Cost effectiveness of oseltamivir treatment of influenza: a critique of published methods and outcomes

Abstract

Objective: The objective of this review was to determine, from a systematic assessment of published data, the cost effectiveness of the neuraminidase inhibitor antiviral medication oseltamivir in comparison with usual care (i.e. over-the-counter medication such as analgesics and antipyretics for symptomatic relief) for the treatment of influenza. How the findings of each of the studies considered related to the methods used for each analysis and the assumptions made were specifically reviewed.

Results: The online search found 80 individual articles, 66 of which did not meet the pre-defined screening criteria. The 14 studies remaining reported cost, cost-effectiveness and cost-utility analyses for oseltamivir treatment in various groups: healthy adults and adolescents, children, elderly, and individuals at increased risk.

Conclusion: Despite the range of values assumed for key probabilities such as the diagnostic certainty of influenza among people presenting with influenza-like illness, and how much work time is lost due to illness in healthy adults, base-case analyses consistently showed oseltamivir treatment to be cost effective or even cost saving for the four population groups studied, a conclusion that is in-line with previous reviews on this topic. However, clarity was frequently lacking in the published data in terms of various model assumptions and results, particularly with regards to the exact distributions of the constituting elements of savings and of quality-adjusted life years gained.

Keywords::

• oseltamivir
• influenza
• pharmacoeconomics
• cost-effectiveness analysis
• health-economic evaluation

Introduction

Annual influenza epidemics result in significant economic losses to society as a consequence of the considerable morbidity and mortality associated with the disease. Each year, influenza can affect up to 20% of the population, the most recent estimates for the US suggesting that seasonal epidemics are associated with a loss of ∼80 million working days, ∼226,000 excess hospitalisations, ∼35–41,000 excess deaths and a total economic burden, based on projected statistical life values, of ∼$87 billion per year^1–5. Although a range of factors contribute to the overall cost burden, the financial impact results primarily from the loss of working time/productivity associated with influenza-related absenteeism (indirect costs) and the cost of medical resources required to treat patients with influenza and influenza-related complications (direct costs), with the indirect costs typically accounting for the majority (>80%) of the total costs⁶.

A large proportion of the excess financial burden of seasonal influenza epidemics on the healthcare sector results from the treatment of patients who develop secondary complications; most commonly these are respiratory complaints, such as pneumonia and bronchitis, and otitis media^7–9. Such patients generally require an increased level of medical intervention, may require hospitalisation, and are at a higher risk of death. In general, it is children, the elderly and patients with a chronic co-morbidity (e.g. cardiovascular disease, diabetes or asthma) or those in residential care who are at greatest risk of developing influenza-related complications, with elderly patients being at highest risk of mortality. Children have the highest rates of infection in the overall population, prospective surveillance having demonstrated annual attack rates of 15–42% in pre-school and school-age children¹⁰. Infected children play a central role in the spread of the disease, as they are frequently the point of entry of the virus into the household from which adults become infected¹¹. However, it is important to note that although the risk of serious complications and death is significantly lower in otherwise healthy adolescents and adults aged 15–64 years old, as the majority of influenza infections occur in this group, they represent the greatest proportion of complications reported⁷. In addition, as these patients typically represent the working population, influenza infection within this group leads to a substantial increase in societal costs, primarily resulting from absenteeism and reduced productivity at work^12,13.

Immunisation is the primary means of influenza prevention and, although vaccination has been proven to be cost effective in reducing influenza incidence and reducing economic burden, coverage is limited in practice to specific population subgroups, with current strategies focusing primarily on older patients (>65 years of age), higher-risk adults and children <5 years of age¹⁴. Furthermore, the efficacy of vaccination can be limited by factors such as the immuno-competency of the recipient (an issue particularly in the elderly) or a mismatch between the vaccine and the circulating influenza strain. As a consequence, a significant proportion of the population each year remains susceptible to influenza infection, giving rise to the use of antiviral drugs such as neuraminidase inhibitors (NAIs) to treat influenza cases. Obviously, evaluation of the pharmacoeconomics around the use of such antiviral drugs is one of the aspects to be discussed before recommending their widespread use.

Pharmacoeconomic analyses should be valid, transparent and consistent with published guidelines. In addition, they should be clear on which elements of costs and benefits they are addressing and in which population groups this is being done. For this purpose, decision models – such as decision trees – are widely used in the analysis of interventions against influenza. In decision models, it is crucial to clearly specify the cohort that enters the model; this is often a group of people with influenza-like illness (ILI) rather than influenza itself, which may be a better reflection of the practical clinical situation (although during epidemic situations the overlap will be high), as laboratory testing/confirmation is not routinely done in most countries for patients presenting with ILI. Patients with ILI may exhibit influenza-like symptoms due to other causes – including respiratory syncytial virus infection – and this should be considered in any model. The exact proportion of ILI patients who have influenza will also depend on the time of year and whether the virus is actually circulating within the relevant time period, as well as on the vaccination status of the population.

The scope of this paper is to consider the health economics of treating ILI patients with oseltamivir, an oral NAI that significantly reduces the duration of seasonal influenza when taken within 48 hours of symptom onset^15–17. A meta-analysis including more than 3,500 adults and adolescents in 10 studies reported the benefits of oseltamivir treatment in confirmed influenza cases, including a 55% reduction in lower respiratory tract complications, a 59% reduction in hospitalisation and 27% fewer antibiotic prescriptions¹⁸. In infected children, oseltamivir therapy was associated with a 44% reduction in the incidence of otitis media when compared with placebo¹⁸.

The pharmacoeconomics of oseltamivir have been comprehensively evaluated in a range of published studies in a variety of populations and scenarios, some of which have been the subject of previous reviews aiming to assess a wide range of influenza treatment and prevention strategies^19,20. Possibly, because of the large scope of these reviews, drawing conclusions on relative cost effectiveness has not been easy. Although Lynd et al¹⁹ showed oseltamivir to be cost saving from a societal perspective and cost effective from a healthcare perspective, the authors emphasised that variability in methods and assumptions created uncertainty about how widely results could be generalised. Nevertheless, these reviews represent a good point of reference for the current assessment. Since these publications, additional studies have been published warranting this review.

It was noted that although papers identified in this search also reported economic evaluations of other influenza treatments and of vaccination, explicit comparison of oseltamivir with these interventions is outside the scope of this review. In addition, the cost effectiveness of the use of oseltamivir for post-exposure prophylaxis is also considered outside the scope of this paper.

Methods

To identify articles for this review, a search of four online databases (Medline, Biosis, SciSearch and Embase) was performed. The search terms used were (oseltamivir) AND (influenza OR flu) AND (cost^* OR pharmaco economic^* OR pharmacoeconomic^* OR pharmaco-economic^* OR health economic^*). The use of the asterisk after ‘cost’ allowed searching for terms such as cost effectiveness, cost utility, cost benefit and all their variants. The search was limited to English language publications only.

All publications found in the search were screened according to inclusion and exclusion criteria. Publications eligible for inclusion were peer-reviewed papers in scientific journals that reported pharmacoeconomic analyses of oseltamivir for the treatment of seasonal influenza. Furthermore, publications should refer to the European and Northern American context, to achieve relatively comparable societal settings and healthcare systems.

For each article selected by the above screening procedures, two assessors (MJP and/or PB) evaluated the reported outcomes, as well as the analytical and modelling approaches used, the perspective taken and assumptions made. The elements of particular interest were:

1.	What proportion of ILI cases was actually (assumed to be) due to influenza?
2.	What was the probability that patients started treatment within 48 hours of first symptoms?
3.	What were the specific base-case assumptions, including probabilities of influenza-related complications?
4.	Which sensitivity analyses were performed?
5.	Which specific features were underlying the calculation of cost effectiveness?
6.	How was cost utility calculated and how were utility scores determined?
7.	How was the amount of work loss per episode of influenza estimated?

Results

The online search found 80 individual articles, 62 of which did not meet the pre-defined screening criteria. Of the 18 articles remaining, four more were excluded, two^21,22 because they referred to the specific Japanese situation and two^23,24 because the treatment intervention analysed was either a ‘generic’ NAI, for which the efficacy and costs were calculated by combining data for different NAIs and did not differentiate data for oseltamivir separately. The principal features of the 14 articles assessed are shown in Table 1, categorised by population group (three papers appear more than once, as they analysed patients from more than one group). Each article is described in greater detail in the text below, using the same grouping method. All the articles assessed used a decision-analytical model to calculate outcomes and their associated costs.

A summary of the most important assumptions made in each study, and the key assumptions underlying the findings, is shown in Table 2, inclusive of the results of the base-case analysis in each paper.

Otherwise healthy adults and adolescents

Cost and cost-benefit analysis

Two studies used cost or cost-minimisation analysis to compare oseltamivir treatment with usual care, both studies taking a societal perspective.

The first of these was a study from the US by Lee et al²⁵, which evaluated the use of antiviral treatment with and without vaccination. The study was based on a decision-tree model constructed for otherwise healthy individuals aged 18–50 years old. Symptom relief and treatment of side effects were ascribed a monetary value calculated from a willingness-to-pay analysis on 210 patients at a primary care clinic. The model was explicitly evaluated for influenza-infected patients who had either previously received vaccination or had not been vaccinated. Oseltamivir was assumed to reduce work loss by 0.5 days in comparison with usual care and to reduce the duration of symptoms by 1 day. The authors showed how cost-benefit was strongly influenced by the amount and valuation of work time saved. In the base-case analysis, comparison of oseltamivir treatment with no intervention/no vaccination, produced a net monetary benefit of US$29.39 and approximately zero respectively, for influenza-infected vaccinated and non-vaccinated individuals, respectively. A key factor influencing the cost-benefit calculation in this study was the valuation of disutility. In the valuations generated by a willingness-to-pay survey, avoiding influenza was worth much less (about US$15 per day of relief) to patients than avoiding nausea (more than US$60), which was an assumed side effect of oseltamivir and associated with a probability of 9% in the model. In addition, the authors also conservatively assumed that oseltamivir treatment would have no effect on hospitalisation rates, complications, physician visits or severity of symptoms (only an impact on antibiotic requirement, assumed to be 35% lower than usual care). Although methodologically innovative, the assumption that virologically identified influenza infection was present limits the relevance of the study, as physicians are generally uncertain about the actual involvement of influenza infection and merely just the diagnosis of ILI will be available.

The second study to use cost analysis was done in the Netherlands by Postma et al²⁶, who analysed treatment with oseltamivir from a societal perspective, next to prevention by vaccination as an alternative. The analytic framework applied was basically that of a cost-minimisation analysis. Oseltamivir treatment was found to be cost saving relative to usual care for all types of workers in both age groups analysed, and was highest for full-time workers aged 45–65 years old (€135 cost savings per patient) and lowest for half-time workers aged 25–44 years old (€40). The base-case model assumed that 24% of people with ILI would visit their physician for treatment and that the real influenza incidence in the population was approximately 8.5%. Patients presenting too late (after 48 hours) were not included in the model as they would not be expected to receive treatment (and so accrue neither costs nor benefits). Given the extremely low levels of influenza-related deaths in healthy working adults, mortality was not included in the model. In this study, patients treated with oseltamivir were assumed to return to normal activity earlier (a 24% reduction in the number of illness days) than those treated with symptomatic relief alone. Based on previous studies^27,28, 1.5 work days were assumed to be lost due to ILI, a smaller value than other papers assessed in this review (including Lee et al²⁵). However, a further 1.75 days were deemed lost during the first 3.5 days spent at work, which were assumed to be at 50% of normal efficiency. Probabilistic sensitivity analysis showed that even for the group with the lowest cost savings (half-time workers aged 25–44 years old), treatment was cost saving 100% of the time.

Summarising, two studies analysed costs, savings and net benefits for antiviral treatment in healthy working adults. Both studies found potentials for cost savings in this group, in particular for those not vaccinated and working full time.

Cost-effectiveness analysis

The study by Armstrong et al²⁹ was the only one in this review that just used cost-effectiveness analysis to compare oseltamivir with usual care for influenza treatment. It also differed from all the other studies that were assessed by taking the perspective of a US-managed care organisation. As the analysis was performed from the perspective of the healthcare provider, no allowance was made in the study for lost work days. For influenza-positive patients, cost effectiveness of oseltamivir was found at an incremental cost per successfully treated patient of US$677 and at US$39 per symptom-free day gained. As an approximation, the latter would correspond with US$14,000 per symptom-free year gained and US$28,000–70,000 per quality-adjusted life year (QALY) gained if a 20–50% reduction in quality of life is assumed for influenza³⁰. In this study, the base-case analysis did not take into account the impact of diagnostic accuracy. Therefore, a sensitivity analysis was performed assuming a diagnostic accuracy of 60% (assuming that 40% of the ILI cases treated would not be influenza) with a related cost-effectiveness of US$723 and US$41 per successfully treated patient and symptom-free day, respectively. Cost-utility, bivariate, multivariate and probabilistic sensitivity analyses were not performed in this study. The applicability of the findings to clinical practice is limited to those patients who indeed present and start treatment within 48 hours of the onset of illness, as this was explicitly assumed in the model. Subsequent to publication, the event probabilities used and some of the assumptions drawn from the original analysis were challenged in a letter to the publishing journal by Poulios et al³¹ who recalculated that oseltamivir would provide more health gains for both the cost-effectiveness measures reported in the original paper.

Cost-utility analysis

A further six studies of the treatment of healthy adults and adolescents were identified that included a cost-utility analysis, giving the cost per QALY, with or without other outcomes. Of these, two (both from the US) took only a societal perspective, three (from Europe) took both a societal and a healthcare payer's perspective and one (from Canada) took only a healthcare payer's perspective.

In the earlier of the two US studies that took a societal perspective, Muennig and Khan analysed the cost utility of oseltamivir for treating unvaccinated people presenting with ILI³². The cohort modelled in the study referred to all healthy adults in the US; of the 49% of adults who were assumed to develop ILI (which may be considered rather high), the model assumed that 10% would visit their physician for treatment. Although the analysis took a societal perspective, and included the time spent by the patient travelling to the clinic or hospital and being treated, the cost of lost work time was not accounted for. This is surprising given that the authors did add costs for caregiver support, and assumed cost savings because bed-bound patients were not using transport to work or school. Despite this somewhat unconventional approach, the analysis found that the incremental cost per QALY relative to usual care was US$27,600, well below the US$50,000 benchmark value for the US usually used to signify that an intervention is likely to be cost effective³³. The Quality of Well-Being (QWB) scale was used to quantify quality of life in people with ILI. The authors acknowledged that technically this instrument does not produce utilities, but it was preferred to the Health Utilities Index because the health dimensions it measures are more appropriate to ILI and may certainly well approach utilities. Further interpretation of the results is complicated because information presented on the model and the sensitivity analysis is limited.

In their study from a societal perspective in the US, Rothberg and Rose analysed the cost utility of treating healthy working adults with oseltamivir relative to usual care³⁴. In their model, oseltamivir is only given after a positive rapid test for influenza (which is included in the cost for this strategy), an approach that does not reflect normal clinical practice. From their analysis, the authors showed that test-and-treat with oseltamivir is $0.20 per year more costly than usual care, but produces 0.024 fewer illness-days per year, or 0.0104 extra quality-adjusted days. This corresponds to an incremental cost per QALY of US$7,020 per QALY gained. One of the main reasons why this analysis may have underestimated the cost effectiveness of oseltamivir is related to the assumed percentage of actual influenza among ILI cases. The authors applied a maximum value of 31%, which is very low relative to that used in other studies. In addition, the paper appears to assume that antiviral treatment has no effect on complications with the exception of related use of antibiotics. As in the Muennig & Khan study³² described above, the Quality of Well-Being Scale was used to derive a utility value for influenza (0.6); however, although varying this value between 0 and 1 substantially changes the cost-utility ratios, it is unclear what impact this has on the cost utility of the oseltamivir strategy, as reporting of incremental cost-effectiveness ratios is limited. The work time lost was assumed to be 1.9 days, but no ‘illness-at-work’ productivity loss was assumed in this model as was done in the Dutch study²⁶; oseltamivir was assumed to shorten duration of illness by 24 hours.

Three studies performed in the UK³⁵, Denmark³⁶ and Finland³⁷ all employed a model developed specifically to economically analyse oseltamivir treatment in comparison with alternative pharmacological interventions from both the healthcare payer and societal perspective. Where available, each of the authors based their assessments on local costs and resource use data. The model, described in detail by Sander et al³⁵, accounts for the probability of patients starting treatment ± 48 hours from symptom onset, the probability of a patient being influenza positive or negative, the probability of a patient developing complications (in this case pneumonia and bronchitis, with each reported complication occurring in a different patient) and whether the patient is treated as an outpatient, is hospitalised or dies. Treatment with oseltamivir was assumed to have an effect on the incidence but not on the clinical course of complications. In the data presented, it is assumed that usual care represents symptomatic medication. In the utility calculation, quality adjustment of illness days is done using weightings derived from valuations of health status during 21 days of a clinical trial. Likert scale scores are transformed first to visual analogue scale (VAS) scores, then to time trade-off (TTO) equivalent scores, using an algorithm based on quality of life valuation work done at York University³⁸. Due to a typo in one of the tables of the papers, it remains unclear which QALY weight is exactly used for ILI with usual care as compared to the 0.94 for oseltamivir. It was estimated that it is likely to be around 0.90.

In the UK study reported by Sander et al³⁵, the base-case analysis showed oseltamivir was cost effective from the healthcare payer's perspective (an incremental cost-effectiveness ratio of £14 per day of normal activity gained, and an incremental cost-utility ratio of £5,600 per QALY gained). From the societal perspective, oseltamivir was dominant over usual care. The model assumed that 1.4 fewer days of activity would be lost with oseltamivir and that 100% of illness days (before resumption of normal activity) represent lost productivity due to ILI. The authors acknowledged the 100% figure to be higher than that in other studies and potentially biased upwards, but when this was reduced to 25% in sensitivity analyses, the effect on results was small. Furthermore, the authors found that results were sensitive to the assumed rate of diagnostic certainty (70% in the base case). Yet, sensitivity analysis using a rate of 34% still produced a cost per QALY gained of £11,872 from the healthcare payer's perspective (still well below the £30,000 threshold³⁹) and oseltamivir was still dominant from a societal perspective. A similar pattern of results was derived from an analysis that assumed that 25% of patients started treatment after the 48 hours (so incurring cost but accruing no benefits); the base case assumed no late starters (i.e. 100% of patients commenced treatment <48 hours after onset of symptoms). A discount rate of 1.5% was used to adjust the value of life years lost due to premature death in this study, but omitting discounting was found to have little effect on the results in sensitivity analysis.

The description by Vindt Holm et al³⁶ of their analysis for Denmark is less extensive than that provided by Sander et al³⁵ for the UK, which somewhat complicates the assessment of their methods and assumptions. However, quality of life methods, assumptions and outcomes are reported in depth, which is very useful. Although not clearly stated, the assumption on productivity loss seems to be the same as that in the UK study. However, in contrast to that study and based on the probabilities reported in the paper, in this study oseltamivir does not appear to reduce the risk of complications and, despite a lower risk of hospitalisation, death was more likely with oseltamivir. However, as this would be contrary to the clinical data and to the basic assumptions of all other oseltamivir models, the authors would assume that this is an error in the way in which the data are presented. The cost-effectiveness analysis, based on days of normal activity gained (assumed treatment difference of 1 day) showed oseltamivir to be dominant from the societal perspective and an incremental cost per QALY gained of €5,060 from the perspective of the healthcare payer. Diagnostic certainty was assumed to be 70%, in-line with the value observed in oseltamivir clinical trials and it was assumed that oseltamivir would only be prescribed to patients presenting within 48 hours of symptom onset. Only a limited number of the base-case assumptions were tested in the sensitivity analysis in this study, a single multivariate sensitivity analysis (excluding all complications, hospitalisations and mortality) performed from the perspective of the healthcare payer, indicating that oseltamivir remains cost effective (cost per QALY gained of €11,400) in this conservative setting. However, data for univariate analyses of other key base-case model assumptions such as diagnostic certainty (70%) and proportion of late starters (0%) were not presented.

A third study based on the application of the same modelling approach, but in this case applied to a Finnish setting, was published by Järvinen et al³⁷. In this analysis, oseltamivir was assumed to reduce the time taken to return to normal activity by 1.4 days but exact production losses regarding various assumptions on labour participation rates, part-time jobs and days off work anyway due to holidays and weekends assumed are not listed clearly. In contrast to the analyses of Sander et al³⁵ and Vindt Holm et al³⁶, the incidence of sinusitis was also included as a complication in this model. In the base-case analysis (70% diagnostic certainty and 90% of patients commencing treatment <48 hours after symptom onset), oseltamivir treatment was found to dominate usual care from the societal perspective. This result proved robust, as a lowering of the averted indirect costs of production losses by 50% – taking into account our above critique – still showed the desirable combination of cost savings accompanied by QALY gains. Excluding indirect costs fully, oseltamivir provided a net cost per QALY gained of €13,400 from the perspective of the healthcare payer, likely to be considered cost effective in Finland. Univariate sensitivity analysis around the key base-case assumptions (diagnostic certainty/commencement of treatment <48 hours/effect of oseltamivir on incidence of complications, hospitalisation and mortality/cost of working day lost) revealed a range that generally remained well below €30,000 per QALY gained. The only exception was diagnostic certainty, where a reduction to 27% resulted in an increase to €40,000 per QALY gained, although no rationale was provided as to why such a low figure was considered. Although effects of oseltamivir on mortality were assumed, the relative contributions to overall QALYs gained of respective shares for life years and quality were not provided in the publication.

A similar model was used by O'Brien et al ⁴⁰ to compare the cost effectiveness and cost utility of oseltamivir with standard antibiotics for empirical treatment of complications in Canadian healthy adults. In this study, the analysis was performed only from the perspective of the healthcare provider and so no allowance was made for work time lost due to illness. A relatively high incremental cost utility of CAN$57,900 per QALY was found (cost per influenza day averted of CAN$49), based on a diagnostic certainty for influenza of 69%⁴¹. In the base-case analysis, it was assumed that no patients would start treatment after 48 hours of symptom onset, but the effect of up to 100% starting late was tested in a sensitivity analysis. At the 100% value, the results of a bivariate sensitivity analysis shown in the paper suggest some retention of efficacy if treatment is started late, which is not explained further and is in contrast to what is stated in their Methods section. The model used in this study was very similar to various other models described above, except that O'Brien et al calculated quality adjustment factors directly from the VAS scores from the trials (normalised to an interval from 0 to 1) without transforming them to TTO scores. In particular, scores from the 7 days after treatment initiation in the clinical trials were used. Reduced hospitalisation rates due to oseltamivir were directly modelled on the reductions in pneumonia incidence (as has also been done by other researchers included in this review).

Summarising, most cost-utility analyses directed at oseltamivir treatment of healthy adults and adolescents seem to indicate acceptable costs per QALY gained from the perspective of the healthcare payer and cost savings from a societal perspective. However, some confusion remains on the specific methods used, the assumptions made and the reported calculations, potentially limiting the usefulness of some of these studies.

Children

Three studies among children met the criteria for assessment. The first study by Reisinger et al⁴² used the same decision-modelling approach as described by Sander et al³⁵. The economic analysis was performed from the perspective of a UK healthcare payer (incorporating UK costs and resource use data) and included analyses for the full population (aged 1–12 years) and a subgroup analysis for children aged 1–5 years. In addition to bronchitis and pneumonia, the incidence of otitis media as a complication of influenza infection in children was included. For the base-case analysis, 100% presentation and 60% diagnostic certainty were assumed. For the whole study population, oseltamivir was shown to be cost effective at £11,200 per QALY gained and £9.52 per day of normal activity gained. For the subgroup aged 1–5 years, oseltamivir was dominant over usual care for both cost utility and cost effectiveness. With the exception of a reduction in diagnostic certainty to a very low level (10%), in all sensitivity analyses performed (univariate: 47.5% diagnostic certainty; 90% of patients presenting <48 hours; reduced incidence of influenza-related hospitalisation; multivariate: 47.5% diagnostic certainty; no complications, hospitalisations and mortality) the cost per QALY remained well below the £30,000 ‘acceptable’ limit. Quality of life valuations of the influenza episode used in the utility analysis were derived from healthy adults as no corresponding data for children were available⁴³. Although the study provides very useful data regarding the cost effectiveness of oseltamivir in children, it provides only a limited insight into the distribution of cost savings over the various categories of resource utilisation and the respective shares of life years and quality gained.

The use of oseltamivir for empirical treatment and a test-and-treat strategy in otherwise healthy children in the US was the subject of a study by Rothberg et al⁴⁴. The authors took a societal perspective, assuming that just over 50% of parents would stay at home to look after an ill child and so would lose work time in proportion to the time the child was away from school. The base case assumed a mean of 4.67 days, based on one of only a few valid studies done in this area by Nettleman et al⁴⁵. Oseltamivir was assumed to enable the child to return to school 45 hours sooner than usual care. Utility was valued using quality adjustments based on the EQ-5D instrument (EuroQOL) and included the duration of influenza itself, side effects of antiviral treatment and reduction in incidence of otitis media. However, oseltamivir was assumed to have no effect on the risk of hospitalisation. For two of the scenarios tested (i.e. for children aged 2 and 7 years old) oseltamivir was shown to be dominant over usual care for both empirical treatment and test-and-treat and in 15 year olds, cost effectiveness of oseltamivir was approximately within the range of $23,000–25,000 per QALY. Pre-treatment testing was always inferior to empirical treatment because influenza was assumed to be present in 68% of the cases in the model and tests were not sensitive enough.

An assessment in children aged 1–12 years old from a Finnish perspective was performed by Järvinen et al³⁷, a general critique of their approach having already been provided above. In contrast to their analysis in adults, for children, otitis media as a complication of influenza infection was included in place of sinusitis in this analysis. For the base case (70% diagnostic certainty and 90% patients commencing treatment <48 hours after symptom onset), an incremental cost-utility ratio of €15,404 per QALY gained was determined from the perspective of the healthcare payer. From a societal perspective, inclusion of the cost of parental work loss (on average 0.4 working days) due to childcare resulted in a cost-utility ratio for oseltamivir treatment of €5,750. A univariate sensitivity analysis around the key base-case assumptions (diagnostic certainty; commencement of treatment <48 hours; effect of oseltamivir on incidence of complications, hospitalisation and mortality; cost of working day lost) revealed a range that generally remained below €30,000 per QALY gained, the exceptions being diagnostic certainty, where a reduction to 27% resulted in an increase to €54,900, and 50% late starters, where an increase to €36,000 was observed. With regards to the methodology applied, no specific research towards quality of life in children with influenza or parental/societal preferences regarding prevention of influenza in these children is explicitly reported. In addition, quality of life weights as derived for otherwise healthy adults were also applied in this evaluation.

Summarising, cost utility for oseltamivir treatment in young children is highly dependent on the specific assumptions applied for absenteeism of parents and quality of life weights assumed for these children. In general, treatment with oseltamivir is cost effective from the healthcare payer's perspective and from a societal perspective when parental work loss resulting from childcare is included in the analysis.

Elderly

One of two studies performed among elderly people that met our screening criteria was that by Rothberg et al⁴⁶. Although the incremental cost per QALY for empirical oseltamivir compared with usual care is not stated in the paper, it can be calculated from the data for the base case (an unvaccinated 75-year-old patient) that the net costs per QALY is US$5,561 ([155.56–118.86]/[9.9849–9.9783]). For a vaccinated person, cost utility is worse, with the cost per QALY rising by a factor of 6 or more. The diagnostic certainty assumed for the base case in this analysis was 35%, with other rates tested in sensitivity analyses. The authors show how changes in this rate affect the incremental cost utility of oseltamivir over usual care for healthy older adults with an assumed risk of hospitalisation of 4%. During epidemics when influenza virus is circulating and diagnostic certainty is over 70%, empirical oseltamivir is most cost effective. At only 35% certainty, it is still the best option for unvaccinated people, but test-and-treat (with oseltamivir) is more cost effective for those already vaccinated and, as influenza probability falls further, the most cost-effective strategies are test-and-treat for the unvaccinated and usual care of pain relief only for the vaccinated.

Postma et al⁴⁷ performed the one study that calculated outcomes in net costs per life year gained, assuming the within-trial reductions for pneumonia to approximate reductions in mortality. The model was developed on previous modelling work^48,49 and directly incorporated the findings from the meta-analysis by Kaiser et al¹⁸. As the model was developed directly on outcomes for ILI patients, diagnostic certainty was only implicitly included as observed in the oseltamivir trials at 65%¹⁸. Extensive sensitivity analysis was performed on the percentage of patients not presenting within 48 hours of symptom onset (0% (base case) up to 75%) and therefore assumed to be too late to benefit from their oseltamivir prescription. For elderly not belonging to the high-risk group, cost effectiveness was estimated to be below €10,000 per life year gained, this figure increasing to €12,200 per life year when 75% of the patients were assumed to present >48 hours. A probabilistic sensitivity analysis was additionally included.

Two further studies included elderly people in the analyses, but only as part of a population deemed to be ‘high risk’ – results for the healthy elderly groups in this analysis were not reported separately, so these studies are described in the next section.

Summarising, two studies directed at oseltamivir treatment in the elderly indicate that the health-economic profile crucially depends on the vaccination status of the patient presenting with ILI and the diagnostic certainty. The latter obviously increases during epidemics when influenza is widely circulating and when unvaccinated ILI patients and vaccinated patients with breakthrough influenza may benefit from oseltamivir at acceptable costs per QALY and life year gained.

High-risk groups

The US study among elderly people described above⁴⁶ used sensitivity analyses to model cost effectiveness of oseltamivir over usual care for those at elevated risk of hospitalisation from influenza, using risk categories from a US study of vaccination effectiveness⁵⁰. In those at high risk (24% risk of hospitalisation in case of influenza infection; typically those with heart or lung disease), empirical oseltamivir was consistently more cost effective regardless of vaccination status and the same was true for medium-risk individuals (10% risk of hospitalisation; typically those with diabetes, stroke, dementia, or renal or rheumatological disease), except at times when influenza probability was <10%, in which case testing before oseltamivir treatment is more cost effective for those who are vaccinated.

Three other studies have compared oseltamivir with usual care for people at elevated risk of influenza complications. In the UK study by Sander et al⁵¹, high-risk adults were defined as those aged under 65 years old with co-morbidities and all those aged ≥65 years old, but, as mentioned above, results for the otherwise healthy elderly group are not reported separately. Using a model applied by the author group in other studies (see above), and taking the perspective of the healthcare payer, oseltamivir was found to be cost effective compared with usual care at a cost per QALY gained of £225 (incremental cost of £3 per day of normal activity gained). The high level of diagnostic certainty (70%) assumed in this analysis was the same as in the other two studies by the authors, but at a lower certainty rate of 34%, modelled in their sensitivity analysis, the cost per QALY was still £911. The authors noted that data on the effect of oseltamivir treatment on the typical disease course in this patient group are very limited. Therefore, they included conservative scenarios in their sensitivity analyses. It remains unclear in the paper what was the crucial driver for the savings and what was the exact evidence for that parameter.

Järvinen et al³⁷ reported data for an analysis from the Finnish perspective for the same at-risk population as that previously described by Sander et al⁵¹. The authors reported a very favourable cost-effectiveness ratio in the baseline (70% diagnostic certainty and 90% presentation ≤48 hours) of €754 per QALY but with a wide range in uncertainty analysis, with results varying from oseltamivir treatment dominating usual care up to usual care dominating. Quality gains due to oseltamivir were assumed to be less pronounced than for the other two population groups analysed in the model (otherwise healthy adults and children) and discussed previously, although significant mortality benefits were attributed to the treatment. In the absence of a mortality benefit and based only on quality gains, the cost-effectiveness ratio was still only €3,000 per QALY, but the uncertainty range failed to include the option of oseltamivir being dominant.

Finally, Postma et al⁴⁷ addressed the high-risk groups separately in their paper. The methods used were similar to those described above, with results for both elderly and non-elderly patients provided. In particular, at a baseline diagnostic certainty of 65% and 100% of patients being treated ≤48 hours, savings were consistently estimated to be greater than the costs in various options investigated. In only one of the options investigated was oseltamivir not estimated to be dominant (i.e. cost saving) – the exclusion of indirect savings due to averted production losses for the non-elderly at increased risk. In this scenario, costs per life year gained were negligible, at only €400.

Summarising, studies indicate favourable costs per QALY or life year gained for oseltamivir treatment of risk groups. Indeed, some models indicate potentials for cost savings. Limitations to interpreting the studies relate to the facts that some studies do not explicitly separate elderly from the overall high-risk group and savings are not always reported in detail, which hampers insights into the relative contributions of individual components. Obviously, for this high-risk group, vaccination status is another parameter of relevance for cost effectiveness that warrants further research.

Discussion

The great majority of papers that met the selection criteria for this assessment used cost-effectiveness or cost-utility analysis based on decision models, providing a reasonable degree of comparability between them. The key assumptions used in the models, however, were much more varied, making it more difficult to integrate the results of the analyses and to draw a unified conclusion about the cost effectiveness of oseltamivir relative to usual care. As shown above, despite the range of values assumed for key probabilities such as the diagnostic certainty of influenza among people presenting with ILI, and how much work time is lost due to illness in healthy adults, base-case analyses generally showed oseltamivir treatment to be cost effective or even cost saving; the findings were consistent across the four population groups that were studied. Clarity was sometimes lacking on the exact distributions of savings – specifying, for example, shares of work days lost in the total financial savings – and of QALYs. Some simple pie charts might have helped in this respect. With QALYs often being built from both quality and life year shares, clarity on each part's respective contributions would have been helpful in further interpreting many of the papers discussed. Also, it was found that, in particular for medical high-risk groups and the elderly, explicit consideration of the vaccination status with regard to cost effectiveness was sometimes lacking. The authors feel that this aspect warrants further investigation, with oseltamivir probably more likely to be cost effective in still unvaccinated individuals in these high-risk groups.

The two most common methods used to derive utility values for the papers that were looked at were the EQ-5D instrument and quality of life scores from clinical trial data, based on VAS data or transformed to TTO equivalents. Most of the papers that were assessed did not test the effect of different disease valuations in sensitivity analyses or did not report on this clearly. Authors of several of the papers that were assessed pointed out that they had only limited data on which to base their analyses, particularly on the course of influenza and the effect of antiviral treatment on complications, hospitalisation and quality of life. This was a bigger problem for children, the elderly and high-risk patients. As would be expected, the earlier analyses had fewer clinical trials from which relative efficacy of NAI treatment could be gauged. In healthy adults, most papers assumed diagnostic certainty rates of 60–70%, which are typical of periods when the influenza virus is circulating^41,52,53. The study by Rothberg and Rose³⁴ assumed the lowest rate of all those that were assessed (only 13%), but still found oseltamivir to be cost effective. Another parameter with the potential to affect the economics of oseltamivir treatment is the requirement to commence treatment within 48 hours of symptom onset. Although the majority of analyses allowed for this in their model, few papers reported the effect on the final result of changes in this parameter.

Most papers in this review assumed that oseltamivir would reduce the incidence of complications, most commonly pneumonia and bronchitis in adults and otitis media in children, and that further benefits would accrue from this (e.g. fewer hospital and physician visits and less antibiotic use). Evidence is starting to emerge that influenza is associated with an elevated incidence of cardiovascular complications, even in healthy adults⁵⁴. The economic impact of these sequelae is yet to be fully evaluated and understood, and was not included in any of the articles selected in this review. Also, oseltamivir treatment during pandemics and prophylactic use of oseltamivir was outside the scope of this paper. Furthermore, it was noted that as all studies were directed to the Western world, this analysis lacks generalisability beyond these countries (e.g. for the developing world).

The scope of this review was limited by the limited level of detail of some of the papers that were assessed, relating to the reporting not only of the model and key assumptions but also of sensitivity analyses. Access to more information and/or clearer explanations might have enhanced the interpretation of some of the affected studies. Interpretation of the results in children is also partly limited by the age groups analysed in the two studies that were assessed. Although the clinical study by Whitley et al¹⁷ showed that oseltamivir shortened illness duration not only in infants under 2 years old but also in children aged 2–5 years old and in those over 5 years old, the economic analysis based on that study, which was assessed in this review⁴², showed greater cost effectiveness in a subgroup aged 1–5 years old than in those aged 1–12 years old. It would be interesting to study how the economic benefits in very young children, aged under 2 years old, compare with these other groups, particularly given that infants under 2 years old are treated as a distinct group in respect of vaccination recommendations. The Rothberg et al study⁴⁴ showed that economic benefit of oseltamivir relative to usual care in a 2-year-old infant was similar to that in a 7-year-old child.

The National Institute for Health and Clinical Excellence (NICE) guidance report stated that if oseltamivir were assumed to reduce hospitalisations and deaths, the incremental costs per QALY relative to usual care would be £3,900–25,000 for the at-risk group^55,56. Similar order of magnitude results were found for healthy adults. If assumed reductions on hospitalisations and deaths were not inserted, generally unfavourable cost-effectiveness results were found. The NICE Appraisal Committee seemed to feel that some of the crucial assumptions often made were not supportable. In particular, one assumption causing concern was that people who died in the model would have had a normal life expectancy. Also, the NICE Committee believed that the diagnostic certainty assumed for influenza was often too high⁵⁵. Finally, NICE recommended that oseltamivir be used only for the at-risk group (including those over 65 years old), and only while influenza is circulating, to assure relatively diagnostic certainty.

Conclusions

Despite the range of values assumed for key probabilities such as the diagnostic certainty of influenza among people presenting with ILI, and how much work time is lost due to illness in healthy adults, base-case analyses generally showed oseltamivir treatment to be cost effective or even cost saving and the findings were consistent across the four population groups that were studied (healthy adults, children, the elderly and high-risk groups). This conclusion is in line with previous reviews on the topic. However, clarity was frequently lacking on model assumptions and results (e.g. the exact distributions of the constituent elements of savings and of QALYs), which would further help in the overall assessment of cost effectiveness.

Table 1. Designs of studies screened for analysis in the review (all used decision-analytical simulation models).

Study and reference	Country	Perspective	Analysis done	Sensitivity analyses done
Otherwise healthy adults and adolescents
Lee et al 2002^25	USA	Societal	CBA	UV, PR
Postma et al 2005^26	Netherlands	Societal	CBA	PR
Armstrong et al 2000^29	USA	MCO	CEA	UV
Muennig & Khan 2001^32	USA	Societal	CUA	UV, BV, MV
Rothberg & Rose 2005^34	USA	Societal	CEA, CUA	UV, MV, PR
Sander et al 2005^35	UK	Societal and payer	CEA, CUA	UV, MV, PR
Vindt Holm et al 2004^36	Denmark	Societal and payer	CEA, CUA	MV
Järvinen et al 2007^37	Finland	Societal and payer	CEA, CUA	UV, PR
O'Brien et al 2003^40	Canada	3rd party payer	CEA, CUA	BV, PR
Children aged up to 15 years
Järvinen et al 2007^37	Finland	Societal and payer	CEA, CUA	UV, PR
Reisinger et al 2004^42	USA/UK	Healthcare payer	CEA, CUA	UV, PR
Rothberg et al 2005^44	USA	Societal	CUA	UV, BV, MV, PR
Elderly
Järvinen et al 2007^37	Finland	Societal and payer	CEA, CUA	UV, PR
Rothberg et al 2003^46	USA	Societal	CUA	UV, BV, MV, PR
Postma et al 2007^47	Netherlands	Societal	CEA	UV, BV, PR
High-risk populations
Järvinen et al 2007^37	Finland	Societal and payer	CEA, CUA	UV, PR
Rothberg et al 2003^46	USA	Societal	CUA	UV, BV, MV, PR
Postma et al 2007^47	Netherlands	Societal	CEA	UV, BV, PR
Sander et al 2004^51	UK	Healthcare payer	CEA, CUA	UV, MV, PR

BV, bivariate; CBA, cost-benefit analysis (defined as considering monetary consequences only, no health gains explicitly calculated, nor related to money); CEA, cost-effectiveness analysis (defined as studies formally relating monetary amounts to clinical outcomes, such as infections, complications and life years); CUA, cost-utility analysis (defined as those studies formally relating monetary amounts to quality-adjusted life years); MCO, managed care organisation; MV, multivariate; PR, probabilistic; UV, univariate.

Table 2. Key assumptions made and main outcomes in each study analysed in the review.

Study and reference	Diagnostic certainty*	Proportion who started treatment late (>48 hours after start of illness)	Productivity loss assumed for influenza treated with usual care	Valuation of utility	Outcomes assumed to be improved by oseltamivir relative to usual care	Cost effectiveness and cost utility of oseltamivir in the base case
Otherwise healthy adults and adolescents
Lee et al 2002^25	100%	0%	2.8 days	–	Antibiotic use, symptom duration	No benefit for oseltamivir over usual care
Postma et al 2005^26	65%	0%	1.5 days + 1.75 days from 50% reduced productivity at work	Utility not valued	Antibiotic use, complications, hospitalisation, physician visits, illness duration	Cost savings of €40–135 per patient
Armstrong et al 2000^29	60–100%	0%	Not done (payer perspective)	–	Complications, hospitalisations, physician visits, illness duration	ICER per symptom-free day gained at US$39
Muennig & Khan 2001^32	Not stated	5%	Not accounted for	Quality of Well-Being scale	Complications, hospitalisations, deaths, physician visits, illness duration	ICER US$26,700 per QALY
Rothberg & Rose 2005^34	<31%	0%	1.9 days	Quality of Well-Being scale	Antibiotic use, illness duration	ICER US$7,020 per QALY
Sander et al 2005^35	70%	0% (25% in SA)	8.8 days (7.4 for oseltamivir)	TTO scores derived from transformed Likert-scale QOL scores	Complications, hospitalisations, illness duration, daily utility	ICER £15 per day of normal activity gained and £5,600 per QALY
Vindt Holm et al 2004^36	70%	0%	Full length of time lost to normal activity (but not clearly stated)	TTO scores derived from transformed Likert-scale QOL scores	Complications, hospitalisations, illness duration, daily utility	Oseltamivir dominant from the societal perspective; ICER at €5,060 per QALY from the healthcare perspective
Järvinen et al 2007^37	70%	10%	1.4 days	Various literature sources	Complications, mortality, illness duration, utility	Oseltamivir dominant from the societal perspective; ICER at €13,400 per QALY from the healthcare perspective
O'Brien et al 2003^40	69%	0% (up to 100% in SA)	Not done (payer perspective)	VAS scores of QOL normalised to 0–1 scale	Complications, illness duration, ILI utility	ICERs CAN$49 per influenza-day averted and US$57,900 per QALY
Children aged up to 15 years
Järvinen et al 2007^37	70%	10%	0.4 days	Various literature sources	Complications, mortality, illness duration, utility	ICER ranging from €57,50 to €15,400 per QALY within both perspectives analysed
Reisinger et al 2004^42	10–60%	Modelled effect of late starters (probabilities not stated)	HC perspective only	QOL weighting from Dutch disease disutility data	Incidence of otitis media and pneumonia; hospitalisation and death (but not from pneumonia); illness duration	ICER per normal-activity day gained at £10, incremental costs per QALY at £11,200
Rothberg et al 2005^44	68%	Not included in model	4.67 days per parent off work to care for child	EQ-5D	Incidence of otitis media, illness duration	Oseltamivir dominant in children aged 2 and 7 years; ICER ≈ US$25,000 per QALY in 15 years olds
Elderly
Rothberg et al 2003^46	35%	Not included in model	Patients considered to be not employed	EQ-5D	Hospitalisation, complications requiring antibiotics, illness duration	ICER for unvaccinated man aged 75 years at US$5,560 per QALY
Postma et al 2007^47	65%	0–75% in SA	Patients considered to be not employed	Utility not valued	Hospitalisation, death, use of antibacterials and illness duration	ICER at €1,760 per life year gained
High-risk populations
Järvinen et al 2007^37	70%	10%	Not included	Various literature sources	Complications, mortality, illness duration, utility	ICERs at €8 per day of normal activity gained and €745 per QALY
Rothberg et al 2003^46	35%	Not included in model	Not included	EQ-5D	Hospitalisation, complications requiring antibiotics, illness duration	Oseltamivir cost effective in most scenarios
Postma et al 2007^47	65%	0–75% in SA	1.5 days + 1.75 days from 50% reduced productivity at work	Utility not valued	Hospitalisation, death, use of antibacterials and illness duration	Oseltamivir dominant over usual care
Sander et al 2004^51	70%	Modelled a probability of 10%	Not included	TTO scores derived from transformed Likert-scale QOL scores	Incidence of bronchitis and pneumonia, hospitalisation (ILI), illness duration	ICERs at £3 per normal-activity day gained and £225 per QALY

EQ-5D, EuroQOL 5-domain; ILI, influenza-like illness; QOL, quality of life; QALY, quality-adjusted life year; SA, sensitivity analysis; TTO, time trade-off; VAS, visual analogue scale; HC, healthcare; ICER, incremental cost-effectiveness ratio.

^* Probability that an ILI case is influenza.

Acknowledgements

Declaration of interest: This manuscript was prepared with the assistance of Roger Nutter and Scott Malkin of Gardiner-Caldwell Communications Ltd, Macclesfield, UK. Financial support was provided by F. Hoffmann-La Roche Ltd, Basel, Switzerland.