness in clinical practice has been limited due to inability to achieve consistent reductions in the portosystemic gradient,11 the frequency of adverse effects and contraindications,12, 13 the need for long-term therapy,14 and risk of rebound bleeding.15

The search for alternatives led to consideration of endoscopic variceal ligation (EVL), effective for secondary prevention,14 as a potential modality for primary prophylaxis. Although clinical trials comparing EVL to no treatment show that ligation reduces bleeding-related and overall mortality,16–19 trials comparing EVL to β-blocker therapy have been less consistent: some trials show “no difference” in VB while others show that EVL reduces bleeding risk.20–25 Two meta-analyses comparing EVL with β-blocker therapy have shown that ligation reduces the risk of VB, although it may not increase survival.26, 27

Because both β-blocker therapy and EVL are effective in reducing the risk of VB, the issue of cost-effectiveness deserves consideration. Despite publication of several cost-effectiveness analyses for whether and how to screen patients with cirrhosis for esophageal varices,28–32 there has been no direct comparison of the costs and effectiveness of β-blocker therapy and EVL when used as primary prophylaxis. A recent editorial indicated that differences in the quality of life and cost between the 2 treatments are poorly defined.33 Therefore, we sought to determine whether and under what conditions β-blocker therapy or EVL is more “cost-effective”. In this analysis, we consider both quantity (cost-effectiveness) and quality (cost-utility) of life.


BB, β-blockade; EV, esophageal varice; EVL, endoscopic variceal ligation; ICER, incremental cost-effectiveness ratio; ICUR, incremental cost-utility ratio; LY, life year; QALY, quality-adjusted life year; TIPS, transjugular intrahepatic portosystemic shunt; VB, variceal bleeding.

Patients and Methods

Decision Model.

We created a Markov decision model targeted to patients with cirrhosis, portal hypertension, and medium-to-large esophageal varices who would be candidates for primary prophylaxis with either nonselective β-blocker therapy or EVL. The relevant literature from which probability estimates were derived suggests that this analysis is best applied to patients with Child class A or B cirrhosis. The main comparison is between initial use of β-blocker therapy and initial use of EVL to eradicate esophageal varices (Fig. 1). Patients who do not tolerate β-blocker therapy subsequently undergo EVL. Upon starting on a definitive method of primary prophylaxis, patients enter the Markov portion of the model (Fig. 2).

Figure 1.

Decision tree comparing β-blocker therapy versus variceal ligation for primary prophylaxis.

Figure 2.

Markov states for the model, as represented by rectangles. The circled state (i.e., first episode of esophageal variceal bleeding) is a transient state that lasts only until bleeding ceases; patients then go to 1 of 2 postbleed states for the remainder of the cycle. If no bleeding occurs during that cycle, patients then proceed to either of 2 “history of bleed” states.

In the Markov portion of the model, in which cycle length is 1 month, patients begin in a state of never having bled from varices. For the first episode of variceal bleeding, a transient state represented by the oval in Fig. 2, for which a transjugular intrahepatic portosystemic shunt (TIPS) may or may not be required, patients enter into 1 of 2 post-bleed states. Patients subsequently can either remain in those states by rebleeding during the next cycle or (more likely) go into 1 of 2 “history of variceal bleeding” states: one after TIPS and the other without TIPS. States with and without TIPS were modeled separately because the subsequent risk for variceal bleeding is much lower after TIPS placement, whereas mortality that is not related to bleeding is higher with TIPS.34 Death from bleeding or other causes may occur in any cycle from any state. Primary outcomes include direct costs, life-years, and quality-adjusted life-years (QALYs). QALYs measure health outcomes by assigning a weight ranging from 0 (death) to 1 (optimal health) for each time period, and correspond to the health-related quality of life during that period; these measures are then aggregated across time periods.35 Secondary outcomes are bleeding episodes resulting from esophageal varices, the proportion of persons with one or more one bleeding episodes, the need for TIPS for refractory bleeding, and all-cause mortality.

Model Assumptions.

Our model contains several assumptions that are based on available literature and/or on practice patterns. First, intolerance to β-blocker therapy occurs within the first month, and endoscopy with ligation is performed promptly upon discontinuation of the drug, such that time without prophylaxis is negligible. Second, ligation requires a mean of 3.3 sessions to eradicate varices,26 with surveillance endoscopy performed every 6 months thereafter to detect and ligate recurrent varices. Third, ligation is used as initial treatment for bleeding, with TIPS reserved for rescue treatment. Fourth, multiyear risks of bleeding or death obtained from the literature were assumed to be constant over time and were converted to cycle probabilities. Fifth, based on published data,3, 36 the risk of rebleeding after an episode of variceal bleeding was assumed to be 60% in untreated patients during the ensuing 12 months, with half of that risk (30%) occurring during the first month37 and the remaining 30% risk occurring in a constant fashion over the following 11 months.36 Based on published literature, ligation alone to the point of variceal obliteration reduced the risk of rebleeding by 50%;38, 39 the combination of ligation and β-blocker therapy reduced the risk of rebleeding by 75%.40, 41 Lastly, noncompliance was not explicitly modeled; however, early noncompliance can be considered to some extent by reducing the proportion of persons who tolerate β-blocker therapy, and later noncompliance would result in decreased efficacy, consistent with increased relative risks for bleeding. Both variables were examined in sensitivity analysis.

Perspective and Data Inputs.

As the perspective for the model is the health care system, we did not consider indirect costs such as time lost from work. We used a 5-year time horizon with 1-month cycle lengths because of the limitations of both the published literature and life expectancy in this population. We applied a discount rate of 3% to costs, life years, and QALYs. Point estimates and ranges used in sensitivity analyses for probabilities, utilities, and costs were derived from the published literature and Medicare allowable reimbursement (Table 1).

Table 1. Probabilities, Utilities, and Costs Used in the Analysis
 Value (Range)Source (Reference)
  1. Abbreviations: EV, esophageal varices; TIPS, transjugular intrahepatic portosystemic shunt.

Tolerating β-blocker therapy0.80 (0.60-1.00)(1, 32)
Annual risk of EV bleeding0.15 (0.10-0.30)(2, 12)
TIPS for refractory bleeding0.10 (0.00-0.50)(3, 34)
Death from bleeding episode0.15 (0.10-0.50)(3, 10, 12)
Rebleeding during first 30 days (untreated)0.30 (0.00-0.50)(37)
Rebleeding during days 31-3650.30(36)
Rebleeding after TIPS (annualized)0.125 (0.10-0.15)(54, 55)
Death from other causes (per year)0.143(7, 32, 56)
Relative risk of bleeding on β-blocker therapy0.59 (0.30-0.80)(9, 10, 12)
Relative risk of bleeding after prophylactic ligation0.35 (0.25-0.50)(26)
Relative risk of rebleeding after ligation alone0.50(38, 39)
Relative risk of rebleeding after ligation and β-blocker0.25(40, 41)
Compensated cirrhosis, on β-blocker therapy0.88 (0.80-1.00)(32, 57, 58)
Compensated cirrhosis, prophylactic ligation0.90 (0.80-1.00)(32, 58)
Intolerant to β-blocker therapy0.80 (0.70-0.90)(32, 58)
Bleeding episode (1 cycle)0.30 (0.20-0.50)(32, 58)
Postbleed, no TIPS required0.63 (0.50-0.75)(32, 58)
TIPS placement (1 cycle)0.30 (0.20-0.50)(32, 58)
Post-TIPS, no rebleeding0.60 (0.50-0.70)(32, 58)
Costs ($)  
β-blocker (1-month cost)20 (10-50)
Ligation (per session)600 (400-1000)(59)
Endoscopic surveillance (every 6 months after eradication of varices)618 (300-1000)(59)
Bleeding episode, no TIPS15,000 (10,000-20,000)(32, 60)
Bleeding episode, TIPS insertion16,700 (10,000-25,000)(61)
TIPS follow-up (monthly)115 (90-150)(49)
Discount rate3% (0-5%)(62)

Sensitivity Analysis.

We performed one-way sensitivity analyses for each variable using the ranges shown in Table 1. Based on these results, the more clinically relevant or sensitive variables were subjected to 2-way sensitivity analysis. Probabilistic sensitivity analysis was performed by varying all variables simultaneously over their plausible ranges. This process involved assigning distributions to 27 variables (11 event probabilities, 7 costs, 3 relative risks, and 6 utilities) used in the model. For the Monte Carlo simulation, 1000 iterations were run for both strategies using unique combinations of the 27 distributions, with values for each variable selected randomly. The differences of the strategies for each run were recorded to determine how often each was considered more cost-effective. TreeAge Pro 2005 software, Healthcare version release 1.4 (TreeAge Software Inc., Williamstown, MA), was used for all analyses.


Base Case Results.

When only life-years are considered, initial use of β-blocker therapy has a 5-year discounted cost of $6,745 and an effectiveness measure of 3.150 life-years, while initial variceal ligation costs $9,402 and has an effectiveness measure of 3.177 life-years (Table 2). The incremental cost-effectiveness ratio (ICER), which is the difference in costs divided by the difference in effectiveness (in this case, life-years) is $98,407 per life-year saved. The magnitude of this value exceeds the usual benchmark for cost-effectiveness, which is around $50,000 per life-year saved,35 suggesting that initial use of ligation is not cost-effective when compared with β-blocker therapy.

Table 2. Base Case Results—Primary Outcomes
StrategyCost ($)Life yearsIncremental cost-effectiveness ratio*Quality-adjusted life yearsIncremental cost-utility ratio*
  • *

    Incremental cost-effectiveness ratios were calculated using additional significant figures for life years and QALYs.

β-blocker therapy67453.1502.67

When both life-years and quality are considered, however, there is a decrement in the effectiveness measure for both strategies, and more so for β-blocker therapy (Table 2). The incremental cost-utility ratio (ICUR) is $25,548 per QALY, suggesting that the gain in effectiveness with initial ligation is “worth” the increase in cost.

Five-year rates for the other outcomes are shown in Table 3. These results imply that, over 5 years, for every 1000 patients with high-risk varices, initial ligation prevents variceal bleeding in 77 persons, 65 bleeding episodes, 5 TIPS procedures, and 7 deaths, as compared with initial β-blocker therapy. Expressed as the number needed to treat,42 13 patients would have to undergo ligation for one person to avoid variceal bleeding during this 5-year period over the protection afforded by initial β-blocker therapy.

Table 3. Base Case Results—Secondary Outcomes
StrategyBleeding episodes per 1000 personsPersons who bled (%)TIPS (%)Death (%)
β-blocker therapy32125.32.757.7

Sensitivity Analysis.

Relative to the arbitrary benchmark for cost-effectiveness of $50,000 per life year (or per QALY) saved, results were not sensitive to most variables over the ranges in Table 1. Variables to which the results were sensitive are shown in Tables 4 and 5. When considering only life-years, β-blocker therapy is preferred unless the threshold condition for any of the 5 variables in Table 4 is met. All baseline probabilities in Table 4 are in a direction opposite to those threshold probabilities required to make EVL cost-effective when compared with β-blocker therapy. And although all threshold values are contained within the ranges used in 1-way analysis, most are at the extremes of those ranges. The relative risk of bleeding while on β-blocker therapy was 0.58 in the base case. If the relative risk were < 0.40, indicating enhanced efficacy, then β-blocker therapy would be cost-saving. With regard to time, the relative cost-ineffectiveness of ligation was not sensitive to the 5-year time horizon, as the ICER for ligation still exceeded $60,000 with a 10-year time horizon. For the scenario where all persons in the ligation group receive both ligation and β-blocker therapy for an episode of bleeding, assuming an 80% tolerance rate for β-blocker therapy and the additional reduction in the risk of re-bleeding, the ICER was $63,250 per life-year saved. Lastly, the monthly cost of β-blocker therapy would have to exceed $64 before the ICER for ligation would fall to $50,000.

Table 4. Results of 1-Way Sensitivity and Threshold Analyses for Cost-Effectiveness: Ligation Versus β-Blocker Therapy
VariableBaseline probability (range)Range of ICERs ($ per life year)*Threshold for cost-effectiveness
  • *

    Base case incremental cost-effectiveness ratio = $98,407.

  • Threshold for cost savings using β-blocker therapy is < 0.40.

Annual risk of variceal bleeding0.15 (0.1-0.3)$164,322 to $42,027≥0.26
Response to β-blocker therapy0.45 (0.25-0.65)$40,268 to $601,400<0.31
Relative risk of bleeding on β-blockers≈0.58 (0.30-0.80)($266,347) to 25,902≥0.69
Relative risk of bleeding with ligation0.35 (0.25-0.50)$42,322 to $815,975<0.27
Relative risk of rebleeding on β-blockers0.25 (0.10-0.50)$167,003 to $98,585>0.48
Table 5. Results of 1-Way Sensitivity and Threshold Analyses for Cost Utility: Ligation Versus β-Blocker Therapy
VariableBaseline probability (range)Range of ICURs ($ per QALY)*Threshold for cost-effectiveness
  • *

    Base case incremental cost-utility ratio = $25,548.

Response to β-blocker therapy0.45 (0.25-0.65)$13,160 to $54,539≤0.63
Relative risk of bleeding on β-blocker therapy≈0.58 (0.3-0.8)$211,827 to $9,082≥0.46
Relative risk of bleeding with ligation0.35 (0.25-0.50)$13,862 to $66,375≤0.46
Months of observation60 months (10-60 months)$228,258 to $25,077>24 months

As in the base case, the results are quite different when QALYs are considered. In general, the ICUR favored EVL for most variables. EVL was preferred on the basis of this ratio unless the relative risk of bleeding with β-blocker therapy was less than 0.46; the relative risk of bleeding following successful EVL exceeded 0.46; or the probability of responding to β-blocker therapy exceeded 0.63 (Table 5). In contrast to the results of the 1-way sensitivity analyses for life years, the baseline probabilities are in a direction consistent with the threshold probabilities, and the threshold values are closer to the midpoint of the ranges used in sensitivity analysis. In terms of time horizon, EVL became cost-effective after 24 months, which is well within the timeframe of this analysis. Reducing the disutility for β-blocker therapy to zero would not affect the clinical decision, because the ICUR for ligation remains below $50,000 per QALY. With regard to cost, EVL remains cost-effective even if the monthly cost of β-blocker therapy is zero. With regard to the utility of 0.30 (range, 0.20-0.50) for the single cycle representing a bleeding episode, there was no value (from 0 to 1) that affected the base-case decision.

We performed 2-way sensitivity analyses of the relative risks of bleeding with both treatments (Fig. 3A,B). As with base-case and one-way sensitivity analyses, the results differ based on whether life years or QALYs are considered, but only in a quantitative sense. When cost per life year is considered, the absolute difference in relative risk between β-blocker therapy and ligation must exceed 0.35 for ligation to be cost-effective. (In the base-case, the absolute difference was 0.23.) When cost per QALY is considered, the absolute difference in relative risk need only exceed 0.12 for ligation to be cost-effective relative to β-blocker therapy.

Figure 3.

Two-way sensitivity analysis of the relative risks of bleeding with both treatments. Two-way sensitivity analyses of the incremental cost (A) per life year saved and (B) per QALY saved at varying levels of relative risks of bleeding with β-blocker therapy and ligation.

Due to the more subjective nature of the utility values for the model, we increased the upper limit for the utility of the postbleed state without TIPS to 0.80. The ICUR for EVL compared to β-blocker therapy was $2441 greater than when the upper limit was 0.75 and only $7008 more than the base case, which is a relatively small change. Further, we linked the increased upper limit for the postbleed utility to comparable increases for 3 other utilities (bleeding states with and without TIPS, and the postbleed state with TIPS) to determine how a simultaneous increase in all 4 utilities would affect the base case results. At the new upper bounds for all 4 utilities (0.80, 0.77, 0.40, and 0.40), the ICUR for EVL increased from $25,553 to $33,213 – a small increase that would not affect the choice of therapy.

In probabilistic sensitivity analysis, the 27 input values were varied among 1000 simulations. Results are expressed as the proportion of iterations in favor of either strategy at a particular dollar amount known as the “willingness to pay” (Fig. 4A,B). When considering life-years alone, β-blocker therapy is more cost-effective than ligation at least 75% of the time when the willingness to pay is $60,000 or less (Fig. 4A). When the willingness-to-pay threshold is $100,000 per life-year saved, the 2 strategies are cost-effective the same percent of the time. However, when QALYs are considered, ligation is more cost-effective than β-blocker therapy 92% of the time when the willingness to pay is $50,000 per QALY, and essentially 100% of the time when it is $100,000 (Fig. 4B).

Figure 4.

Probabilistic sensitivity analysis. Proportions of simulations that are cost-effective according to the willingness to pay. Depending on willingness to pay for (A) a life year or (B) a QALY, the curves show the proportions of the time when β-blocker therapy or ligation is the more cost-effective choice.


Because of the morbidity, mortality, and cost of variceal bleeding, current guidelines recommend that patients with cirrhosis undergo diagnostic endoscopy to look for varices.43 For patients at high risk because of moderate-to-large esophageal varices (F2 or F3), primary prophylaxis is recommended.43, 44 Beta-blocker therapy reduces the risk of variceal bleeding, but is often either contraindicated by comorbid conditions or is not well tolerated. In light of recent clinical trials20–25 and meta-analyses26, 27 demonstrating comparable or better effectiveness of ligation for primary prophylaxis, we conducted the current analysis to determine the cost-effectiveness and cost-utility of these 2 modalities.

We found that whether ligation or β-blocker therapy is more “cost effective” depends on whether quality, not just quantity of life, is considered. If only quantity of life is considered, then ligation is not cost-effective within the ranges of most factors considered in sensitivity analysis. In the base case, the incremental cost per life-year saved for ligation is $98,407. However, if quality of life is considered, then the added cost of ligation appears to be worth the gain in QALYs, with an ICUR of $25,548/QALY.

Other analyses have considered the costs and/or outcomes of primary prophylaxis for patients with high-risk esophageal varices.28, 32 Teran and colleagues used a Markov model to examine the costs and effects of β-blocker therapy, sclerotherapy, shunt surgery, and observation. Beta-blocker therapy resulted in cost-savings and a modest increase in quality-adjusted life expectancy when compared with no treatment. This analysis did not consider ligation, and did not examine the ICERs among the strategies.32 An analysis by Aoki and colleagues also used a Markov model to compare ligation, β-blocker therapy, and observation for the number of bleeding-free life years in a hypothetical population of 50 year-old men with high-risk esophageal varices.28 Baseline analyses indicated 3.19 bleeding-free life years with ligation, 3.04 life years with β-blocker therapy, and 2.01 life years with observation. In probabilistic sensitivity analyses, 77% of the hypothetical cases had more bleeding-free life years after EVL. Only the risks of bleeding after EVL and after β-blocker therapy influenced the treatment choice. Although this work thoroughly considered outcomes, costs and quality of life were not considered. In a recent clinical trial comparing EVL to β-blocker therapy, Jutabha and colleagues reported that the cumulative direct costs of both treatments were no different.22 However, the simple cost analysis from this study is based on a small sample size. Further, the true costs incurred by patients participating in the trial were not included. We recognize that other strategies for diagnosis and treatment of varices have been examined, including empiric β-blocker therapy for all patients with cirrhosis without looking endoscopically for varices.31 However, this strategy begins at a different point (no endoscopy, suspected or confirmed cirrhosis) than ours, which begins when moderate-to-large varices are identified endoscopically.

In the current analysis, base case results were insensitive to most variables. For those variables to which the ICER was sensitive, the threshold probabilities for changing the base case findings fell in opposite directions to base case probabilities, whereas those for the ICUR fell in the same direction. The threshold probabilities were at the extremes of the ranges used for life years and closer to the mid-range for QALYs, suggesting that the life year results are more robust than those for QALYs.

Our utility values were adapted from previous work and are not based on values derived from large, representative patient cohorts. This limitation would seem to reduce the validity of our findings for QALYs. However, sensitivity analysis using widely varying utility ranges had no impact on the main findings. Further, the disutility value of 0.02 assigned to β-blocker therapy was conservative, such that a greater value would have improved the cost-utility of ligation relative to β-blocker therapy. Interestingly, even when the disutility for β-blocker therapy is zero, ligation remains cost-effective relative to β-blocker therapy because β-blocker therapy results in more QALY deductions for reduced health states during and following increased bleeding episodes.

In this analysis, all β-blocker use was unmonitored, meaning that the hemodynamic response was not measured. Several studies have shown that β-blocker therapy is particularly effective for both primary and secondary prophylaxis when the hepatic venous pressure gradient falls to less than 12 mm Hg or achieves a reduction of ≥25%.45–48 Measuring the response to β-blocker therapy by hemodynamic monitoring has been found to be cost-effective, particularly when nitrates are added prior to abandoning drug therapy altogether and proceeding to ligation.49 For this reason, as part of an extended sensitivity analysis, we examined the cost-effectiveness of monitoring the response to β-blocker therapy (data not shown). However, because the use of nitrates in combination with β-blockers for primary prophylaxis has recently fallen into disfavor, we considered β-blocker therapy without adding nitrates for nonresponders. Our findings again depended on whether quality of life was considered. When only life years are considered, monitoring the hemodynamic response to β-blocker therapy was not cost-effective compared to unmonitored β-blocker therapy unless the cost of monitoring was less than $560, the annual probability of bleeding exceeded 20%, or the time horizon exceeded 7 years. When compared to ligation, however, hemodynamic monitoring was nearly always cost-effective. When quality of life is considered, hemodynamic monitoring was nearly always cost-effective compared to unmonitored β-blocker therapy, but it was not cost-effective when compared with ligation. The comparison of hemodynamic monitoring to ligation was sensitive to procedure costs and the disutility of β-blocker therapy. Monitoring became cost-effective compared to ligation if the cost of monitoring was less than $1480, the cost of ligation exceeded $775, or the disutility of β-blocker therapy was less than 0.014.

This analysis has several limitations. First, we did not consider severe complications from TIPS, ligation, or β-blocker therapy.27 We included costs for monitoring and treatment of TIPS stenosis. With regard to EVL, although minor complications such as chest pain, dysphagia, and fever are relatively common, they are transient and would, at most, add only to cost. The added cost for minor complications was indirectly addressed through sensitivity analysis, where the upper limit of the cost for EVL was $1,000. In contrast, major complications are uncommon. A meta-analysis comparing EVL with β-blocker therapy showed that severe adverse events were less common with EVL (9 of 285 for EVL versus 39 of 311 with β-blocker therapy; relative risk = 0.34, 95% CI, 0.17-0.69), with no difference in treatment-induced bleeding (8 cases for EVL versus 9 for β-blocker withdrawal).27 While including costs for complications of EVL would have increased the cost of both strategies (though more so for EVL) and would have made β-blocker therapy appear more cost-effective, considering the bleeding rate from β-blocker withdrawal as well as other causes, its other side effects may well have exceeded the cost of ligation-induced complications. Second, we did not explicitly consider nonadherence, making our analysis more of a “cost-efficacy” analysis, meaning that we determined how cost-effective both strategies could be if adherence were high. We captured nonadherence to some extent by including intolerance to β-blocker therapy (which resulted in the use of EVL) and varying this factor in sensitivity analysis.

A third limitation is that we did not consider the costs and effects of liver transplantation or development and management of HCC. Because this analysis pertains to patients with Child class A and B cirrhosis, liver transplantation would be infrequent. Similarly, the annual incidence of hepatoma in these patients is low, approximately 0.5%-4% per year.50–53 It should be noted that mortality from hepatoma is included in nonbleed death rates, although treatment costs are not. These relatively downstream events are likely to occur at similar rates irrespective of the method of primary prophylaxis used. In total, we believe that the effects of liver transplantation and hepatoma on our results would be small and differentially negligible.

We did not include a “control” strategy of no treatment because we believe that some type of primary prophylaxis, usually β-blocker therapy, is considered to be the standard of care for patients with moderate to large esophageal varices, making it important to understand the incremental cost-effectiveness of any other modality relative to β-blocker therapy. Also, we did not include β-blocker co-therapy to prevent rebleeding in the base case ligation strategy, despite its effect in reducing the risk of rebleeding beyond that of EVL alone.40, 41 However, we did consider such co-therapy in sensitivity analysis, and found that the ICER decreased from $98,407 to $63,250. Although still beyond the $50,000 per life year threshold, the reduced ratio makes ligation more sensitive to variations of other factors, some of which would make EVL cost-effective for life years as well as QALYs. Finally, the truncating effect of the 5-year time horizon tends to favor β-blocker therapy, as at the end of 5 years, there were 7 more persons alive and 77 fewer who had bled per 1000 simulated patients undergoing EVL. As a result, more life years and QALYs would accrue to ligation beyond 5 years, narrowing the cost-effectiveness ratio that favors β-blocker therapy at 5 years and further reducing the cost-utility ratio that already favors EVL.

We conclude that whether EVL is “cost-effective” relative to β-blocker therapy for primary prevention of esophageal variceal bleeding depends on whether life years or QALYs are considered. At a $50,000 willingness to pay, EVL is not cost-effective compared to β-blocker therapy if only life years are considered; however, ligation is cost-effective if QALYs are considered. Further study is needed to determine the extent to which these findings may be affected by the use of β-blocker co-therapy after a first bleed, an extended time horizon, and by patient-derived utilities.