Cost-effectiveness of hepatic venous pressure gradient measurements for prophylaxis of variceal re-bleeding

Authors


Dr M. R. Arguedas, UAB Liver Center, MCLM 281, 1918 University Boulevard, Birmingham, AL 35294-0007, USA.
E-mail: Arguedas@uab.edu

Summary

Background : Measurement of the hepatic venous pressure gradient may identify a sub-optimal response to drug prophylaxis in patients with a history of variceal bleeding. However, the cost-effectiveness of routine hepatic venous pressure gradient measurements to guide secondary prophylaxis has not been examined.

Methods : A Markov model was constructed using specialized software (DATA 3.5, Williamstown, MA, USA). Three strategies involved secondary prophylaxis without haemodynamic monitoring using beta-blockers alone, beta-blockers plus isosorbide mononitrate or endoscopic variceal ligation alone. Four strategies involved secondary prophylaxis with beta-blockers plus isosorbide mononitrate or beta-blockers alone, accompanied by one or two hepatic venous pressure gradient measurements to identify haemodynamic non-responders, who underwent endoscopic variceal ligation as an alternative. The total expected costs, variceal bleeding episodes and total deaths were calculated for each strategy over 3 years.

Results : The two most effective strategies were combination therapy alone and combination therapy with two hepatic venous pressure gradient measurements. The incremental cost-effectiveness ratio of the latter strategy was $136 700 per year of life saved compared with combination therapy alone. The ratio improved as the time horizon was extended or the rates of variceal re-bleeding were increased.

Conclusions : The cost-effectiveness of haemodynamic monitoring to guide secondary prophylaxis of recurrent variceal bleeding is highly dependent on local hepatic venous pressure gradient measurement costs, life expectancy and re-bleeding rates.

Introduction

Variceal re-bleeding is a significant cause of morbidity and mortality in patients with portal hypertension. Over the past three decades, multiple treatment regimens have been investigated in order to decrease the incidence of variceal re-bleeding in patients with portal hypertension, including portal-systemic shunts, endoscopic therapy and pharmacological agents. The use of non-selective beta-blockers, alone or in combination with isosorbide mononitrate, has been shown to decrease variceal bleed rates in multiple studies of patients with oesophageal varices.1–5 They have therefore become the most common therapy in the secondary prophylaxis of variceal bleeding.

The benefit of beta-blocker therapy has been attributed to a decrease in portal pressure, as demonstrated by a decrease in the hepatic venous pressure gradient (HVPG).6 Variceal bleed rates are directly related to HVPG measurements, as demonstrated by the absence of variceal bleeding in patients with an absolute HVPG of less than 12 mmHg,7, 8 and rates of 5–8% over 2–3 years when secondary prophylaxis results in a 20% reduction in HVPG from baseline.9–11 However, several studies have demonstrated that 49–64% of patients do not respond to pharmacotherapy with a drop in HVPG to less than 12 mmHg or by 20%.9–12 These studies have also shown that, in patients who fail to ‘respond’ to therapy, variceal bleed rates are high, ranging from 39% to 62% over 2–3 years.

It would be beneficial to identify patients who do not respond to pharmacological therapy in order for alternative therapies to be implemented. Due to a lack of a non-invasive test to evaluate reliably the response of portal pressure to pharmacotherapy,13 measurement of HVPG has been suggested as a way to identify patients who fail to attain an adequate haemodynamic response with medical therapy.14, 15 HVPG measurement has not been studied as a tool for the modification of prophylaxis in patients with bleeding oesophageal varices. We used data from previous studies to construct a Markov model of prophylaxis for recurrent variceal haemorrhage, comparing therapy guided by HVPG measurement with all other primary therapies. In our model, HVPG measurement was used to identify patients who do not respond to pharmacotherapy in order to offer alternative therapy, including endoscopic variceal ligation.

Materials and methods

The model

A Markov model was developed using specialized software (DATA 3.5, Williamstown, MA, USA) to estimate the costs and outcomes associated with secondary prophylaxis strategies for variceal bleeding.

In a Markov model, members of a cohort are divided amongst several mutually exclusive health states, and movements of the cohort across these states are modelled over time.16, 17 The health states are defined to capture the salient characteristics of the disease and the treatments under consideration. The time horizon of the model is divided into a series of evenly spaced periods or cycles. At the end of each cycle, the members of the cohort are re-allocated across the health states, with these movements guided by transition probabilities that reflect the natural history of the disease in question. The model permits the estimation of the total costs, clinical events and life expectancy in patients undergoing a particular medical intervention. For the base-case analysis, each cycle consisted of 6 months and the cohort was followed for six cycles (3 years).

On entry, patients surviving an initial episode of variceal bleeding underwent secondary prophylaxis for re-bleeding. The typical patient represented a 55-year-old individual with cirrhosis and no other significant co-morbidities with an average Child–Pugh score of 8. During each cycle, patients could occupy one of several health states designed to incorporate the natural history after variceal bleeding. Specifically, patients could experience recurrent variceal bleeding, procedure-related complications, death as a result of re-bleeding or death due to progression of the underlying liver disease (Figure 1). Transition probabilities, in order to move the patients of the cohort through different health states over time, were obtained from the literature by performing a Medline™ search. The search terms included variceal bleeding, portal hypertension, secondary prophylaxis, hepatic venous pressure gradient and portal pressure. Further data were obtained from the articles retrieved by manually reviewing the reference list of each of the articles retrieved via Medline. At the end of each cycle, the model accrued units of effectiveness (life expectancy), number of re-bleeding episodes and costs for each patient in a particular secondary prophylaxis strategy.

Figure 1.

Outline of model and health states. EVL, endoscopic variceal ligation; HVPG, hepatic venous pressure gradient; VB, variceal bleeding.

Table 1 shows the cost and probability estimates for our baseline analysis.1–4, 7–12, 18–31 For our analysis, we took the perspective of a third-party payer (Medicare).

Table 1.  Model variables: baseline values and ranges used in sensitivity analyses
VariableBase caseRangeReference
  • Probabilities presented semi-annually unless otherwise specified.

  • † 

    Medicare reimbursement data obtained from the University of Alabama at Birmingham business office.

Probabilities*
Re-bleeding without haemodynamic monitoring (per 6 months)
 Endoscopic variceal ligation0.1070.05–0.1712, 18–20
 Beta-blockers0.1250.03–0.191
 Combination0.0850.05–0.1012, 20–22
Haemodynamic response
20% reduction compared with baseline
 Beta-blockers0.360.19–0.559–11, 22, 23
 Combination0.510.40–0.679, 12, 23, 24
Less than 12 mmHg
 Beta-blockers0.120.10–0.219–11, 22, 23
 Combination0.140.10–0.2012, 23
Re-bleeding according to haemodynamic response (per 6 months)
20% reduction compared with baseline
 Beta-blockers0.020–0.0510
 Combination0.040–0.0512
Less than 12 mmHg
 Beta-blockers0N/A7–10, 22, 25
 Combination0N/A 
Complications due to endoscopic variceal ligation (per episode)0.0010.0001–0.0126
Intolerance to drugs
 Beta-blockers0.050.05–0.201, 3, 4
 Combination0.130.07–0.1919–21, 24
Death due to variceal bleeding (per episode)0.200.05–0.3027
Death due to other causes (per 6 months)0.0350.016–0.302
Death due to endoscopic variceal ligation complications (per episode)0.10.01–0.25Assumption
Costs (US$)
Esophagogastroduodenoscopy365150–500Medicare
Beta-blockers126100–45028
Isosorbide mononitrate125100–45028
Hepatic venous pressure gradient2500500–2500Medicare
Variceal bleeding hospitalization15 1105300–35 000Medicare, 29–31
Complications from endoscopy10 0002000–20 000Medicare

The strategies

We compared seven strategies of secondary prophylaxis for variceal bleeding.

Strategy 1 (beta-blockers without haemodynamic monitoring).  In this strategy, secondary prophylaxis for variceal bleeding consisted of non-selective beta-blockers. HVPG measurements were not performed.

Strategy 2 (combination beta-blockers and isosorbide mononitrate without haemodynamic monitoring).  In this strategy, secondary prophylaxis for variceal bleeding consisted of beta-blockers and isosorbide mononitrate. HVPG measurements were not performed. We considered this strategy as ‘usual care’, given the results of recent trials which document greater reductions in portal pressure and lower re-bleeding rates with combination therapy.9, 12

Strategy 3 (endoscopic variceal ligation).  The entire cohort underwent secondary prophylaxis with endoscopic variceal ligation. Obliteration was achieved with an average of 3.5 sessions. Follow-up surveillance endoscopy for recurrent varices was performed at 6-month intervals.

Strategy 4 (beta-blockers with one HVPG measurement).  In this strategy, secondary prophylaxis consisted of non-selective beta-blockers. A single HVPG measurement was performed 3 months after drug initiation. Response to therapy was defined as an HVPG decrease to < 12 mmHg. Responders continued beta-blocker therapy, whereas non-responders underwent endoscopic variceal ligation as secondary prophylaxis.

Strategy 5 (beta-blockers with two HVPG measurements).  In this strategy, secondary prophylaxis consisted of non-selective beta-blockers. HVPG measurements were performed prior to and 3 months after drug initiation. Response to therapy was defined as HVPG < 12 mmHg or a decrease of ≥ 20% compared with baseline. Responders continued beta-blocker therapy, whereas non-responders underwent endoscopic variceal ligation as secondary prophylaxis.

Strategy 6 (combination therapy with one HVPG measurement).  In this strategy, secondary prophylaxis consisted of combination non-selective beta-blockers and isosorbide mononitrate. A single HVPG measurement was performed 3 months after drug initiation. Response to drugs was defined as an HVPG decrease to < 12 mmHg. Responders continued pharmacotherapy, whereas non-responders underwent endoscopic variceal ligation as secondary prophylaxis.

Strategy 7 (combination therapy with two HVPG measurements).  In this strategy, secondary prophylaxis consisted of combination non-selective beta-blockers and isosorbide mononitrate. HVPG measurements were performed prior to and 3 months after drug initiation. Response was defined as HVPG < 12 mmHg or a decrease of ≥ 20% compared with baseline. Responders continued pharmacotherapy, whereas non-responders underwent endoscopic variceal ligation as secondary prophylaxis.

Clinical data and assumptions

The probabilities of recurrent variceal bleeding, haemodynamic response to pharmacotherapy, intolerance to pharmacotherapy, complications due to endoscopy and death due to variceal bleeding or other causes are shown in Table 1. All rates were transformed into probabilities with the use of the formula p = 1 − ert, where r is the rate of the event occurring over a defined time period (t).

Regardless of the initial secondary prophylaxis strategy, once an episode of recurrent variceal bleeding occurred, we included the costs and outcomes of hospitalization with 3.5 sessions of endoscopic variceal ligation.

Although non-variceal bleeding mortality was included in our model, morbidity due to complications from chronic liver disease, such as progressive hepatocellular dysfunction, portosystemic encephalopathy, ascites, spontaneous bacterial peritonitis and hepatocellular carcinoma, was not included. We assumed that these complications would occur with equal frequency amongst all strategies and would not contribute excess costs or mortality to any one strategy. The probability of liver transplantation was not included in the model, as it was likely to add costs and life expectancy equally amongst the different strategies.

The medical literature reviewed for this study did not report complications associated with the measurement of HVPG. This also reflects the experience at our institution, where serious adverse events associated with HVPG measurements have not been observed. Therefore, transition probabilities associated with complications of HVPG measurement were assumed to be negligible.

Costs

All costs are reported in 2002 US dollars. The costs utilized in the model represent the average Medicare reimbursement rates at our institution according to diagnosis-related groups and Current Procedural Terminology codes. The costs of drugs represented the average wholesale price.28

Cost-effectiveness analysis

The model was used to estimate patient life expectancy, number of episodes of recurrent variceal bleeding and expected health care costs under each of the seven strategies. These strategies were then compared using the methods of incremental cost-effectiveness analysis.32 Following this approach, successively more effective and more costly strategies were compared with each other using incremental cost-effectiveness ratios. The incremental cost-effectiveness ratio comparing two strategies is defined as the difference in the strategies' costs divided by the difference in their effectiveness. The incremental cost-effectiveness ratio measures the additional expenditure required for each additional unit of health gain obtained by choosing a more costly but more effective intervention.

Our results are reported in terms of US dollars per year of life saved as well as in US dollars per episode of recurrent variceal bleeding averted. As recommended by the US Panel on Cost-Effectiveness in Health and Medicine, all future costs and health benefits were discounted to their present value using an annual discount rate of 3%.33

Sensitivity analyses

Sensitivity analyses were performed to explore the degree to which our results were influenced by uncertainty regarding the parameter values used in the model. In one-way sensitivity analyses, results were re-calculated as the values of model parameters were varied separately. In addition, two-way sensitivity analyses were performed to examine the results of simultaneously varying select pairs of variables that were found to be influential in one-way sensitivity analyses.

Results

Base-case analysis

The undiscounted and discounted results of the base-case analysis are shown in Tables 2 and 3. Discounted analysis demonstrated a cost of $9504 per year of life saved on combination therapy without haemodynamic monitoring. The remaining strategies were more expensive, including beta-blocker therapy and endoscopic variceal ligation without haemodynamic monitoring. Combination therapy with two HVPG measurements was estimated to cost $13 605 per year of life saved.

Table 2.  Base-case results (costs per year of life saved) (undiscounted)
Strategy Costs*Life expectancy*Incremental cost-effectiveness ratio
  • HVPG, hepatic venous pressure gradient.

  • Costs (2000 US$) and effectiveness (life expectancy) are reported per patient.

  • † 

    Incremental cost-effectiveness ratio reported as US$/year of life saved.

  • ‡ 

    The incremental cost-effectiveness ratio reported is that of combination therapy with two HVPG measurements compared with combination therapy without haemodynamic monitoring. All other strategies are strongly dominated.

Combination therapy96372.47
Beta-blocker11 2822.42Dominated
Endoscopic variceal ligation12 2082.44Dominated
Combination therapy (1 HVPG)12 9892.47Dominated
Beta-blocker (1 HVPG)13 3002.46Dominated
Combination therapy (2 HVPG)13 7552.51117 657
Beta-blocker (2 HVPG)13 7992.50Dominated
Table 3.  Base-case results (costs per year of life saved) (discounted)
Strategy Costs*Life expectancy*Incremental cost-effectiveness ratio
  • HVPG, hepatic venous pressure gradient.

  • Costs (2000 US$) and effectiveness (life expectancy) are reported per patient.

  • † 

    Incremental cost-effectiveness ratio reported as US$/year of life saved.

  • ‡ 

    The incremental cost-effectiveness ratio reported is that of combination therapy with two HVPG measurements compared with combination therapy without haemodynamic monitoring. All other strategies are strongly dominated.

Combination therapy95042.25
Beta-blocker11 1862.20Dominated
Endoscopic variceal ligation12 0432.22Dominated
Combination therapy (1 HVPG)12 8292.24Dominated
Beta-blocker (1 HVPG)13 1472.24Dominated
Combination therapy (2 HVPG)13 6052.28136 700
Beta-blocker (2 HVPG)13 8262.27Dominated

Amongst the strategies which did not include HVPG measurement, beta-blocker therapy alone was associated with an average life expectancy of 2.20 years and resulted in 67 episodes of variceal bleeding per 100 patients. Combination therapy added 0.05 years to the average life expectancy and resulted in 15 fewer episodes of re-bleeding. Endoscopic variceal ligation resulted in a life expectancy of 2.22 years and 61 variceal bleeding episodes per patient.

Amongst patients undergoing one HVPG measurement, those on beta-blockers had a life expectancy of 2.24 years and experienced 54 episodes of variceal bleeding, whereas those on combination therapy had an equal life expectancy but one less episode of variceal bleeding (53 per 100 patients). Patients on beta-blockers who underwent two HVPG measurements had a life expectancy of 2.27 years and suffered 44 episodes of variceal bleeding. Combination therapy with two HVPG measurements added 0.01 years and resulted in a marginally lower absolute re-bleeding rate (0.2%) compared with patients undergoing haemodynamic monitoring on beta-blockers alone.

The strategy of combination therapy without haemodynamic monitoring was associated with greater or equal life expectancy and lower costs which ‘dominated’ the strategies of beta-blockers alone, endoscopic variceal ligation, beta-blockers with one HVPG measurement and combination therapy with one HVPG measurement. Compared with beta-blockers with two HVPG measurements, combination therapy with two HVPG measurements was more effective (2.28 vs. 2.27 years survival) and cheaper ($13 605 vs. $13 826). Therefore, the incremental cost-effectiveness ratio of this latter strategy compared with combination therapy without haemodynamic monitoring was $136 700 per year of life saved and $51 262 per episode of re-bleeding prevented.

Sensitivity analyses

The results of the model were robust in most one-way sensitivity analyses, yet variables emerged which were important in changing the overall results (Table 4). Several variables were found to influence significantly the incremental cost-effectiveness ratio of the two most effective strategies (combination therapy and combination therapy with two HVPG measurements). Extension of the time horizon was particularly influential and resulted in a dramatic decline in the incremental cost-effectiveness ratio (Figure 2). At a time horizon of 1 year, the incremental cost-effectiveness ratio was $2274 000 per year of life saved, whereas the ratio decreased to $21 000 per year of life saved at 5 years. An increase in non-variceal bleeding-related mortality resulted in a modest effect on cost, with a decrease in the incremental cost-effectiveness ratio from $225 000 per year of life saved to $133 000 per year of life saved when the non-variceal bleeding mortality was reduced from 30% to 2% per year. Exaggeration of the cost of hospitalization for variceal bleeding from $2500 per episode to $50 000 per episode resulted in a decline in the incremental cost-effectiveness ratio from $178 000 to $38 000. Of course, the cost of HVPG measurement was also influential. At costs of $500 and $1000 per procedure, the cost-effectiveness ratios were $15 000 per year of life saved and greater than $50 000 per year of life saved, respectively.

Table 4.  Results of one-way sensitivity analyses
VariableBaseline valueValuesICER ($/life expectancy)*
Combination therapy with haemodynamic monitoring
(2 HVPG) vs. combination therapy alone
  • HVPG, hepatic venous pressure gradient; ICER, incremental cost-effectiveness ratio.

  • Rounded to the nearest thousand.

  • † 

    Strategy strongly dominated (more expensive and less effective).

Probabilities
Probability of haemodynamic response (2 HVPG)0.510.10Dominated
0.6087 000
Re-bleeding on combination therapy given a haemodynamic response (20% reduction)0.040.0161 000
0.10Dominated
Non-variceal bleeding mortality0.070.025105 000
0.30220 000
Costs
Cost of variceal bleeding hospitalization15 1105000156 000
35 00072 000
Cost of HVPG measurement250050015 000
5000295 000
Figure 2.

One-way sensitivity analysis of the time horizon. ICER, incremental cost-effectiveness ratio.

The strategy of beta-blockers with two HVPG measurements was associated with lower costs and greater effectiveness than combination therapy with two HVPG measurements only as long as the probability of achieving a haemodynamic response to combination therapy was greater than 49%.

Two-way sensitivity analyses were performed on selected variables. Figure 3 demonstrates the effect of varying the time horizon and the cost of HVPG simultaneously. This figure demonstrates that combination therapy with two HVPG measurements was associated with an incremental cost-effectiveness ratio of less than $50 000 per year of life saved at a low cost of HVPG and as the time horizon was extended.

Figure 3.

Two-way sensitivity analysis of the time horizon and cost of the hepatic venous pressure gradient (HVPG) measurement.

In order to address uncertainty, we performed 1000 trials through a first-order Monte Carlo simulation. Haemodynamic monitoring with two HVPG measurements was more expensive and more effective in 69% and 80% of the trials, respectively. In trials in which haemodynamic monitoring was both more costly and more effective, the median cost-effectiveness ratio was $133 360 per year of life saved (5th and 95th percentiles, $12 687 and $197 740, respectively). At 3 years, the proportions of trials beneath the $200 000, $100 000 and $50 000 willingness-to-pay thresholds were 98%, 45% and 19%, respectively.

Discussion

The purpose of our model was to evaluate the cost and efficacy of routine HVPG measurement to guide secondary prophylaxis of recurrent variceal bleeding. In our model, the two most effective strategies were combination therapy alone and combination therapy with two HVPG measurements. Due to a slightly lower re-bleeding rate, combination therapy was found to be both cheaper and more efficacious than beta-blocker therapy alone or endoscopic variceal ligation alone. Both strategies which included only one HVPG measurement were found to be less efficacious than combination therapy alone. Specifically, combination therapy with one HVPG measurement identified the 14% of patients in whom an absolute decrease to less than 12 mmHg was achieved, but the additional 37% of patients who experienced a 20% reduction in portal pressure were incorrectly classified as ‘non-responders’ and underwent endoscopic variceal ligation.

In the beta-blocker strategy, the proportion of patients in whom HVPG < 12 mmHg was achieved was very similar to that observed in the combination therapy group (12% and 14%, respectively). However, the proportion of patients misclassified as ‘non-responders’ was much less with beta-blockers. The proportion of patients who achieved a 20% reduction in HVPG in the beta-blocker group was significantly lower than that in the combination therapy group (24% vs. 37%, respectively). Therefore, a smaller ‘protected’ group would be missed with a single HVPG measurement and fewer patients would be switched to an inferior strategy (endoscopic variceal ligation). The net result would be that, amongst patients on beta-blockers, even one HVPG measurement would result in improved survival compared with no haemodynamic monitoring.

On the other hand, both strategies that included two HVPG measurements demonstrated a slightly improved survival. In these strategies, pharmacotherapy was only replaced by endoscopic variceal ligation in the 50% of patients who did not experience a drop in HVPG to < 12 mmHg or by ≥ 20% of the baseline value. By the use of two HVPG measurements, only patients who did not experience a haemodynamic response by either criterion were elected for alternative prophylaxis with endoscopic variceal ligation.

Combination therapy with two HVPG measurements was expensive. However, it became cost-effective at 1 year compared with standard prophylaxis with combination pharmacotherapy by the manipulation of several variables. As the time horizon was extended, the extra cost of the initial HVPG measurement was diluted, resulting in an incremental cost-effectiveness ratio of less than $50 000 after ∼ 3.5 years. As the cost of HVPG measurement was decreased, this strategy crossed the incremental cost-effectiveness ratio threshold at $600 per HVPG measurement. Combination therapy with two HVPG measurements also became more attractive as the rates of recurrent variceal bleeding and death from bleeding increased. For example, in a ‘best-case’ scenario, biased towards an HVPG measurement strategy by incorporating a low cost to HVPG measurements ($500 vs. $2500 utilized in the base-case analysis) and a high probability of re-bleeding in patients in whom a haemodynamic response was not performed or identified (15% vs. 12.5%), the cost-effectiveness ratio of this strategy could be as low as $5000 per year of life saved. On the other hand, in a ‘worst-case’ scenario, biased against HVPG measurements by increasing the cost of HVPG ($3000 vs. $2500) and decreasing the probability of re-bleeding in non-monitored patients (i.e. 10% vs. 12.5%), prohibitively high ratios ($300 000 per year of life saved) were observed. Therefore, the identification of variables that influence the cost-effectiveness of routine HVPG measurements allows for the postulation of specific groups of patients with variceal bleeding who would benefit most from this management approach.

Two subsets of patients with a history of variceal bleeding were identified as having the most potential for benefit from aggressive prophylaxis: patients with an extended life expectancy and patients at high risk for recurrent variceal bleeding. Patients in whom life expectancy is expected to be prolonged due to compensated cirrhosis or non-cirrhotic portal hypertension may benefit more from HVPG measurement because of the early selection of the most efficacious prophylactic strategy with the distribution of the cost of the initial HVPG procedure over a greater number of years. Patients at high risk for variceal re-bleeding are also better candidates for more aggressive prophylaxis and may be identified by methods other than the estimation of portal pressure. Variceal morphology has been shown to be highly predictive of variceal bleeding in a key study by the North Italian Endoscopic Club.34 In this study, factors such as variceal size, ‘red weal markings’ and ‘cherry-red spots’ were used in combination with the Child–Pugh class to create a highly predictive index for calculating the initial bleeding risk. In addition to variceal morphology, variceal pressure measured with an endoscopic probe has been shown in several studies to provide useful prognostic information for both initial variceal bleeding and recurrent bleeding.25, 35, 36 Although routine HVPG measurement in the secondary prophylaxis of variceal bleeding is expensive in our model, this strategy is more likely to be cost-effective in the setting of an extended life expectancy and/or a high risk of recurrent bleeding.

Although the above conclusions can be drawn from our analysis, several limitations of the model were identified. A review of the pattern of recurrent variceal bleeding revealed that a substantial portion of re-bleeding occurs during the first 6 weeks after the initial bleed, and the rate of re-bleeding is not linear over time.37 This pattern makes the manipulation of the time horizon less valid and may result in a decreased influence of serial HVPG measurement on therapy due to a substantial portion of re-bleeding occurring prior to the initial measurement at 12 weeks. To address this observation, we modified our model to reflect a higher re-bleeding rate at 6 weeks after the index bleed. Even though this did not significantly change the overall results at 3 years, it did slightly improve the cost-effectiveness of HVPG monitoring (data not shown).

The current HVPG measurement technique, using a single hepatic vein measurement, has recently been disputed as a reliable indicator of portal pressure in patients with cirrhosis due to a reported variation in measurements between different hepatic veins.38 However, most of the patients in this study with significant variation in their HVPG measurements had gradients of < 10 mmHg, and only 13% of patients had measurements on both sides of the 12 mmHg threshold.

The main identified limitation of our model was a lack of exhaustive data for the calculation of all probabilities for recurrent variceal bleeding, haemodynamic response to pharmacotherapy, intolerance to pharmacotherapy, complications due to endoscopy, death from variceal bleeding and death from cirrhosis. Subset analysis of compensated and decompensated cirrhotics, as well as patients with non-cirrhotic portal hypertension, could not be performed due to a lack of adequate data, which may have identified a selective group of patients who would have benefited the most from HVPG measurements.

Finally, although this model demonstrates a reduction in overall mortality due to a reduction in variceal bleeding events, multiple clinical trials of successful variceal bleeding prophylaxis have failed to demonstrate a statistically significant decrease in overall mortality, as would be expected with a decrease in variceal bleeding rates. It is noteworthy that none of these studies has adequate statistical power to detect differences in survival, but meta-analyses have demonstrated a trend towards improved survival by preventing episodes of variceal bleeding, especially when prophylaxis is compared with placebo or ‘non-active treatments’.1 Given these small differences in survival (years of life saved) between strategies, which lead to high cost-effectiveness ratios, an alternative measure of effectiveness that incorporates both the quality of life as well as survival, such as quality-adjusted life years, may have been utilized. Conceivably, preventing death and the detrimental effects on the quality of life from variceal bleeding may have resulted in more favourable cost-effectiveness ratios of strategies that incorporate HVPG monitoring. Unfortunately, patient-derived health-state utilities associated with specific complications of cirrhosis are lacking, and those available have been derived from physicians and may not accurately reflect patient preferences.

Despite these shortfalls, the identification of patients who fail to respond to pharmacotherapy continues to appear to be beneficial. Screening of these patients for poor response by HVPG measurement or another means of testing is warranted. Doppler ultrasound is currently being investigated for the estimation of portal pressure by both endoscopic39 and transabdominal40 approaches, with promising data on the identification of portal hypertension, but a lack of capability for calculating changes in portal pressure. Several studies have demonstrated a correlation of variceal pressure with variceal bleeding rates.25, 35, 36 However, this method has not been as well studied as HVPG measurement and has not been used to evaluate long-term changes in pressure in response to standard pharmacotherapy for bleeding prophylaxis. After consideration of these other modalities for the objective estimation of variceal bleeding risk, HVPG measurement remains the gold standard.

This model of the cost-effectiveness of HVPG measurement to guide secondary prophylaxis for variceal haemorrhage justifies further investigation of this technique as a tool to decrease morbidity and mortality from variceal bleeding. Although this strategy appears to be expensive, additional research is warranted until portal pressure can be estimated by an alternative technique or pharmacotherapy associated with a more uniform decrease in portal pressure is instituted.

Acknowledgements

M. R. Arguedas received a 2002 Lister Hill Center for Health Policy Intramural Grant.

Ancillary