EID Home | Ahead of Print | Past Issues | EID Search | Contact Us | Announcements | Suggested Citation | Submit Manuscript

Comments

Dispatch

Cost-Benefit of Stockpiling Drugs for Influenza Pandemic

Ran D. Balicer,*† Michael Huerta,* Nadav Davidovitch,*† and Itamar Grotto*†
*Ministry of Health, Jerusalem, Israel; and †Ben-Gurion University of the Negev, Be'er-Sheva, Israel

Appendix 1. Cost-Benefit of Stockpiling Antiviral Agents for Influenza Pandemic

Detailed Methods, Data, and Results

We used a static spreadsheet model (Microsoft Excel 2000, Microsoft Corporation, Redmond, WA, USA) to estimate the effect of pandemic influenza on the Israeli population. We defined 3 separate strategies for the use of antiviral drugs during a pandemic, analyzed the effect of these strategies on pandemic outcomes, and estimated the economic consequences of each scenario.

Health-related Outcomes of the Pandemic

We divided the population (6,748,000 in Israel at the end of 2003) into 3 age categories: ≤18 years, 19–64 years, and ≥65 years (1). Each age category was further grouped by low- or high-risk for serious complications of influenza infection by using the US population age-specific proportions (2). When we adjusted for the age structure of the Israeli population, the high-risk group numbered ≈899,000 persons (14% of total population). Point estimates and ranges of health-related variables used in the base-case model are detailed in Appendix Table 1.

We constructed a baseline nonintervention scenario by using age- and risk-specific rates to estimate the expected numbers of patients, physician visits, hospitalizations, deaths, and lost work days (2) and calculated the economic value of each of these outcomes. These outcomes were then recalculated for each intervention strategy, as described in detail below, yielding the economic benefit associated with each strategy. The resulting figure was then compared to the costs associated with purchasing enough antiviral stocks to implement the strategies.

Economic Outcomes

Point estimates of economic variables used in the base-case model are detailed in Appendix Table 2. Cost of patient visits (including cost of prescription drugs and diagnostic tests) was based on current data provided by a major healthcare organization in Israel (3). Hospitalization and workday costs were provided by the Israeli Central Bureau of Statistics (1). The benefit of each strategy was defined as the cost savings it yielded, relative to nonintervention. We calculated both direct costs to the healthcare system and overall costs to the economy, the latter including the value of lost workdays but not indirect costs from excess deaths.

According to previous assessments by Meltzer et al. that addressed a similar scenario (2), the value of lives lost made up 83% of all estimated costs of pandemic influenza. When applying these estimates to our study, the addition of the value of lives lost increases our calculated cost-benefit ratios >6-fold.

Drug Selection and Costs

Drug resistance may appear in approximately one third of patients treated with the M2 inhibitors amantadine or rimantadine (4,5), and early rimantadine resistance has been described when this drug was used for treatment and postexposure prophylaxis in families (5). In vitro susceptibility testing of recent human H5N1 isolates from east Asia indicate that this strain is resistant to the antiviral drugs amantadine and rimantadine but susceptible to the neuraminidase inhibitor oseltamivir (6). Furthermore, oseltamivir has been shown to be effective against H5N1 and H9N2 viruses in mice (7). M2 inhibitors have several other major disadvantages, including higher rates of adverse effects and no proven efficacy for reducing influenza complication rates. Of the 2 currently available neuraminidase inhibitors, oseltamivir and zanamivir, only oseltamivir is licensed for prophylaxis in Israel (8), the United Kingdom (9), and the United States (10). We therefore limited our analysis to oseltamivir, at a daily dose of 75 mg when used for prophylaxis (11) and 150 mg when used for treatment (12). Per diem oseltamivir costs were calculated for each strategy by using the projected number of neuraminidase inhibitor recipients and drug prices quoted in March 2004 by the manufacturer's representative in Israel (Roche Pharmaceuticals [Israel] Ltd, pers. comm.). Parameters related to drug costs are detailed in Appendix Table 2. We based our calculations on a 10-year shelf life for the drug, when stored as bulk active powder, and discounting was performed by using the locally accepted annual rate of 3%. The cost-benefit ratio of each strategy was calculated by dividing strategy-specific drug costs by treatment-derived economic benefits. Ratios were calculated separately for direct health-related costs and for overall costs to the economy. Since the different strategies were not mutually exclusive, incremental cost-benefit analysis was not performed.

Probability of a Pandemic

We adjusted all cost-benefit outcomes for the estimated probability of a pandemic occurring per year. We based our calculations on the recent historic incidence of 3 influenza pandemics over the last century, thus adopting a conservative point estimate of 1 pandemic every 33 years for the base-case and applied a wide range of estimates for sensitivity analyses.

Nonintervention

The baseline scenario, modeled by using estimates derived by Meltzer et al. from previous pandemics (2) and data collected in Israel during interpandemic periods (3), provides an estimate of health-related and economic outcomes that would be expected were the pandemic allowed to run its natural course. These estimates by Meltzer et al. are based on a wide range of possible attack rates (15%–35%). Illness and death rates in these estimates are considerably lower than those estimated for the pathogenic potential of the currently circulating H5N1 avian strain. These estimates were selected to be in line with our general approach to underestimate the pandemic's effect and potential benefits provided by each of the interventions, as explained below. The parameters used for these calculations are detailed in Appendix Tables 1 and 2, and the formulas are detailed in Appendix 2. This scenario serves as a reference category against which the alternative strategies can be compared.

Intervention Strategies

We assumed that strain-specific vaccine would be unavailable during the initial months of a pandemic. If appropriately stockpiled, antiviral drugs can be directed either at therapy or prophylaxis. Prophylactic strategies can be divided into long-term preexposure prophylaxis and short-term, epidemiologically directed postexposure prophylaxis. We compared each of 3 strategies with nonintervention, alternately targeting either the entire population or only populations at high risk for complications.

Strategy 1: Therapeutic Antiviral Use

When given therapeutically to influenza patients within 48 hours of symptom onset and if continued for 5 days, neuraminidase inhibitors can reduce the duration of clinical symptoms by an average of 1 day (13), hospitalization rates by 59% (14), and antimicrobial drug use by 63% (13) (Appendix Table 3). Since the effects of treatment on reducing death rate have not yet been studied or quantified, we have not included in our model this or other yet-unproven potential benefits of these drugs, in keeping with our approach toward underestimating any benefits of intervention. Moreover, effects on death rate reduction would not alter our cost-benefit calculations, since only illness-related costs were taken into account.

We evaluated 2 variations of therapeutic antiviral use: nonselective treatment available to all patients (strategy 1A) and selective treatment limited to use in patients at high risk (strategy 1B). We adjusted this model for treatment initiation rates, since not all patients would be expected to reach a physician and initiate treatment within the 48-hour window of therapeutic opportunity. The proportion of patients likely to seek physician care was estimated by using data available from studies of interpandemic influenza in Israel (3), and the wide range of estimates was used in the sensitivity analysis to address treatment-seeking behavior during a crisis.

Strategy 2: Preexposure Prophylaxis

We evaluated 2 variations of preexposure antiviral use: mass prophylaxis made available to the entire population (strategy 2A) and selective prophylaxis, limited to groups at increased risk for complications (strategy 2B). We assumed a pandemic duration of 50 days (17) and calculated drug costs accordingly.

We performed a systematic review and meta-analysis of studies evaluating the protective efficacy of neuraminidase inhibitors when used for preexposure (seasonal) prophylaxis. The methods and results of this meta-analysis are described below.

A computerized search was conducted by using MEDLINE (January 1966–December 2004) and Embase (January 1980–December 2004) databases. The following combination of keywords was used: (influenza) and ([oseltamivir or Tamiflu] or [zanamivir or Relenza]) and (prevention or prophylaxis or chemoprophylaxis). This search was limited to articles published in English. In addition, we searched these databases using the names of authors of studies identified in the primary search and in the studies' reference sections. We also contacted drug companies for information on unpublished trials.

Our search identified 24 candidate papers. These papers were then independently reviewed by 2 of the authors (R.D.B. and I.G.), 1 of whom was blinded to authors' names, journal, date of publication, and site of study. Of the candidate papers, we selected randomized, controlled, double-blind trials that met all of the following criteria: 1) evaluated preexposure (seasonal) prevention of naturally occurring influenza with zanamivir or oseltamivir (oseltamivir 75–150 mg/day, zanamivir 10 mg/day); 2) included a study sample whose participants were 18–69 years of age; 3) included a healthy, community-based population without specific baseline disease; 4) provided data on intent-to-treat analysis; and 5) reported the incidence of laboratory-confirmed influenza in placebo and control groups. Two studies met these inclusion criteria (11,15). Appendix Table 4 summarizes the main features of the studies included in the analysis.

The 2 authors who reviewed the papers also abstracted information from each of the selected studies. In cases in which >1 article was published on the same study, all articles were assessed for data consistency. All data were abstracted by using a standardized protocol and computerized report form. Reported relative risk (RR), or incidence data necessary for computing RR, were abstracted based on intent-to-treat analysis.

For all studies included in the analysis, cases were defined as clinical influenza confirmed by isolation of influenza virus, by reverse-transcriptase polymerase chain reaction (RT-PCR), or by testing paired serum samples for rise in antibody titer against circulating influenza virus. For each study, we calculated the RR and 95% confidence interval (CI) for influenza infection in the intervention group compared to the control group. We calculated the overall RR for preexposure prophylaxis. Since oseltamivir and zanamivir are both neuraminidase inhibitors and the results of the studies that evaluated these drugs were comparable, we combined the results of these studies in our overall estimate.

In each study, 95% CIs of the RR were calculated by using Breslow's method (18). Since the results of the individual studies were homogenous, the overall RRs and 95% CIs were calculated by using precision-based estimates, as described by Fleiss (19) and Kleinbaum et al. (20), which assume a homogeneity of effect between studies (fixed-effects model). The protective efficacy of the interventions was calculated as 1 – RR.

Heterogeneity of RRs across n studies was tested with the formula

χ² heterogeneity = ΣW_iM_i² – (ΣW_iM_i)² / ΣW_i

where M_i is an individual measure of association and W_i is a weighting factor equal to the reciprocal of the squared individual variance. Significance was evaluated with n – 1 degrees of freedom.

Since only 2 studies of preexposure prophylaxis were included, no sensitivity analysis was applied. All computations were performed by using PEPI software for epidemiologic analysis (21).

The results of each study and overall estimate of preexposure studies are presented in Appendix Table 5. The overall protective efficacy of this intervention was 71% (95% CI 57%–80%). This result was used as the point estimate for the protective efficacy of neuraminidase inhibitors when used in a setting of preexposure prophylaxis (Appendix Table 3).

Strategy 3: Postexposure Prophylaxis

In this scenario, stockpiled antiviral agents are administered as short-term prophylaxis to exposed close contacts in addition to treating the index patients, a strategy termed "ring prophylaxis" (22) or "targeted prophylaxis" (16) (strategy 3A). Only 1 study (16) published to date used dynamic mathematical modeling to examine the expected effectiveness of this pandemic control measure on a population level during a pandemic. This stochastic simulation model by Longini et al. suggested that during an influenza pandemic, postexposure prophylaxis targeted at close contacts (i.e., household, daycare centers, play groups, and schools) might prevent 36% of all infections while providing prophylactic antiviral courses to 54.8% of the population. We have adopted these results as point estimates, while allowing a wide range of values for our sensitivity analysis (Appendix Table 3).

While these estimates are the only ones published to date on the efficacy of this strategy during an influenza pandemic, the inherent limitations of stochastic models such as the one used by Longini et al. must be acknowledged when this strategy is considered in practice. The authors of that study modeled communities of 2000 people, with predefined mixing patterns between subpopulations and a relatively limited number of "outsiders" entering the population. The paucity of evidence-based data on contact and transmission probabilities regarding both epidemic and pandemic influenza, as well as the complexity of real-life mixing patterns within a large population, may lead to substantial alterations in the efficacy of targeted prophylaxis, which cannot be modeled. Longini et al. have shown that these estimates are sensitive to various (currently unpredictable) parameters of the population and the pandemic strain, such as population compliance, delay in treatment initiation, and the outbreak basic reproductive number (16,23). In addition, as this strategy requires relatively rapid use of large antiviral stocks, it may potentially lead to premature depletion of the stockpiles. This postexposure prophylaxis strategy is less likely to lead to such early depletion of antiviral stocks, when used in the first stages of the pandemic, as long as relatively few suspected cases are identified. In these first stages, longer prophylactic courses may be considered. As the scale of the outbreak increases, application of this strategy becomes more problematic, as it may lead to rapid consumption of stockpiled drugs, and should therefore be monitored closely.

Breakthrough Cases

Under strategies 2 and 3, clinical influenza would develop in a varying proportion of participants who received prophylaxis despite treatment (breakthrough cases). We assumed that these patients, although becoming ill, would nonetheless benefit from the neuraminidase inhibitors that they had been receiving, and they were credited with the effects of therapeutic neuraminidase inhibitor treatment, such as shorter duration of illness and fewer hospitalizations and deaths (see strategy 1).

Sensitivity Analyses

Very little evidence-based data are currently available to allow accurate predictions regarding the effect of the next influenza pandemic. This model, as well as similar published models that attempted to make inferences regarding an impending pandemic, is based on estimates derived mainly from sparse data on previous pandemics and on the characteristics of interpandemic influenza. We have systematically chosen the most conservative estimates available in the literature regarding the different parameters, but such estimates are still associated with much uncertainty. A series of sensitivity analyses was therefore conducted to establish the robustness of the various outcomes of this model. These analyses applied a wide range of values for the parameters relating to pandemic probability, health-related pandemic outcomes, and antiviral drug efficacy. We considered the economic parameters as relatively stable, as they are based on verifiable data. Appendix Tables 1 and 3 list the variables used for strategy outcome estimates and the ranges used for sensitivity analyses.

The sensitivity analysis examined the effect that the modification of each parameter had on the main outcomes (cost-benefit ratios), thus assessing to which of the variables the model was most sensitive. This analysis showed that the most important parameters were the severity of the pandemic (illness and death rates) and the annual probability of a pandemic. Reducing the annual probability of a pandemic to 1 every 100 years, only the strategy of treating patients at high risk represented a cost savings, with a cost-benefit ratio of 1.23. When a combination of low-range estimates (according to Meltzer et al.) of the various health-related pandemic outcomes modeled were assumed, the cost-benefit ratios of therapeutic treatment and postexposure prophylaxis were decreased, but these strategies remained cost-saving, with cost-benefit ratios of >2.27 and 1.47, respectively. We then analyzed our data to define the ranges of cost-benefit ratios when applying various combinations of the parameters in Appendix Table 1. The results of these analyses are presented in Appendix Table 6. Only when we assumed an annual pandemic probability of 1 per 100 years, together with the most unfavorable sets of estimates for the other parameters, did all of the intervention strategies become non–cost-saving. However, if the annual probability of a pandemic remains >1 every 80 years, stockpiling antiviral drugs to treat patients at high risk remains consistently cost-saving, even when one assumes that the most unfavorable sets of estimates within the ranges applied for all other parameters. Since current events in Southeast Asia suggest that the probability of a pandemic may now be 2–3 times greater than our upper limit (1 every 10 years), cost-benefit ratios can be adjusted by multiplying current high-end estimates by the same factor. Additionally, health-related effects of an H5N1 pandemic may be similar or worse than those seen in 1918, with higher attack rates, complication rates, hospitalizations, and death rates. If, for example, under our assumed high-end attack rates, the rates of physician visits and hospitalizations are double our upper value, the cost-benefit ratios of therapeutic treatment of all patients and postexposure prophylaxis would be 6.55 and 3.12, respectively. We did not pursue higher ranges for these values, as our basic conclusion regarding the cost-benefit of currently purchasing antiviral drugs does not change.

Appendix 1 References

1. Statistical abstract of Israel 2003, no. 54. Jerusalem: Israel Central Bureau of Statistics; 2003.
2. Meltzer MI, Cox NJ, Fukuda K. The economic impact of pandemic influenza in the United States: priorities for intervention. Emerg Infect Dis. 1999;5:659–71.
3. Cohen Y. Tamiflu pharmaco-economic dossier, part C: evaluation of influenza cost of illness in Israel. Netanya (Israel): Quintiles Israel; 2000. p. 1–46.
4. Saito R, Oshitani H, Masuda H, Suzuki H. Detection of amantadine-resistant influenza A virus strains in nursing homes by PCR-restriction fragment length polymorphism analysis with nasopharyngeal swabs. J Clin Microbiol. 2002;40:84–8.
5. Hayden FG, Belshe RB, Clover RD, Hay AJ, Oakes MG, Soo W. Emergence and apparent transmission of rimantadine-resistant influenza A virus in families. N Engl J Med. 1989;321:1696–702.
6. World Health Organization Department of Communicable Disease Surveillance and Response. Avian influenza A(H5N1)—update 22: first data on patients from Viet Nam, clinical data from Hong Kong 1997, susceptibility of H5N1 viruses to antiviral drugs. 2004 Feb 12 [cited 2005 May 25]. Available from http://www.who.int/csr/don/2004_02_12a/en/
7. Leneva IA, Roberts N, Govorkova EA, Goloubeva OG, Webster RG. The neuraminidase inhibitor GS4104 (oseltamivir phosphate) is efficacious against A/Hong Kong/156/97 (H5N1) and A/Hong Kong/1074/99 (H9N2) influenza viruses. Antiviral Res. 2000;48:101–15.
8. Israeli Ministry of Health. Oseltamivir phosphate. The Israel Drug Registry. 2005 May 24 [cited 2005 May 25]. Available from http://www.health.gov.il/units/pharmacy/trufot/PerutTrufa.asp?
   Reg_Number=118%2079%2029952%2000&safa=e
9. Kirkbride H, Watson J. A review of the use of neuraminidase inhibitors for prophylaxis of influenza. NI paper for communicable disease and public health (CDPH). UK Health Protection Agency. 2002 Sep 27 [cited 2005 May 25]. Available from http://www.hpa.org.uk/infections/publications/pdf/Neuramindase.pdf
10. Centers for Disease Control and Prevention. Antiviral agents for influenza: background information for clinicians. 2003 Dec 16 [cited 2005 May 25]. Available from http://www.cdc.gov/flu/professionals/pdf/antiviralsbackground.pdf
11. Hayden FG, Atmar RL, Schilling M, Johnson C, Poretz D, Paar D, et al. Use of the selective oral neuraminidase inhibitor oseltamivir to prevent influenza. N Engl J Med. 1999;341:1336–43.
12. Treanor JJ, Hayden FG, Vrooman PS, Barbarash R, Bettis R, Riff D, et al. Efficacy and safety of the oral neuraminidase inhibitor oseltamivir in treating acute influenza: a randomized controlled trial. US Oral Neuraminidase Study Group. JAMA. 2000;283:1016–24.
13. Cooper NJ, Sutton AJ, Abrams KR, Wailoo A, Turner D, Nicholson KG. Effectiveness of neuraminidase inhibitors in treatment and prevention of influenza A and B: systematic review and meta-analyses of randomised controlled trials. BMJ. 2003;326:1235–41.
14. Kaiser L, Wat C, Mills T, Mahoney P, Ward P, Hayden F. Impact of oseltamivir treatment on influenza-related lower respiratory tract complications and hospitalizations. Arch Intern Med. 2003;163:1667–72.
15. Monto AS, Robinson DP, Herlocher ML, Hinson JM Jr, Elliott MJ, Crisp A. Zanamivir in the prevention of influenza among healthy adults: a randomized controlled trial. JAMA. 1999;282:31–5.
16. Longini IM Jr, Halloran ME, Nizam A, Yang Y. Containing pandemic influenza with antiviral agents. Am J Epidemiol. 2004;159:623–33.
17. Oxford JS, Bossuyt S, Balasingam S, Mann A, Novelli P, Lambkin R. Treatment of epidemic and pandemic influenza with neuraminidase and M2 proton channel inhibitors. Clin Microbiol Infect. 2003;9:1–14.
18. Breslow NE. Elementary methods of cohort analysis. Int J Epidemiol. 1984;13:112–5.
19. Fleiss JL. The measurement of interrater agreement. In Fleiss JL, editor. Statistical methods for rates and proportions. New York: Wiley; 1981. p. 212–36.
20. Kleinbaum DG, Kupper LL, Morganstern H. Stratified analysis. In: Epidemiologic research. New York: Van Nostrand Reinhold; 1982. p. 321–76.
21. Abramson JH, Gahlinger PM. Computer programs for epidemiologists: PEPI Version 4.0. Salt Lake City (UT): Sagebrush Press; 2001.
22. Balicer RD, Huerta M, Grotto I. Tackling the next influenza pandemic. BMJ. 2004;328:1391–2.
23. Soares C. Cooping up avian flu. Sci Am. 2005;292:20–2.

EID Home | Top of Page | Ahead-of-Print | Past Issues | Suggested Citation | EID Search | Contact Us | Accessibility | Privacy Policy Notice | CDC Home | CDC Search | Health Topics A-Z

This page posted June 24, 2005
This page last reviewed July 6, 2005

Emerging Infectious Diseases Journal
National Center for Infectious Diseases
Centers for Disease Control and Prevention