Cost-Effectiveness of Stress Ulcer Prophylaxis: Role of Proton Pump Inhibitors
Proton pump inhibitors are cost-effective options for stress ulcer prophylaxis.
There is increasing emphasis on incorporating cost-effective, evidence-based practices into critical care. For instance, histamine (H2) blockers were shown to be more efficacious than sucralfate for stress ulcer prophylaxis (SUP). Accordingly, older practice guidelines recommended H2 blockers for SUP in high-risk critically ill patients.1 However, the literature on SUP has been rather discordant, in part due to variations over time in clinical practice, drug availability, and differences in study criteria and analysis of end points.2 Several reviews, meta-analyses, and cost-effectiveness analyses have tried to reconcile these issues.3-9 Consensus may have emerged regarding the appropriate end point: prophylaxis should aim to prevent clinically signifi cant gastrointestinal bleeding (CSGIB), meaning bleeding that causes either decreases in hemoglobin that require transfusion or hemodynamic instability, rather than just any overt bleeding.3,4,6
Proton pump inhibitors (PPIs), the newest class of antiacid drugs, have not been included in any meta-analyses for SUP. Consequently, although demonstrably superior to H2 blockers for sustained elevation of gastric pH, PPIs were not initially recommended in practice guidelines because of the paucity of direct comparisons with H2 blockers in patients with SUP.1,10 Nonetheless, some researchers proposed and described socalled off-label use of PPIs in SUP.2,11-13 With newer data, guidelines have since been updated to include PPIs for SUP.14 Given the rising demand for limited healthcare resources, it is worthwhile to include PPIs when reanalyzing the cost-effectiveness of options for SUP. Accordingly, the purpose of this study was to determine whether there are differences among select PPIs, H2 blockers, and sucralfate in terms of cost-effectiveness that could infl uence prescribing practices for SUP.
We developed a decision tree to model the probability of CSGIB (pCSGIB) in high-risk patients, without and with prophylaxis, using 1 of 7 drug regimens. Excel (Microsoft, Redmond, WA) and TreeAge software (TreeAge Software, Inc, Williamstown, MA) were used to construct the model. In keeping with previously published costeffectiveness studies, parameters included the following estimates: acquisition costs of the drugs, consumables and labor for administration, pCSGIB, cost of managing a case of CSGIB, probability of side effects, and costs of managing side effects associated with each regimen (
Options selected for analysis included the only 2 drugs specifically approved by the US Food and Drug Administration (FDA) for SUP (IV cimetidine and immediate release [IR] omeprazole). Enteral and IV famotidine and lansoprazole suspension were included because of their growing off-label use just as cimetidine seemed to have been abandoned for SUP.3,15 Pantoprazole, the first IV PPI available in the United States, was included, as was sucralfate, which had earlier been found to be the most cost-effective option for SUP.7-9 Dosages used were cimetidine 300 mg IV every 6 hours; famotidine 20 mg by mouth or nasogastrically (NG) every 12 hours; famotidine 20 mg IV every 12 hours; lansoprazole 30 mg suspension daily; pantoprazole 80 mg IV twice daily; IR omeprazole 40 mg the first day, then 20 mg daily; sucralfate 1 g by mouth or NG every 6 hours. Each regimen was evaluated in the model for a hypothetical 7-day course.
The direct costs of acquiring and dispensing drugs for SUP, as well as adverse financial consequences from any CSGIB, are incurred by hospitals. As such, the hospital perspective was adopted for analysis, with 2005 as the base year. Costs and benefits occur over a short period so neither was discounted, in line with existing recommendations.16 Baseline analysis involved estimation of incremental costs (total costs less savings attributable to each regimen) and incremental effects (cases of CSGIB avoided). Incremental cost-effectiveness ratios (ICERs) were calculated as a measure of additional cost incurred for each extra case of CSGIB averted by each strategy compared with the others. The strategy with the lowest ICER was considered to be most cost-effective (ie, it represented the greatest value compared with the other strategies). One-way, 3-way, threshold, and probabilistic sensitivity analyses were performed to assess the robustness of the results.16-18
Identification of Costs and Benefits
Data for the model were obtained primarily from a systematic review of published literature in peer-reviewed journals. When data were unavailable, alternate sources used were expert opinion and data from the Departments of Pharmacy, Pathology, and Nursing and the Division of Surgical Intensive Care, Department of Anesthesiology, at the University of Iowa Hospitals and Clinics, Iowa City. Studies reviewed were identified by a computer search of Medline and Web of Science. The search key words included [cimetidine], [famotidine], [sucralfate], [lansoprazole], [omeprazole], [pantoprazole], [stress ulcer], [cost-effectiveness], [prophylaxis], [intensive care unit], and [gastrointestinal bleeding]. Criteria for inclusion were use of randomized controlled study design and pCSGIB as one of the end points of the study, or the ability to determine pCSGIB from reported data. All relevant meta-analyses also were reviewed. Values were preferentially extracted from the meta-analyses or the more recent large trials if applicable, unless lower values for pCSGIB had previously been reported by studies that otherwise met the inclusion criteria.7 Baseline and favorable and unfavorable boundary values identified for each variable are presented in
For the baseline analysis, the pCSGIB with no prophylaxis was assumed to be 6%.9 Sucralfate has been studied in many randomized clinical trials (RCTs), so the most favorable value (4.4%) from meta-analyses was cited.4 The pCSGIB for IR omeprazole was determined from a RCT in which it was shown to be at least equivalent to cimetidine19; the definition of CSGIB in this trial was less stringent than the one given above. Applying the stricter definition used in previous meta-analyses, the pCSGIBs were determined to be 3.2% and 3.4% for IR omeprazole and cimetidine, respectively. These are comparable to a 3% pCSGIB reported with famotidine.11 Published data for lansoprazole are limited, but reports suggest that there are no intraclass differences in acid-suppressing efficacy among PPIs.20,21 Accordingly, given the lack of any data demonstrating superiority of either PPIs or H2 blockers in the treatment of SUP, we assumed that IV pantoprazole, famotidine, lansoprazole, and IR omeprazole all had equal efficacy in SUP.
To ensure consistency and avoid extreme values due to peculiar local experiences, the probability of side effects resulting from the different drugs was extracted from the FDA-approved drug package inserts, as published in the Physicians’ Desk Reference.22 Most side effects are rare, so we included them only if they met 1 or both of the following conditions: (1) they were used in previous cost-effectiveness analysis and/or (2) the probability of the side effect occurring exceeded one-tenth of 1%. The probability of clogging of enteral tubes by sucralfate was extracted from the literature.7 Because the association of drugs used for SUP and nosocomial pneumonia appears to be weak and remains controversial, pneumonia was not included as a side effect in this study.3
Sources for cost data are shown in Table 1. Again, to maintain consistency and facilitate reproducibility, acquisition costs of drugs were based on the best prices found in the 2005 edition of the widely used Red Book, an annual industry publication of wholesale drug prices.23 Given higher prices for commercial preparations of enteral PPI suspensions, we assumed that these would be compounded locally by hospital pharmacies as previously described.7 Estimates of labor costs (pharmacy and nursing time) for the preparation and administration of the different regimens were partly obtained from published data as well.12 The costs of treating a case of CSGIB and the costs of treating drug side effects were obtained from American Society of Health-System Pharmacists guidelines, validated by cost calculations from the University of Iowa Hospitals and Clinics (personal communication, Departments of Anesthesiology, Pharmacy, and Pathology, University of Iowa Hospitals and Clinics, Iowa City).7 Costs were adjusted to the year 2005 using a Bureau of Labor statistics conversion factor.24
Baseline analysis identified the ICER for each strategy by using the baseline values for each variable. To test the robustness of these results to uncertainty in the variables, a series of 1-way sensitivity analyses were used to generate a tornado (influence) diagram. We assumed that the willingness to pay (WTP) value to avoid 1 case of CSGIB would be up to twice the cost of treating 1 case of CSGIB. A range of ±25% was applied to the baseline costs to identify favorable and unfavorable values for the influence diagram. The cost of evaluating mental status changes had a wider (±75%) range applied, owing to the need to rely on expert opinion from the University of Iowa Hospitals and Clinics because limited published data were available. The variables found to be most influential on the uncertainty within the model were further evaluated to identify threshold points (values of variables at which the choice of optimal strategy would change). Uncertainty was further tested using a 3-way sensitivity analysis to identify the impact of simultaneous extreme changes in the values of multiple variables.
These all were deterministic sensitivity analyses, yet it is likely that the parameters would vary among different patient populations in different settings, thus affecting the results. Accordingly, we also performed probabilistic sensitivity analysis to evaluate the joint effects of secondorder parameter uncertainty on the results. A Bayesian approach was used to apply probability distributions to the most influential variables: triangular distributions (bounded by a range of ±25%) were selected for drug acquisition costs, as these were bound to be positive and were unlikely to vary much from the best wholesale prices.23 Because pCSGIBs for the different drugs are bounded on the interval 0-1, beta distributions were assumed for those values, with boundaries set using a wider range of ±75% of baseline, given the greater imprecision in the baseline pCSGIBs reported in the literature. Estimates of costs and effects were drawn randomly from these distributions to simulate a hypothetical cohort, better reflecting likely real-world results.17,25 Such stochastic analyses produce different mean ICERs and confidence intervals each time the set of Monte Carlo simulations are run; however, with a sufficiently large number of simulations, the results converge in terms of the relative proportions where each strategy would be the optimal option. We ran 10,000 simulations to generate mean ICERs and cost-effectiveness acceptability curves for each strategy in relation to a range of WTP values.
At baseline values, IR omeprazole was most cost-effective at $12,390.77 per case of CSGIB avoided (
). All other strategies were dominated (ie, they were more expensive and less effective). Lansoprazole suspension was the next most cost-effective option ($13,043.78). The tornado diagram in
showed that results were highly sensitive to pCSGIB with and without SUP (48%) and drug acquisition costs (41%). Further analysis revealed that if the pCSGIB with no SUP was less than 3.5%, none of the drugs would be cost-effective at the designated WTP level of $16,000. Surprisingly, labor costs (nursing and pharmacy time) and the cost of treating an episode of CSGIB together accounted for less than 1% of the model uncertainty. Subsequent 3-way sensitivity analyses varying nursing, pharmacy, and drug acquisition costs simultaneously did not influence the result.
Finally, the probabilistic sensitivity analysis did not alter the choice of optimal strategy with the remaining 3 top-ranked options. However, lansoprazole suspension, although less cost-effective than IR omeprazole, was not dominated as it was in the baseline analysis. Nonetheless, the ICER for lansoprazole suspension ($109,223.80) far exceeded our WTP threshold; therefore, ordinarily it would not be recommended. The acceptability curves of the probabilistic sensitivity analysis showed that within a range of WTP values of $16,000 ± 50%, IR omeprazole, lansoprazole suspension, and enteral famotidine combined were the optimal options in 31%, 30%, and 20% of the simulations, respectively (approximately 80% of the simulations overall).
These analyses support a role for PPIs in preventing stress-related gastrointestinal hemorrhage. They further suggest that the enteral formulations of the PPIs are the most cost-effective class of drugs for SUP, followed by H2 blockers (in particular, enteral and then IV famotidine). The dominant strategy was treatment with IR omeprazole, closely followed by lansoprazole suspension and then enteral famotidine.
As equal efficacy was assumed for all 3 regimens, results can only be attributed to the total costs of the drugs (acquisition plus administration costs) and their side-effect profiles. As demonstrated in the baseline and sensitivity analysis, IV cimetidine, sucralfate, and IV pantoprazole were the least cost-effective options, no doubt due to their peculiar side-effect profiles (cimetidine and sucralfate) and acquisition cost (pantoprazole). Moreover, the cost-effectiveness acceptability curves identified enteral PPIs and/or H2 blockers as the most cost-effective options more than 80% of the time.
Our results do differ from some older cost-effectiveness studies that found sucralfate to be the most cost-effective option.7-9 In part, that is likely because we used updated data for drug effectiveness based on the most recent meta-analyses. We assumed equal efficacy only for H2 blockers and PPIs, based largely on the results of recent RCTs.11,19 Furthermore, acquisition costs do change as generic drugs become available; consequently, we used as our baseline the best wholesale prices available. The fact that the results are sensitive to drug acquisition costs and, to a lesser extent, to labor costs suggests that the option chosen would depend on the particular circumstances of different institutions. For instance, with volume discounts or group purchasing contracts, some hospitals may obtain different drugs at lower prices than the published average wholesale price. At the University of Iowa, for example, the combined daily acquisition and administration cost of lansoprazole suspension is only $0.24 (personal communication, Department of Pharmacy, University of Iowa Hospitals and Clinics, Iowa City), indicating that for this institution (assuming all other values remain constant), lansoprazole suspension would be the most cost-effective choice.
Similarly, larger facilities could achieve economies of scale in their pharmacy and so reduce their unit costs for the different drugs, thus altering the cost-effectiveness rankings. For instance, whereas administration of IR omeprazole is more nursing-intensive, preparation of lansoprazole suspension is more pharmacy-intensive. As such, higher nursing costs (unfavorable) or lower pharmacy costs (favorable) could be expected to identify lansoprazole suspension as the optimal strategy over IR omeprazole. Our analysis did assume that a pharmacist would be solely responsible for the preparation of lansoprazole suspension. In larger hospitals, lower-cost pharmacy technicians might assist with the preparation, thus substantially reducing labor costs in favor of lansoprazole suspension. As noted above, however, the 3-way sensitivity analysis of pharmacy, nursing, and drug acquisition costs did not change the baseline results.
Interestingly, our literature review showed an apparent decline in the reported incidence of CSGIB (albeit mostly in a research context), perhaps due to overall advances in critical care.2 Some even have suggested that SUP may now be unnecessary.26,27 Indeed, our results were sensitive to the pCSGIB without SUP. This was important because the benefits of avoiding a case of CSGIB depend on the difference between the pCSGIB with SUP and the pCSGIB without SUP. A lower pCSGIB without SUP decreases the likelihoodof avoiding a case of CSGIB, thus raising the cost per case of CSGIB avoided. Our results indicate that SUP would remain cost-effective if the pCSGIB without SUP is higher than 3.5%.
It is unclear how much of the apparent decline in pCSGIB is attributable to a true decrease in the incidence versus merely reflecting the meticulous attention to detail characteristic of clinical studies. Moreover, considerable variability exists in the quality and staffing of intensive-care units; the quality of care may not always reflect the idealized circumstances of clinical studies. Therefore, it is too soon to declare that the natural incidence of CSGIB is now so low as to render SUP redundant.
Our assumption that acid-suppressing drugs are not associated with an increased rate of pneumonia deserves further comment. Our model did not include the cost of pneumonia because the literature remains unclear on this issue. Several meta-analyses have examined the relationship between elevated gastric pH from SUP and pneumonia, with conflicting results. Messori et al found that neither ranitidine nor sucralfate compared with placebo had any influence on nosocomial pneumonia, although relative to each other, ranitidine significantly increased the incidence of pneumonia.3 Similarly, Cook et al from Canada found no statistical difference, but a trend toward increased rates of pneumonia with ranitidine relative to sucralfate.28 Conversely, a multicenter (nonrandomized) study by the French ARDS Study Group found an increased rate of pneumonia with sucralfate during mechanical ventilation for acute respiratory distress syndrome.29 Moreover, sucralfate may even increase the risk of acid aspiration and pulmonary edema.7 Clearly this issue is unresolved. However, PPIs and H2 blockers are undoubtedly superior to sucralfate in raising gastric pH and reducing the incidence of CSGIB.1 Even if raised gastric pH were linked to nosocomial pneumonia, that relationship would be seen with both PPIs and H2 blockers, in which case the choice of optimal strategy would not be altered.
Our study does have some limitations. The main one is that values for the variables were extracted from multiple publications. Also, few RCTs addressed the effectiveness of SUP with PPIs. Even fewer included an arm without SUP. Ethical considerations make it unlikely that future studies would simultaneously compare different drug regimens with a placebo arm. We attempted to account for these deficiencies in the data by using a wide range of values during sensitivity analysis. Tolerance does develop during therapy with H2 blockers, and sucralfate adversely affects absorption of other enterally administered drugs. However, without precise data about the incidence and severity of these problems, we could not account for these clinical considerations, although they may impact the choice of agent for SUP. Given the results obtained, these data would not have changed the outcome.
Given current clinical practice, and until the pCSGIB without prophylaxis is consistently low enough, SUP remains cost-effective and recommended. Our findings demonstrate that when appropriate for the individual patient, the most cost-effective strategies for SUP are enteral drugs, particularly the PPIs IR omeprazole and lansoprazole suspension, followed by the H2 blocker enteral famotidine. Based on these results, guidelines should reflect the utility of enteral PPIs as cost-effective options for SUP compared with H2 blockers and IV drugs.