Amantadine, oseltamivir and zanamivir for the prophylaxis of influenza (including a review of existing guidance no. 67): a systematic review and economic evaluation

Description of intervention

Amantadine is a symmetrical C-10 primary amine with a cage-like structure, which is water soluble in hydrochloride salt form. 30 Amantadine hydrochloride exerts an antiviral effect on influenza type A by means of inhibition of the M2 ion channel, which results in the blocking of viral replication. 30 The antiviral activity of amantadine is restricted to influenza A. In addition, amantadine has weak dopamine agonist activity.

Licensed indications

Amantadine hydrochloride is indicated for:

• the treatment of and prophylaxis against signs and symptoms caused by influenza A infection (as Lysovir, Alliance Pharmaceuticals)

• the treatment of Parkinson’s disease (but not drug-induced extrapyramidal symptoms) (as Symmetrel^®, Alliance Pharmaceuticals)

• the treatment of herpes zoster (as Symmetrel).

Dosage and administration

Lysovir is available as reddish-brown, hard, gelatine capsules containing 100 mg amantadine hydrochloride, which are ingested orally. Symmetrel is available as 50 mg/5 ml syrup.

Contraindications

Amantadine hydrochloride is contraindicated in patients who:

• have epilepsy

• have a history of gastric ulceration

• have severe renal impairment

• are pregnant, wish to become pregnant or are breastfeeding

• have known hypersensitivity to amantadine or any excipients.

Cautions

Amantadine hydrochloride should be administered with caution to patients who:

• have hepatic impairment

• have renal impairment

• have congestive heart disease (as the drug may cause exacerbation of oedema)

• experience confusion or hallucinations

• have underlying psychiatric disorders

• are elderly

• are receiving concomitant medications with potential to affect the central nervous system (CNS).

Abrupt withdrawal of amantadine therapy should be avoided in patients with Parkinson’s disease.

It should be noted that, while resistance to amantadine is well documented,30 it has been reported that levels of resistance among influenza isolates have risen dramatically on an international scale. 31 Development of resistance can occur relatively rapidly during treatment and can lead to the failure of prophylaxis, for example within the management of outbreaks of influenza in long-term care settings. 32

Adverse events

Adverse events associated with amantadine hydrochloride include anorexia, nausea, nervousness, insomnia, dizziness, inability to concentrate, convulsions, hallucinations, blurred vision, GI effects, livedo reticularis, peripheral oedema and skin rashes. It has been documented that adverse effects can occur frequently among recipients. 33 Central nervous system adverse events have been described as occurring most notably within the elderly population.

Description of intervention

Oseltamivir is a neuraminidase inhibitor that exerts an antiviral effect on influenza A and B. 34 The drug inhibits viral release, preventing subsequent infection of adjacent cells. The SPC emphasises that oseltamivir is not a substitute for vaccination and that use should take into account official recommendations and variability of epidemiology and impact across patient populations and geographical locations.

Licensed indications

Oseltamivir is indicated for:

the post-exposure prophylaxis of influenza in patients aged 1 year and above who have had contact with a clinically diagnosed influenza index case when influenza is circulating in the community. The SPC states that the administration of oseltamivir should be decided on a case-by-case basis and that seasonal prophylaxis in subjects aged 1 year and above may be considered in exceptional circumstances (such as in the case of mismatch between vaccine and circulating strains of influenza or in the event of a pandemic).

the treatment of influenza in patients aged 1 year and above who present with influenza symptoms when influenza is circulating in the community. Treatment is effective when initiated within 48 hours of onset of the first symptoms.

Dosage and administration

Tamiflu is administered orally and is available as grey-yellow capsules containing 75 mg oseltamivir (as phosphate), 45 mg oseltamivir (as phosphate) or 30 mg oseltamivir (as phosphate), and as a powder (as phosphate) for reconstitution with water (12 mg/ml) as an oral suspension. The administration of 75 mg doses can be made up of one 75 mg capsule or one 30 mg capsule plus one 45 mg capsule or one 30 mg capsule plus one 45 mg dose of suspension. It should be noted that the BNF lists only the 75 mg dose of Tamiflu in capsule form. The administration of suspension is recommended in patients who are not able to swallow capsules. The SPC recommends that powder for oral suspension should be reconstituted by a pharmacist before it is dispensed to the patient.

Contraindications

Oseltamivir is contraindicated in patients who have hypersensitivity to oseltamivir or any of its excipients.

Cautions

Oseltamivir should be administered with caution to patients who:

• have renal impairment

• are pregnant or breastfeeding

• have conditions of such severity or instability that imminent hospitalisation may be required

• are immunocompromised

• have chronic cardiac and/or respiratory disease.

The dose should be reduced if creatinine clearance in patients is < 10–30 ml/minute and administration should be avoided if creatinine clearance is < 10 ml/minute.

Adverse events

Adverse events associated with oseltamivir include nausea, vomiting, abdominal pain, diarrhoea, dyspepsia, headache, fatigue, insomnia, dizziness, conjunctivitis, epistaxis, skin rashes, and – in very rare cases – hepatitis, Stevens–Johnson syndrome and toxic epidermal necrolysis. Neuropsychiatric disorders in children have also been reported.

Description of intervention

Zanamivir is a neuraminidase inhibitor that inhibits the replication of influenza A and B. 34 The SPC states that zanamivir is not a substitute for vaccination, as protection only lasts for as long as the drug is administered, and that the use of zanamivir should be decided on a case-by-case basis according to circumstances and the population in need of protection. The SPC recommends that the drug should be used only when reliable epidemiological data confirm the circulation of influenza in the community. Use of zanamivir should take into account official recommendations, epidemiological variation and varying impact of influenza across patient populations and geographical locations.

Licensed indications

Zanamivir is indicated for:

• the post-exposure prophylaxis of influenza A and B in adults and children aged 5 years and above who have had contact with a clinically diagnosed case of influenza in a household. Relenza may be considered for use in seasonal prophylaxis in exceptional circumstances, for example when there is mismatch between circulating or vaccine strains or in the event of a pandemic.

• the treatment of influenza A and B in adults and children aged 5 years and above who present with ILI when influenza is active in the community.

Dosage and administration

Relenza is available in the form of predispensed dry powder for inhalation in blisters containing 5 mg zanamivir per blister, delivered by means of oral inhalation using a Diskhaler^® device. Each inhalation delivered (quantity released via mouthpiece of the Diskhaler) contains 4 mg zanamivir (the remainder appears to be lost in the inhalation process and is presumably retained within the Diskhaler apparatus).

Contraindications

Zanamivir is contraindicated in patients who:

• are pregnant or breastfeeding

• are hypersensitive to any ingredient of the preparation.

Cautions

Zanamivir should be administered with caution to patients who:

• have asthma and chronic pulmonary disease

• have uncontrolled chronic illness

• are immunocompromised

• are pregnant.

According to the BNF, zanamivir should be used with caution in pregnancy and is contraindicated in breastfeeding women. However, according to the FDA, pregnancy and breastfeeding are cautions rather than contraindications. Other inhaled drugs, such as asthma medication, should be administered before zanamivir.

Adverse events

The following adverse events associated with zanamivir are described as occurring very rarely: bronchospasm, respiratory impairment, angioedema, urticaria and skin rashes.

Amantadine

A total of eight full papers reporting eight RCTs that investigated the prophylactic use of amantadine against influenza were identified. Characteristics of these studies are presented in Table 3.

The original HTA review reported by Turner et al. 10 assessed the use of amantadine in influenza prophylaxis in children (aged under 18 years) and the elderly (aged over 65 years) only, as a Cochrane review of the use of amantadine in adults had recently been reported. 50 This Cochrane review has been subsequently updated33 and was handsearched to identify any additional citations for inclusion in the current review. Turner et al. 10 identified three trials of amantadine prophylaxis undertaken in children. 51–53 However, these studies and an additional trial54 are not included in this technology assessment report, as the dosage of amantadine is not established in children under 10 years of age according to licensed indications. Two prevention trials in the elderly were also included in the original assessment. 55,56 Of these studies, only the findings presented by Pettersson and colleagues55 are included in this update, while the trial reported by Leeming56 was excluded, as twice the currently licensed dose was administered to participants.

An additional seven trials of amantadine prophylaxis were identified by our searches. These included two studies that evaluated seasonal prophylaxis in healthy adults. 57,58 Further trials described amantadine prophylaxis in outbreak control in healthy adolescents in a boarding school59 and in adults in semi-isolated engineering school populations. 60,61 A further three reports of the prophylactic efficacy of amantadine against experimentally-induced influenza were identified. 57,62,63 One of these papers presented results from two separate trials examining the use of amantadine in seasonal prophylaxis and experimentally-induced influenza studies. 57

Four trials included in the Cochrane review of amantadine and rimantadine in influenza A in adults33 have been included in our assessment. 55,57,59,61 Justifications of study exclusions are reported in Appendix 6.

An abstract was available in English for a double-blind, placebo-controlled trial by Plesnik et al.,38 which suggested that amantadine at 100 mg/day reduced the incidence of serologically-confirmed infection and was well tolerated. However, as the full text was not available in English, this citation could not be included and is not presented in the review.

Oseltamivir

Nine studies (of which six were reported in full papers and a further three were abstracts) were identified that investigated the use of oseltamivir in prophylaxis against influenza in six original RCTs. Characteristics of these trials can be seen in Table 4.

Four oseltamivir prevention trials were covered in the original HTA review;10 these were studies WV15825,64,65 WV15673,66 WV1569766 and WV15799. 49 Data for trials WV15673 and WV15697 were reported in the publication by Hayden et al. 66 both individually and combined across the two studies. All of these trials are included in the current assessment. An additional publication, by Hayden et al. ,47 examining the efficacy of oseltamivir in post-exposure prophylaxis within households present findings of an RCT published subsequent to the HTA review. 10 A further paper describes a trial of experimentally-induced influenza. 67 An additional publication45 describes a pooled analysis of data from oseltamivir post-exposure prophylaxis trials that are already included in the review. 46–49

An abstract in English was obtained for the trial by Kashiwagi et al. ,41 in which oseltamivir was administered to healthy adults at 75 mg once daily versus placebo for 6 weeks. This trial was previously reviewed by Jefferson et al. 68 However, the report was not available in full in English and was therefore excluded from this review.

Zanamivir

A total of 10 published reports of eight original RCTs were included in the assessment, of which eight were full papers and two were abstracts providing further reports of included studies. A further trial was identified within the sponsor submissions and is included, giving a total of nine RCTs. 44 These are presented in Table 5.

Turner and colleagues10 evaluated five zanamivir prevention trials: studies NAIA2010,69 NAIA3005,70,71 NAIA30010,46 NAIA200972 and NAIB2009. 72 NAIA2009 and NAIB2009 were reported as a single trial in the published literature. All of these trials are included in the present assessment, with the exception of trial NAIA2010 reported by Schilling et al. ,69 in which the dose of zanamivir used was twice that of current licensed indications.

An additional six citations relating to zanamivir were identified by the systematic searches for inclusion in the clinical effectiveness review. Findings from a trial of zanamivir seasonal prophylaxis in at-risk adolescents and adults have been presented. 75 A report on the use of zanamivir in post-exposure prophylaxis within households has also been published,47 while two additional papers and one abstract provide reports of the use of zanamivir in outbreak control in at-risk elderly subjects within long-term care settings. 76–78 An additional paper describes a pooled analysis of data from zanamivir post-exposure prophylaxis trials that are already included in the review. 45

Amantadine

The quality of the included eight RCTs relating to the prophylactic use of amantadine was relatively poor. No new amantadine prevention trials published since the original HTA assessment10 were identified. A considerable number of the older amantadine trials utilised a dose of 200 mg/day as opposed to the currently licensed adult dose of 100 mg/day18 and were therefore not considered to be suitable for inclusion in this review (see Inclusion and exclusion criteria, p. 17). Other amantadine prevention trials incorporated the use of doses in line with the current licence alongside inappropriate doses, but did not present data appropriate for inclusion separately and were therefore also excluded. Details of these studies can be found in Appendix 6.

Much of the amantadine prophylaxis evidence was not reported clearly, with a lack of detail on, for example, methods of randomisation of study subjects. 57,59–63 It was unclear in a number of trials whether allocation of treatment group was concealed. 55,57–59,62,63 One study publication failed to state clearly the number of participants randomised. 62 As only one report presented details of baseline characteristics of participants,55 it was generally not possible to assess whether baseline comparability between treatment groups had been achieved. It is therefore possible that potentially confounding variables may not have been adequately balanced among participants randomised to each trial arm. An additional four publications failed to state the eligibility criteria for participation in the trials. 55,60–62 A number of co-interventions were identified with the potential to affect outcomes, including vaccination,57–61 intake of medications that may affect study outcomes,64 and previous exposure to the experimental challenge strain. 62 The blinding of participants, those administering the intervention and outcome assessors was similarly difficult to judge and while many publications reported that a double-blind design was used, no further details were presented. Although all studies included at least 80% of randomised participants in the final analysis and only one study failed to report reasons for participants’ withdrawal,60,61 adherence to the intention-to-treat analysis was variable between studies.

Oseltamivir

The quality of the oseltamivir prophylaxis evidence presented was considerably more robust in terms of study design and reporting than that for amantadine. However, the randomisation methods used and concealment of allocation were unclear in the reporting of some studies. 48,49,73–74 All studies stated the number of participants randomised, and only one report failed to describe clearly the baseline characteristics and eligibility criteria,74 with all others judged to have achieved baseline comparability among subjects. A number of authors identified vaccination status,48,49,64,66 and recent use of antivirals48,64 or antibiotics48 as potentially confounding co-interventions. Clarity of reporting of blinding was variable among studies, and one study was described as being open-label in design. 48,73–78 All studies retained at least 80% of randomised subjects for use in the analysis and all, with the exception of two reports,64,67 presented reasons for withdrawal, but the analysis of only two studies could be considered to adhere to the intention-to-treat principle. 47,67,73^-74

Zanamivir

The evidence base for the use of zanamivir in prophylaxis against influenza could also be considered to have a greater degree of internal validity than the trials of amantadine prophylaxis. However, there was a lack of detail on methods of randomisation46,47,72 and concealment of allocation. 46,72,76,77 All studies outlined the number of subjects who were randomised to each group and described baseline characteristics, with baseline comparability considered to have been reached to varying degrees in all trials. Baseline comparability was considered to be relatively weaker in one trial. 78 Vaccination status46,47,70,75,76,78 and recent use of antivirals46,76,78 were identified as co-interventions. More information was available on blinding procedures used in the zanamivir research (with additional information obtained from sponsor submissions) than for oseltamivir and amantadine prophylaxis trials, although there were some gaps in reporting in a number of studies. 46,47,70–72,75 However, all studies included more than 80% of randomised subjects in analyses, described reasons for withdrawal and utilised intention-to-treat analysis.

Amantadine

The included evidence focusing on amantadine prophylaxis against influenza was taken from relatively old trials of lower quality that were conducted across a broad range of population subgroups. There was considerable variability between trials in terms of vaccination levels, setting and duration of prophylaxis. Eight references reporting eight RCTs were identified. The Cochrane review investigating amantadine and rimantadine in influenza A incorporated the use of meta-analysis in their study. 33 However, the large degree of heterogeneity and variation in primary outcomes used in terms of cases prevented between the studies included in our review would suggest that the use of statistical meta-analysis would be inappropriate; as such, the results of these trials are presented in the form of a narrative synthesis.

Evidence for amantadine prophylaxis in children under 10 years is not presented in this systematic review; such data were excluded as amantadine dosage is not established in this age group according to licensed indications. The limited evidence that exists relating specifically to this younger age group was reported within the original HTA review. 10 No clinical trial evidence relating to the use of amantadine in the paediatric population has been published subsequently.

Oseltamivir

Nine references reporting six original RCTs of oseltamivir for the prophylaxis of influenza were identified.

Zanamivir

Ten published articles presenting the results from eight RCTs were identified for inclusion in the systematic review of clinical effectiveness. An additional unpublished report was identified in the sponsor submissions and included in the assessment, resulting in a total of nine RCTs. The use of inhaled zanamivir only is considered within this assessment, hence trial arms in which doses of intranasal zanamivir were administered were excluded.

Roche submission to NICE

The Roche submission to NICE20 reports the use of a mathematical model to estimate the cost-effectiveness of oseltamivir for the seasonal prophylaxis and post-exposure prophylaxis of influenza. The cost-effectiveness model was submitted to NICE for scrutiny by the Assessment Group. The model presented within Roche’s submission is based on the simulating anti-influenza value and effectiveness (SAVE) model, and as such the structure and parameter set is similar to the model reported by Sander et al. 91 Twenty variations of the SAVE model were made available to the Assessment Group. The model compares oseltamivir prophylaxis with amantadine prophylaxis, zanamivir prophylaxis and no prophylaxis in the seasonal and post-exposure settings for four populations: otherwise healthy adults (including children > 12 years), at-risk adults (including children > 12 years), children aged 1–12 years and children aged 1–5 years. The analysis for children aged 1–5 years includes only usual care as a comparator for oseltamivir due to restrictions in the licensed indications of amantadine and zanamivir prophylaxis. It should also be noted that amantadine is licensed only in children aged 10 years or over; this prophylactic option is, however, included in the analysis for children aged 1–12 years. The base-case analysis was undertaken from the perspective of the NHS; secondary analysis was also reported from the societal perspective. The model is reported to use a lifetime horizon, whereby all important events occur within a 1-year time horizon with longer-term adjustments for quality-adjusted life-years (QALYs) lost as a result of premature death due to ILI. Cost-effectiveness is expressed in terms of the incremental cost per QALY gained, although this is based on pairwise comparisons of oseltamivir versus an alternative prophylactic option. In line with current recommendations from NICE,96 health outcomes were discounted at a rate of 3.5%; owing to the time frame used within the model, costs were not subjected to discounting.

The submission states that for oseltamivir versus amantadine and usual care, a cost-effectiveness analysis was undertaken. 20 The model assumes that oseltamivir and zanamivir are equivalent in terms of preventative efficacy and the submission reports a cost minimisation exercise for this comparison. However, the submission does not report the results of any head-to-head trials of zanamivir and oseltamivir prophylaxis (i.e. superiority, non-inferiority or equivalence trials) which provide any evidence to support the assumption of equivalence. Furthermore, the systematic review of clinical effectiveness presented in Chapter 3 did not identify any clinical evidence which could be considered to validate this assumption. Consequently, the use of a cost minimisation analysis for oseltamivir and zanamivir appears to be unjustified; even if equivalence trials were available, the comparative prophylactic effects would remain subject to uncertainty and should therefore be considered within the health economic analysis. Importantly, the Roche submission states that the preventative efficacy estimates have a considerable impact on the cost-effectiveness of oseltamivir prophylaxis. 20

Vaccination is not explicitly considered within the model, either as an option for influenza prevention or as a characteristic of the patient cohort. The studies used to estimate the preventative efficacy of zanamivir and oseltamivir included some patients who had been vaccinated and some patients who had not been vaccinated.

The model uses a deterministic decision tree approach which is reported to be appropriate as it captures a simple ILI pathway and events do not occur more than once. 20 The Roche submission argues that the results are conservative as the benefits of a contact case receiving prophylaxis and subsequently not infecting other individuals are not captured (herd immunity effects). The structural assumptions employed in the model are identical for seasonal and post-exposure prophylaxis settings. The model is reported to be based on ILI rather than true influenza alone, as it is intended to capture the impact of both true influenza and other ILI on costs and health outcomes.

The structures of the seasonal prophylaxis and post-exposure prophylaxis models are simple. For the post-exposure model, an individual who has been in contact with an ILI index case in a household may visit his or her GP to receive prophylaxis or may do nothing. For the seasonal prophylaxis model, the individual may or may not have been in contact with an index case when prophylaxis is initiated. The model assumes that one household member can obtain prescriptions for three contacts in the household. Contact cases may or may not go on to develop ILI. Individuals who develop ILI may be treated using oseltamivir (at-risk populations only) or usual care. Individuals who develop ILI may or may not develop complications. ILI complications are treated in an inpatient or outpatient setting depending on the severity of the complication. The model includes three complications: bronchitis, pneumonia and otitis media in children. Patients who develop ILI complications may survive or may die.

The model includes different attack rates for the seasonal prophylaxis models and for the post-exposure prophylaxis models; post-exposure attack rates are assumed to be higher than those for the seasonal prophylaxis models as contacts have by definition had previous exposure to an index case who may have influenza (Gavin Lewis, Head of Health Economics, Roche, personal communication). The submission states that the attack rates used in the post-exposure prophylaxis model are intended to represent the proportion of patients who, after being exposed to ILI, go on to develop ILI. 20 However, these are sourced from the oseltamivir post-exposure prophylaxis trial reported by Hayden et al. 48 and represent only laboratory-confirmed influenza, rather than all ILI. The attack rate for adults in the seasonal prophylaxis models was taken from Hayden et al. (assumed to be 4.8%). 66 The attack rate for children in the seasonal prophylaxis models was reported to be in the region of 10%,20 although the basis of this assumption is not reported in the submission. The methods used to derive upper and lower CIs around these attack rates are unclear from the submission.

The preventative efficacies of oseltamivir and zanamivir prophylaxis were sourced from a meta-analysis reported by Halloran et al. 45 The effectiveness of amantadine prophylaxis was derived from Monto et al. , although it should be noted that within this study patients received amantadine at a dose of 200 mg, which does not reflect its current licensed indications. 97 The model assumes that seasonal prophylaxis is effective across the whole influenza season; this is likely to be optimistic as patients may become susceptible to infection after they stop taking prophylaxis (see Chapter 3). Seasonal prophylaxis using zanamivir and oseltamivir are assumed to be equivalent to post-exposure prophylaxis using zanamivir and oseltamivir. The relative difference between amantadine as post-exposure prophylaxis and as seasonal prophylaxis was assumed to be the same as the relative difference for oseltamivir in each setting due to a lack of clinical trial evidence. The model does not include the possibility of resistance to amantadine, oseltamivir or zanamivir.

The probability of experiencing specific complications of ILI were sourced from a study reported by Meier et al. 12 It should be noted that these complication rates relate to ILI rather than true influenza alone (despite the claim that the model operates in terms of ILI, the Roche model actually appears to be based on true influenza attack rates). Complication rates due to influenza in children are assumed to be the same for both the 1–5 years age group and the 1–12 years age group. 20 The incidence of pneumonia and bronchitis was sourced from Meier et al. 12 However, the submission states that the incidence of otitis media is likely to be under-reported by Meier et al. Instead, the Roche model uses estimates sourced from oseltamivir clinical trial data;20 however, this estimate is only slightly higher than the estimate reported by Meier et al. (28% in Meier et al. versus 32.4% in the oseltamivir trials).

The probability of hospitalisation was taken from two US studies;98,99 these may not reflect UK practice. The model assumes that the probability of hospitalisation due to bronchitis is the same as that for other ILI. The probability of hospitalisation due to specific complications of ILI is assumed to be the same across the model populations. The model assumes the length of hospital stay to be 4 days for influenza and 7 days for pneumonia irrespective of patient population. The risk of death due to ILI is assumed to be the same as the risk of death due to ILI complications; this assumption is unlikely to be reasonable as ILI complications are known to increase the risk of death. It is likely that this assumption would overstate the benefits of avoiding a case of influenza.

The model includes HRQoL adjustments for individuals who develop influenza and complications of ILI. Utility estimates for patients experiencing an episode of influenza were derived from Likert valuations of patients with laboratory-confirmed influenza within the oseltamivir treatment trials. These rating scale data were converted to visual analogue scale (VAS) valuations and subsequently converted to time trade-off (TTO) utilities using a similar methodology to Turner et al. 10 Utility scores for patients with ILI, bronchitis and pneumonia were based on a Dutch person trade-off study reported by Stouthard et al. 100 Utility scores are applied for the duration of illness, based on clinical trial data (Gavin Lewis, Head of Health Economics, Roche, personal communication). In addition, the model includes the number of potential QALYs lost due to premature death resulting from ILI complications. Importantly, the model assumes that each potential year of life lost is valued at a state of perfect health; this assumption biases in favour of more effective prophylaxis options. The submission itself notes this assumption as a weakness of the model. 20

The model includes costs associated with drug acquisition, GP consultations, diagnostic tests, antibiotics and associated treatments, and hospitalisation for the treatment of ILI complications. Resource use estimates used in the model were derived from a variety of sources. Estimates of drug prescriptions, tests and investigations performed, primary and secondary care resource use for patients with influenza and certain complications were derived from the National Ambulatory Medical Care Survey (NAMCS);101 this is a US database, and may not reflect UK treatment patterns. Assumptions taken from this database were validated by Roche through a structured interview with one clinical expert. Sources for estimates of unit costs included the Personal Social Services Research Unit (PSSRU),102 the Monthly Index of Medical Specialties (MIMS) database,103 the MEDTAP database and the BNF. 14 Rates of antibiotic use were based on expert opinion.

Importantly, the model does not include the cost of drug wastage, and the cost of each prophylaxis course is calculated on the basis of the mean cost per tablet. The difference between the cost of oseltamivir with and without wastage is most pronounced in the seasonal prophylaxis indication for adults, these costs being £68.88 without wastage and £81.80 when wastage is included (see Modelling resource use and costs associated with influenza and other ILI, p. 76). Consequently, the acquisition cost of oseltamivir as seasonal prophylaxis is underestimated in the Roche submission. However, given the assumption of equivalence between oseltamivir and zanamivir, and the lower cost of a seasonal prophylaxis course using zanamivir, oseltamivir is actually dominated by zanamivir in this indication even when wastage is excluded.

The model assumes a single cost associated with hospitalisation due to ILI or ILI complications; this is quoted as £286 per day. This estimate is based on the cost of an inpatient day for mental health services; the justification for using this hospitalisation cost is unclear. 102 The model does not explicitly include the possibility of patients requiring intensive therapy unit (ITU) care or mechanical ventilation. A further potential problem with the SAVE model is that it assumes that all patients with ILI will incur GP consultation costs; this is not necessarily true as not all patients with ILI (whether influenza or not) will consult their GP. 104 Further, the model does not consider any costs associated with adverse events of prophylaxis or treatment using amantadine, oseltamivir or zanamivir.

The submission includes the details of one-way and probabilistic sensitivity analysis to explore uncertainty surrounding model parameters. The probabilistic sensitivity analysis was undertaken using @risk software alongside Microsoft excel.

Risebrough et al. – Economic evaluation of oseltamivir phosphate for post-exposure prophylaxis of influenza in long-term care facilities

Risebrough et al. 92 report the methods and results of a decision-analytic model to evaluate the cost-effectiveness of post-exposure prophylaxis versus no prophylaxis in long-term care facilities. The model includes three treatment options: post-exposure prophylaxis using oseltamivir, post-exposure prophylaxis using amantadine and no prophylaxis. The analysis was undertaken from the perspective of the single government payer in Canada. Zanamivir was excluded from the analysis because of difficulties in drug administration experienced by elderly patients. The primary health economic outcome for the analysis was reported to be the incremental cost per ILI case avoided compared with usual care (no prophylaxis); however, the model results are presented only in terms of costs and consequences which are not synthesised to produce incremental cost-effectiveness ratios. All patients are assumed to have received prior vaccination for influenza. The model uses a time horizon of 30 days, which is intended to represent the approximate duration of one institutional outbreak.

The model uses a decision tree structure to evaluate the costs and health outcomes associated with each of the three options. The first chance node relates to whether an outbreak occurs within the given care facility. Following an outbreak, patients in the prophylaxis arms begin post-exposure prophylaxis for 12 days using either amantadine or oseltamivir. For patients receiving amantadine, the model includes the possibility of developing amantadine resistance, while adverse events may be experienced by individuals receiving either prophylactic option. The model then includes the possibility that the individual develops ILI from which they may experience a complication, recover without complication, or die. If the ILI case is complicated, the patient may be treated in the care facility or, alternatively, may be transferred to hospital. The model does not include the expected effects of herd immunity. The model does not differentiate between specific complications experienced by individuals developing ILI. The incidence of ILI complications has an impact only on the cost side of the model; the impact of ILI and prophylaxis on HRQoL is not included in the economic analysis.

The authors assume an ILI attack rate in vaccinated residents of 17%. This estimate was reported to have been derived from a number of case–control studies and RCTs. The precise statistical methods used to derive this baseline attack rate (e.g. statistical meta-analysis) is unclear. The model does not include the possibility of patients receiving antiviral treatment following the onset of ILI. At the time of the analysis, the authors reported that there were no RCTs evaluating oseltamivir or amantadine as post-exposure prophylaxis in the nursing home setting. 91 Therefore, the authors assumed that post-exposure prophylaxis using oseltamivir would be at least as effective as seasonal prophylaxis using oseltamivir, and that amantadine would be at least as effective as rimantadine. Relative risk reductions in ILI incidence of 60% and 63% were assumed for amantadine and oseltamivir respectively. The authors assumed that prophylaxis using either amantadine or oseltamivir would result in a 50% relative reduction in antibiotic use, serious complications and death; no evidence is provided to support the validity of this assumption. The model includes the possibility of patients withdrawing from therapy as a result of the incidence of adverse events.

The model includes acquisition costs for amantadine and oseltamivir, serum creatinine tests and oral antibiotics, as well as the cost of hospitalisation for the management of influenza or other respiratory infections and the cost of hospitalisation due to adverse events. A cost is included for death resulting from ILI in an acute hospital. Dose adjustments are included in the cost of amantadine. Acquisition costs for amantadine were taken from the Ontario Drug Benefit Formulary, while the cost of oseltamivir was based on the manufacturer’s wholesale price. Serum creatinine test costs were taken from the Ministry of Health Schedule of Benefits. 106 The costs of hospitalisation due to adverse events were based on authors assumptions. The cost of transfer to an acute care facility for treatment of influenza complications was based on the average of all hospitalisations for influenza or other respiratory procedures per case mix group, derived from the Ontario Case Costing Initiative. 106 Higher costs were assigned to those complications that have potentially life-threatening complexity; the same cost was assumed irrespective of the patient’s outcome. Neither costs nor health outcomes were adjusted for time preferences.

The authors undertook one-way sensitivity analysis and best/worst-case scenario analysis, varying cost and event probability parameter values to identify the key determinants of cost-effectiveness. The sensitivity analysis explored the impact of changing assumptions concerning the relative efficacy of amantadine and oseltamivir versus placebo, the cost of serum creatinine testing, the incidence of adverse events, the attack rate for ILI, the outbreak rate and the rate of amantadine resistance. The sensitivity analysis also explored the impact of including the cost of nurse or pharmacist time to review the patient chart and to calculate the creatinine clearance. Finally, the cost-effectiveness of rimantadine was also explored in the sensitivity analysis. Probabilistic sensitivity analysis was not undertaken within this study.

In the base-case analysis, the study suggests that post-exposure prophylaxis using oseltamivir or amantadine is expected to reduce the incidence of ILI cases, hospitalisation and death compared with no prophylaxis. Both options are also expected to produce cost-savings as compared against no prophylaxis. When compared in terms of the incremental cost per ILI case avoided, oseltamivir is expected to dominate both amantadine and no prophylaxis. The sensitivity analysis suggests that the analysis is sensitive to the amantadine dose calculation. The use of alternative assumptions concerning the attack rate for ILI, the outbreak rate and the rate of amantadine resistance did not affect the base-case conclusions. The sensitivity analysis also suggested that if rimantadine were available in Canada, at 32% of the cost of oseltamivir, it would be the least expensive option; however, the authors suggest that oseltamivir would remain the most effective option. The worst-case scenario for amantadine resulted in improvements in ILI cases avoided, albeit at a greater cost than no prophylaxis. In the worst-case scenario, oseltamivir remained more effective and less costly compared with the amantadine and no prophylaxis options.

Turner et al. – Systematic review and economic decision modelling for the prevention of influenza A and B

Turner et al. 10 report the methods and results of a mathematical decision model to evaluate the cost-effectiveness of amantadine, zanamivir and oseltamivir in the prevention and treatment of influenza A and B. This study formed the assessment report used to inform the 2003 NICE appraisal of oseltamivir and amantadine for the prevention of influenza. 16 The analysis was undertaken from the perspective of the NHS, although reduced time from work is considered within the sensitivity analysis. The model includes eight preventative options: (1) no prophylaxis, (2) vaccination, (3) amantadine prophylaxis, (4) zanamivir prophylaxis, (5) oseltamivir prophylaxis, (6) vaccination plus amantadine prophylaxis, (7) vaccination plus zanamivir prophylaxis and (8) vaccination plus oseltamivir prophylaxis. All antiviral strategies relate to seasonal prophylaxis over a period of 6 weeks (42 days). Post-exposure prophylaxis using amantadine, oseltamivir and zanamivir are not included in the economic model; the model has since been adapted to examine the cost-effectiveness of post-exposure prophylaxis; however, the results of this work have not been released into the public domain. 107 The assessment report also evaluated the cost-effectiveness of treatment options for influenza A and B; however, these options are considered separately from the antiviral prophylaxis options. Cost-effectiveness is expressed in terms of the incremental cost per QALY gained and the incremental cost per influenza illness day avoided. The model uses a time horizon of a single influenza season, and includes QALY losses resulting from premature death due to influenza. The model estimates the cost-effectiveness of prophylaxis in four discrete subgroups: healthy adults, high-risk adults, children and residential care elderly.

The model uses a decision tree approach to evaluate the costs and health outcomes for each prophylactic option. Chance nodes are used to describe the probability of a patient developing influenza (dependent on the prophylaxis option), and QALY losses and costs are assigned to each branch. Costs and benefits for patients with influenza are modified for strategies including vaccination, on the basis that vaccination may reduce the severity of secondary complications. The model includes two complications: pneumonia and otitis media (the latter is included only in the paediatric model).

The model operates on the basis of true influenza rather than ILI. As treatments for influenza are evaluated separately from prophylaxis and vaccination options, the exclusion of ILI may be reasonable because costs and benefits in patients with ILI which is not true influenza are not expected to differ between prophylaxis options (and would therefore cancel each other out in the cost-effectiveness calculations). Baseline attack rates for true influenza were estimated using random-effects meta-analyses of placebo arm outcomes from relevant trials included in the systematic review. The preventative efficacy of each prophylaxis option was estimated by calculating the odds ratio of developing influenza, adjusted for the probability of compliance. The protective benefit of the prophylaxis options was assumed to apply only to the period over which patients are taking prophylaxis. The benefit of prophylaxis in vaccinated patients was assumed to be cumulative, such that the relative benefit of prophylaxis was applied to the baseline influenza attack rate excluding the expected number of cases protected by prior vaccination. The probability that an individual presents to the GP with influenza was based on a UK study of excess ILI consultations over a 10-year period reported by the Royal College of General Practitioners (RCGP)5 and the baseline influenza attack rate derived from the meta-analysis. 10 The probability of presentation was estimated by dividing the number of excess ARI consultations by the expected number of individuals who are expected to develop influenza in each population group. As the number of patients who present with true influenza is unknown, the numerator for this calculation was based on excess ARI consultations, assuming that all excess consultations are due to influenza. This approach, therefore, implies that the rate of non-influenza ILI consultations is constant over the year, and is likely to represent the maximum theoretical impact of influenza over a season. 5

The model includes HRQoL impacts associated with the incidence of influenza, adverse events resulting from the use of amantadine, the incidence of pneumonia and otitis media, and a QALY loss resulting from premature death due to complications. QALY losses due to influenza were derived from VAS scores collected in trials of oseltamivir for the treatment of influenza (studies WV15670, WV15671, WV15730, WV15819, WV15876, WV15978, WV15812 and WV15872). QALYs were derived by recalibrating Likert score data to VAS scores which were then converted into TTO scores. 104 QALY losses due to premature death were estimated on the basis of mean age of death due to influenza within the model subgroup, remaining life expectancy, age-specific utility scores and the discount rate. QALYs lost due to premature death were discounted at a rate of 1.5% in the base-case analysis in line with recommendations from NICE at the time of the assessment. The valuation of serious adverse events due to amantadine was based on an assumed EuroQol-5D (EQ-5D) profile. Adverse events resulting from the use of oseltamivir and zanamivir were assumed to have no impact on HRQoL. The valuation of secondary complications of influenza (pneumonia and otitis media) was based on WHO disability weights for lower respiratory conditions. 109

The model includes the costs associated with GP visits, prophylaxis and vaccination acquisition and inpatient hospital stays. The cost of a GP consultation in the surgery or at home was derived from the PSSRU; this cost was weighted by the frequency of home and surgery visits to generate a mean cost per visit for the elderly population and for the healthy adult population. The mean cost of a GP visit for the paediatric model was assumed to be the same as for the healthy adult model. The cost of antiviral prophylaxis was based on a 6-week course, assuming 50% of the recommended dose. Each drug cost was inflated to account for container fees and pharmacy prescribing fees, although these cost adjustments do not form part of NICE’s methods guidance. 96 The cost of vaccination was taken from payments to GPs for vaccination and included an administration cost. Hospitalisation costs were based on Health-care Resource Groups (HRGs); the HRGs assumed for hospitalisation differed according to the population under consideration. Owing to the short time horizon for the analysis, costs were not subjected to discounting.

Simple uncertainty analysis was undertaken using one-way and two-way sensitivity analyses surrounding the base-case model specification. This included varying assumptions in relation to influenza attack rates, the probability of death and the value of QALY losses due to premature death resulting from influenza complications. Joint uncertainty in model parameters was evaluated using probabilistic sensitivity analysis; parameter uncertainty was propagated through the model using Monte Carlo sampling techniques. However, results are presented as CIs surrounding the cost-effectiveness ratio; CEACs for prophylaxis are not presented in the report.

In the base-case analysis, amantadine, oseltamivir and zanamivir were dominated by vaccination. The combined option of amantadine plus vaccination yielded an incremental cost per QALY gained of £28,920 compared with vaccination alone in the residential care population. The incremental cost-effectiveness ratio (ICER) of amantadine for all other populations was considerably higher, ranging from £124,854 to £909,210. When adverse events were excluded from the model, the results of the probabilistic sensitivity analysis suggested that the probability that amantadine resulted in an incremental cost per QALY gained below £30,000 was around 45% for the elderly residential care population. However, this is a conservative assumption which favours amantadine. For the other populations, the probability that amantadine has an incremental cost per QALY gained below £30,000 was less than 1%. For the combined option of oseltamivir plus vaccination, the incremental cost per QALY gained for the residential population was £64,841 compared with vaccination alone. For all of the remaining populations, the ICERs were markedly less favourable, ranging from £251,004 to £1,693,168. The probabilistic sensitivity analysis suggested that the probability that oseltamivir has an incremental cost per QALY gained that is below £30,000 was 3% or less for all populations. Zanamivir was also dominated by vaccination. For the combined option of zanamivir plus vaccination, the incremental cost per QALY gained for the residential population was £84,682 compared with vaccination alone. The incremental cost per QALY gained ranged from £324,414 to £2,188,039 for the remaining populations. The uncertainty analysis suggested that the probability that zanamivir has an ICER that is below £30,000 per QALY gained was less than 1%.

Scuffham and West – Economic evaluation of strategies for the control and management of influenza in Europe

Scuffham and West93 report the use of a decision model to estimate the ICER of six influenza control strategies compared with no intervention in elderly populations in England, France and Germany. The options included in the model are opportunistic vaccination, comprehensive vaccination, chemoprophylaxis using oseltamivir, chemoprophylaxis using rimantadine, treatment using oseltamivir and treatment using rimantadine. The costs and health effects of zanamivir and amantadine were not included in the model. The analysis was undertaken from the perspective of the health-care financier for each country. The analysis reports marginal health economic outcomes in terms of the cost per hospitalisation averted, cost per death averted, cost per life-year gained and cost per morbidity day averted. The time horizon used within the model was a typical (average) influenza season.

The modelling approach adopted by the authors was not explicitly stated; however, the text indicates that a decision tree modelling methodology was employed. The model estimates the proportion of patients who develop clinical symptoms of ILI, a percentage of whom will visit their GP for treatment and may receive symptomatic treatment or antibiotics for complications of ILI. The model includes the possibility that patients who develop complications may require hospitalisation and the possibility that complications may lead to premature death. The model does not include any herd immunity effects associated with vaccination or prophylaxis.

The model includes the cost of hospitalisation due to complications including influenza and pneumonia, other ARI and congestive heart failure. The model does not include any valuation of the impact of influenza complications upon HRQoL, hence complications appear to be included in the model only in terms of costs avoided. The number of premature deaths due to influenza by age group was taken from a study by Fleming et al. 5 Based on UK hospitalisation data, the authors estimated the years of potential life lost for the healthy 80-year-old population to be 7 years; owing to the likely presence of co-morbidities, the authors assumed that premature death due to influenza would result in a mean loss of 3.5 potential years of life. The authors did not discount costs as almost all relevant events occur within a single influenza season. The potential life-years lost due to premature death resulting from secondary influenza complications was discounted at a rate of 1.5%.

The authors assumed an attack rate for ILI of 10%. This estimate was sourced from excess GP consultation rates, current rates of vaccination and expert opinion. Excess GP consultation rates were taken from a study based on national data collected by the Weekly Returns Service (WRS) of the RCGP and from national data for hospital admissions and deaths. 110 These are modelled independently of ILI attack rates. The probability of after-hours GP consultations was derived from expert opinion, while the percentage of GP home visits was taken from the UK population-based study of incidence, risk factors, complications and drug treatment of influenza reported by Meier et al. 12 The efficacy of chemoprophylaxis was taken from a review reported by Demicheli et al. 94 Based on this review, the authors assumed that neuraminidase inhibitors (NIs), specifically oseltamivir, reduce the incidence of influenza by 55%, while ion-channel inhibitors, specifically rimantadine, reduce the incidence of influenza by 35%. The authors assumed that when taken as prophylaxis, these therapies result in the same proportional reductions as vaccination in terms of GP consultation, hospitalisation and death. The model does not appear to include parameters describing the probability that a patient with ILI has true influenza. However, the estimates of the clinical efficacy of prophylaxis relate specifically to laboratory-confirmed influenza, not ILI. This appears to represent an inconsistency in the parameterisation of the model.

The model includes a number of different resource use items including GP consultations, after-hours visits and home visits, antibiotics, hospitalisations due to influenza and pneumonia, other respiratory illness and congestive heart failure, vaccination acquisition and administration costs, and antiviral prophylaxis and treatment. Unit costs were derived from the PSSRU,111 national sources of hospitalisation data,112 Department of Health publications on prescription costs113 and national tariff estimates. 114 The authors assumed that prophylaxis and treatment did not result in any adverse events. Non-compliance with prophylaxis was included in the model at a weekly rate of 5%.

The authors report the results of a large number of simple sensitivity analyses relevant to each option for the prevention and/or treatment of influenza. This included varying assumptions concerning the years of potential life lost resulting from premature death due to influenza complications, the discount rate for health outcomes, ILI attack rates, excess GP consultations, the number of excess hospital admissions for influenza complications and the number of premature deaths due to ILI complications. Specifically with regard to the prophylaxis options, the sensitivity analysis included varying assumptions regarding GP consultations to receive chemoprophylaxis, compliance rates, the dosage of oseltamivir, the percentage of prophylaxis used during the 4-week peak of the influenza season and drug price. Despite the extensive use of simple sensitivity analysis, the authors did not undertake probabilistic sensitivity analysis, and the impact of joint uncertainty in model parameters is not captured within the analysis.

Under the base-case assumptions, the authors report the marginal cost per life-year gained for oseltamivir to be €197,919 compared with no intervention. The cost per hospitalisation averted for oseltamivir is reported to be €114,774, while the cost per death averted is reported to be €657,544. The cost per morbidity day averted, excluding and including deaths, is reported to be €1198 and €373 respectively. The results of the sensitivity analysis are reported only in terms of the benefit : cost ratio (ratio of the strategy costs minus the costs of hospitalisation averted) and the cost per morbidity day averted. The findings of the sensitivity analysis based on the latter outcome measure are particularly difficult to interpret in a policy context. The analysis is reported to be most sensitive to changes in the timing of the programme, the price and dose of the prophylactic, and the assumed loss in potential life-years due to premature death.

Demicheli et al. – Prevention and early treatment of influenza in healthy adults

Demicheli and colleagues93 report the use of a model to estimate the cost-effectiveness and cost–utility of influenza prevention in healthy adults from the perspective of the Ministry of Defence (MOD). The health economic analysis was undertaken alongside three ongoing Cochrane reviews; the results of these reviews led to marked changes in the scope of the proposed economic analysis and the final economic models presented in the paper. 94 The authors state that potential preventative options to be evaluated within the final model were vaccination, oral amantadine, oral rimantadine and oral oseltamivir. However, costs and health outcomes are presented for three preventative options: vaccination, amantadine prophylaxis and a third option denoted ‘NI prophylaxis’. Although the authors justify the exclusion of zanamivir from the analysis because of trials apparently including only laboratory-confirmed outcomes, the exclusion of rimantadine is not justified within the paper, and the NI option is not directly specified as representing oseltamivir. The primary health economic outcome for the analysis was the incremental cost per avoided case. The time horizon used within the analysis was not explicitly reported; however, the analysis appears to relate to a single influenza season (i.e. a 1-year time horizon).

The authors adopted a decision tree approach to evaluate the differences in benefits and costs of the alternative options for the prevention of influenza. The authors report that they simplified an initially complicated decision tree model structure to include only the possibility of developing influenza and the possibility of experiencing adverse events due to prophylaxis. The model does not include the costs and health impacts of complications due to influenza or ILI and, as a consequence, the model does not include the possibility of death. It is reasonable to argue that the specification of this model is poor, as the results of the analysis ignore key costs and benefits associated with influenza prevention.

The model appears to operate in terms of true influenza cases rather than ILI cases, although this is not entirely clear. Influenza attack rates were derived from influenza sickness rates for 1997 obtained from the Defence Analytical Services Agency (DASA). The model assumes an incidence rate for influenza of 5.7 per 1000; while this value appears very low, incidence rates of up to 400 per 1000 were explored within the sensitivity analysis. The model does not include the possibility of a patient with symptomatic influenza presenting to a health-care professional for treatment. The effectiveness of the amantadine, NIs and vaccination were obtained from three Cochrane reviews of the clinical effectiveness of vaccination and prevention of influenza.

The model includes acquisition costs associated with influenza prevention, which were derived from the Defence Medical Supply Agency and authors’ assumptions. 94 No other cost components appear to be included in the results of the model. The impact of administration costs on overall cost-effectiveness is explored within the sensitivity analysis. A formal price year is not reported. The authors do not mention the use of discounting, which appears to be appropriate given the restrictive scope of the model (i.e. the exclusion of complications and death).

The authors undertook simple sensitivity analysis exploring the impact of improved/worsened preventative efficacy of vaccination and prophylaxis, improved adverse event profiles for vaccination and antiviral prophylactics, duration of prophylaxis and the inclusion of administration costs for prevention. Probabilistic sensitivity analysis was not undertaken in this study.

Costs and health outcomes are not reported in a disaggregated form, and it is difficult to establish whether the results are true incremental comparisons between the options, or whether they are compared marginally against a policy of no prevention. The text appears to indicate the latter to be the case. Under the base-case assumptions, the marginal cost per case avoided for vaccination, amantadine and NI (presumably oseltamivir) are reported to be £2807, £9458, and £88,193 respectively. The uncertainty analysis suggests that under most conditions vaccination is likely to be the most cost-effective option. The key determinant of cost-effectiveness appears to be the influenza incidence rate, for which higher rates are expected to result in more favourable cost-effectiveness ratios for vaccination and prophylaxis. The robustness and reliability of the results of this analysis are severely restricted by the limited scope of the model and the limited reporting of the economic evaluation.

Patriarca et al. – Prevention and control of type A influenza infections in nursing homes

This study95 reports the methods and results of a model of the cost-effectiveness of options for the prevention of influenza A in the elderly nursing home population. The model includes four options for the prevention of influenza A: vaccination without chemoprophylaxis, vaccination with amantadine post-exposure prophylaxis following an outbreak of influenza (30 days’ duration), amantadine post-exposure prophylaxis following an outbreak of influenza (30 days’ duration) with no prior vaccination, and amantadine as seasonal prophylaxis (3 months’ duration) with no prior vaccination. All options are compared with a strategy of no control. Cost-effectiveness is expressed in terms of the incremental cost per illness averted, the incremental cost per hospitalisation averted and the incremental cost per death averted. The perspective of the analysis is not explicitly reported; however, the authors state that only direct costs were included in the analysis. The time horizon for the analysis is unclear; however, the authors state that they did not include future medical costs associated with deaths averted.

The authors used a decision tree model to evaluate the incremental costs and health outcomes for each preventative option. Chance nodes are used to describe the probability that an individual is immune or susceptible to influenza A, the probability of community exposure, the efficacy of vaccination, the possibility of a nursing home outbreak and the possibility that an individual will or will not become ill. Patients who become ill experience one of four possible outcomes: infection and survive, infection and die, hospitalisation and survive or hospitalisation and die. The model is reported to include the impact of herd immunity although the precise methods for including this factor are unclear. Respiratory complications only are included in the model.

The incidence of disease during the course of an outbreak was based on the experience of 41 separate vaccine efficacy studies conducted in nursing homes during the period 1972–85. The probability of an outbreak was estimated according to the results of a case–control study;115 this probability was adjusted for the vaccination and chemoprophylaxis options to account for herd immunity effects. The model assumes an overall attack rate of 43% during influenza outbreaks and 16% at other times. The model does not include the possibility of antiviral treatment for patients who develop ILI. The authors assumed that 80% of residents who completed the course of chemoprophylaxis would be fully protected. The probability of recovery/death with or without hospitalisation following influenza infection for patients receiving amantadine prophylaxis was assumed to be the same as for vaccination. More favourable outcomes were assumed for patients who received both vaccination and prophylaxis, although this was reported to be based on only limited clinical evidence. The impact of adverse events is not included in the effectiveness aspect of the model.

The model includes costs associated with vaccination, acquisition costs for amantadine prophylaxis and costs of diagnostic tests, treatments, ambulance and hospitalisation for influenza infections and associated complications. Administrative costs were excluded from the analysis for the chemoprophylaxis options, but were included for vaccination. The authors state that adverse events associated with amantadine are not associated with excess medical care costs; however, the authors did include the costs of treating fractures and soft-tissue injuries resulting from dizziness or postural hypotension for patients receiving amantadine. Costs of influenza infections and associated complications were sourced from 1986 prospective payment schedules for appropriate diagnosis-related groups and other sources. Physician charges were based on Medicare Part B payments. The authors do not make any reference to the use of discounting within the analysis.

One-way and multiway sensitivity analyses were undertaken surrounding the efficacy of influenza vaccination, the efficacy of chemoprophylaxis, and assumptions concerning risk reductions in hospitalisation and death for patients receiving prophylaxis. The authors also undertook a threshold analysis to determine how much amantadine and vaccination would have to cost before these options would no longer result in savings in direct medical costs. Finally, the authors explored the impact on cost-effectiveness of changing the exposure rate to influenza viruses. Probabilistic sensitivity analysis was not undertaken.

The option of outbreak prophylaxis was excluded from the analysis as it was the least effective and most expensive programme. 95 Marginal cost-effectiveness ratios are presented for vaccination plus chemoprophylaxis versus vaccination alone, continuous chemoprophylaxis versus vaccination alone, and continuous chemoprophylaxis versus vaccination plus chemoprophylaxis. The combination of vaccination and chemoprophylaxis during an outbreak was reported to result in demonstrable improvements in outcome at a modest increase in cost. However, the cost-effectiveness calculations include only the program costs, and do not account for expected cost savings in medical care costs. This omission biases against more effective prevention options. The authors report that changing assumptions regarding efficacy and the risk of hospitalisation and death exerted only minor or negligible effects on the clinical and economic outputs of the model. The authors report that varying exposure to influenza led to a proportionate reduction in the number of cases and a subsequent reduction in the cost-effectiveness of each programme. Increasing the level of coverage of vaccination and chemoprophylaxis led to a progressive decline in morbidity and increases in cost-effectiveness.

Interventions and comparators

The model evaluates the incremental costs and health outcomes of seasonal prophylaxis and post-exposure prophylaxis of influenza using amantadine, oseltamivir and zanamivir in comparison with each other and no prophylaxis.

Model population

Cost-effectiveness estimates for influenza prophylaxis using oseltamivir, amantadine and zanamivir are presented for six discrete subgroups: children aged 1–14 years (with at-risk medical condition or otherwise healthy), adults aged 15–64 years (with at-risk medical condition or otherwise healthy) and elderly adults aged over 65 years (with at-risk medical condition or otherwise healthy). In addition, the analysis considers the impact of prophylaxis for individuals who have been vaccinated against influenza and for individuals who have not been previously vaccinated. Although the model structure is identical for all subgroups, the analyses differ in terms of influenza attack rates, prophylaxis dose, prophylactic efficacy and prognosis following influenza onset.

Health economic outcomes

The primary health economic outcome used in the economic model is the incremental cost per QALY gained. This is calculated for all non-dominated prophylactic options compared with the next most effective option. Options that are dominated (simple or extended) are ruled out of the analysis.

Time horizon and time preferences

The model assumes that all events of interest occur within a single influenza season; hence, the time horizon is effectively 1 year in duration. As such, costs and health outcomes arising within this period are not subjected to discounting. However, as secondary complications of influenza and other ILI may result in premature death, the model also accounts for potential years of life lost beyond this time horizon; these are adjusted to account for the expected level of quality of life. Quality adjusted life-years lost because of premature death resulting from the incidence of influenza-related complications are discounted at a rate of 3.5%, in line with current recommendations from NICE. 96 A summary of the scope of the economic comparisons is presented in Table 27 (note that the duration of prophylaxis is assumed to be in line with licensed indications).

Key model assumptions

• Other ILI which is not influenza may also result in complications (including RSV and Mycoplasma pneumoniae). It should be noted that the complications arising from influenza may differ in reality from those for other ILI such as RSV (this is a limitation in the use of the data from Meier et al. ;12 see Parameters relating to the onset of influenza and other ILI, below). However, since the costs and effects associated with other ILI are the same for each prophylaxis group, these do not affect the resulting estimates of cost-effectiveness.

• Prophylaxis using amantadine, oseltamivir and zanamivir are effective only against true influenza.

• Antiviral prophylaxis is effective in preventing influenza only for the period over which the patient is taking the drug. For seasonal prophylaxis, the model assumes that a patient may be protected over a proportion of the whole influenza season. However, it should be noted that monitoring of influenza activity takes place at a national level and the duration for which activity exceeds the national threshold may not reflect influenza activity at the local level. The importance of this assumption is tested in the sensitivity analysis (see One-way/multiway sensitivity analysis and scenario analysis, p. 90). For the sake of simplicity, the model assumes that the risk of infection is constant for the period when influenza is circulating; this is in line with the previous models, reviewed in Systematic review of existing cost-effectiveness evidence (p. 43).

• The joint benefit of vaccination followed by prophylaxis is assumed to be cumulative (the effectiveness of prophylaxis is applied to any remaining influenza cases which are not effectively protected by vaccination).

• The model assumes that amantadine, oseltamivir and zanamivir would be used as prophylaxis when influenza is known to be circulating in the community (the threshold is currently set at 30 new ILI GP consultations per 100,000 population). 8

• The model assumes that the prescription of seasonal prophylaxis and post-exposure prophylaxis of influenza requires a consultation with a GP. The possibility of multiple courses of antiviral prophylaxis being prescribed to an index case on behalf of other household contacts is explored in the sensitivity analysis. The model assumes that prophylaxis is not given at the same time as influenza vaccination, hence a second visit is required.

• If an individual develops a secondary complication of ILI (whether or not this is due to influenza), the course of the complication is unaffected by the prior use of prophylaxis. Treatment of symptomatic influenza using oseltamivir or zanamivir is assumed to reduce the incidence of complications in at-risk patients. If a patient has already developed a complication while receiving prophylaxis, it is unlikely that antiviral treatment will provide any additional benefit. Given the simple structure of the model, the analysis assumes that patients who receive antiviral prophylaxis and subsequent treatment for symptomatic ILI develop complications after being prescribed treatment. This assumption is likely to favour prophylaxis as it increases the costs of treating symptomatic influenza. Assumptions surrounding the use of antiviral treatment following prophylaxis are explored in the sensitivity analysis.

• Patients who experience adverse events due to prophylaxis are likely to consult their GP for advice.

• Adverse events due to oseltamivir and zanamivir are mild, self-limiting and have no impact on a patient’s HRQoL. Adverse events due to amantadine prophylaxis may be more severe and may result in a reduction in the patient’s quality of life.

• Antiviral treatment of symptomatic influenza and ILI using zanamivir and oseltamivir is given in line with current NICE recommendations. 116 The choice of NI for the treatment of symptomatic ILI is assumed to be independent of the prophylactic strategy under consideration. Antiviral treatment is assumed to incur an additional cost in patients who have previously received prophylaxis. For example, if a patient is prescribed oseltamivir prophylaxis, subsequently develops symptomatic ILI and is given oseltamivir treatment, a separate prescription of the drug is required.

• All patients who develop complications due to influenza and other ILI present to a health-care professional for treatment.

• Patients who develop either uncomplicated or complicated ILI may be prescribed antibiotics.

• Patients who stop taking prophylaxis are assumed to do so at the beginning of the course and hence do not gain any additional protection over patients who do not receive prophylaxis (the impact of assumptions regarding withdrawal rates are explored in the sensitivity analysis – see One-way/multiway sensitivity analysis and scenario analysis, p. 90).

• The costs of diagnostic tests (blood tests, sputum tests, chest X-ray) for patients presenting with respiratory complications are assumed to be included in the unit costs of GP consultation and A&E consultation.

• Owing to limitations in the evidence base, the model assumes that only complicated ILI cases may result in hospitalisation and death. These assumptions are explored in the sensitivity analysis (see One-way/multiway sensitivity analysis and scenario analysis, p. 90).

• The model includes only those health benefits accrued by patients receiving influenza prophylaxis; potential benefits accrued through decreased transmission of influenza as a result of the use of prophylaxis are not considered in the health economic model.

• A proportion of influenza cases are assumed to be resistant to amantadine. Although there is some evidence of resistance for the NIs, these rates are low and are excluded from the base-case analysis. The impact of resistance to oseltamivir is considered within the sensitivity analysis (see One-way/multiway sensitivity analysis and scenario analysis, p. 90).

Baseline influenza attack rate

The baseline influenza attack rate describes the probability that an individual will develop influenza over the influenza season. The model assumes that the probability of developing influenza differs among children, adults and elderly individuals. Different attack rates are also assumed between the seasonal and post-exposure prophylaxis models, as probability of influenza infection is likely to be higher in an individual who has been in frequent close contact with an index case with symptomatic ILI in the household. In terms of seasonal prophylaxis, the clinical trials included in this review do not represent a good basis for estimating the probability of developing influenza as they include different levels of exposure to influenza vaccination across each subgroup; one would expect that this would result in lower attack rates than in the unvaccinated population. For the seasonal prophylaxis model, influenza attack rates were derived from a large meta-analysis of placebo arm groups of clinical trials of influenza vaccination versus no influenza vaccination reported by Turner et al. 10 The model uses the actual patient numbers presented in the summary of each meta-analysis to estimate the mean and distribution of the attack rate. Beta distributions were used to describe the uncertainty surrounding these parameters.

This source does not, however, provide a useful basis for estimating attack rates for the post-exposure prophylaxis models, as individuals eligible for post-exposure prophylaxis have, by definition, been exposed to an index case with symptomatic influenza or ILI. Consequently, one would expect the attack rate for these individuals to be higher than the attack rate in an individual who has not been exposed to an index case. Attack rates for the post-exposure prophylaxis options were sourced from the trials of post-exposure prophylaxis included in the systematic review (see Chapter 3). For the paediatric subgroup the attack rate was taken directly from the subgroup analysis reported by Hayden et al. ,48 as this was the only study which presented a subgroup analysis for the paediatric population. For the working-age adult and elderly populations, the attack rate was taken from a pooled analysis of placebo group attack rates reported in five trials of post-exposure prophylaxis. 46–49,72 It should be noted that patient-level data were not available, hence these attack rates relate to populations that are mixed in terms of subject age. Beta distributions were used to characterise the uncertainty surrounding these attack rates. The attack rates are presented in Table 28.

Probability that an ILI is influenza

The probability of developing ILI during the influenza season was not available from the literature. Instead, the model uses data provided by the RCGP concerning the probability that a case of ILI is true influenza. Within the health economic model, this probability is divided by the true influenza attack rate to provide an estimate of the broader ILI attack rate in each subgroup (accounting for true influenza and other ILIs). Data relating to the probability that ILI is influenza was based on an analysis of swabs taken from individuals with symptomatic ILI collected during routine surveillance over the influenza seasons 2003–4 to 2006–7 (Dr Alex Elliott, RCGP, personal communication). These data relate to those weeks when influenza was known to be circulating in the community, as defined by the 30/100,000 ILI GP consultation threshold;8 they are shown in Table 29.

Table 29 suggests that one would expect fewer influenza cases among the ILI cases when the consultation rate falls to baseline levels. The model assumes that the probability that ILI is true influenza is 0.50 across all subgroups (622/1256). Uncertainty surrounding this parameter was modelled using a beta distribution.

Probability that influenza is influenza A

The probability that a case of influenza is influenza A is based on virological surveillance data provided by the HPA (Dr Piers Mook, HPA, personal communication). These data relate to 12 influenza seasons from 1995–6 to 2006–7; they are shown in Table 30.

The probability that influenza A is the dominant influenza strain during a given influenza season was calculated from the data shown in Table 30; this gives a probability of 0.75 (influenza B is assumed to be dominant during the 2002–3 season). The probability that a case of influenza is influenza A was then modelled separately for those years where influenza A is dominant and those where influenza B is dominant. For years in which influenza A is dominant, the probability that an influenza case is influenza A was estimated to be 0.86 (740/859). For years in which influenza B is dominant, the probability that an influenza case is influenza A was estimated to be 0.30 (83/281). The overall mean probability that a case of influenza is influenza A is estimated to be 0.72. Beta distributions were used to characterise the uncertainty surrounding the probability that influenza A is dominant and the probability that an influenza case is influenza A given the dominant influenza strain during a given influenza season. These data are used to modify the effectiveness of amantadine which is effective only against influenza A.

Preventative efficacy of prophylaxis using amantadine

The systematic review of clinical effectiveness presented in Chapter 3 highlighted a paucity of evidence relating to the efficacy of amantadine in both the seasonal and post-exposure prophylaxis settings. Two studies were available relating to the seasonal prophylaxis of influenza using amantadine;57,58 data relating to the relative protective benefit of amantadine compared with placebo was available only from the study reported by Reuman et al. 57 This study included healthy adults who had not been vaccinated; a mean RR of 0.40 (95% CI 0.08–2.03) was estimated from the event data reported within the clinical trial publication. Owing to the absence of additional or alternative studies, this parameter estimate was applied to all subgroups in the seasonal prophylaxis model, hence the model assumes that the preventative efficacy of amantadine is independent of age and risk status. It should be noted that the systematic searches did not identify any direct evidence of the benefit of amantadine in the paediatric population in line with licensed indications (see Chapter 3); therefore we have extrapolated efficacy estimates from the adult population.

The systematic review did not identify any clinical trials of the effectiveness of amantadine in the post-exposure prophylaxis setting within households (see Chapter 3). However, one study was identified which examined the efficacy of amantadine in outbreak control in healthy adolescents in a boarding school over a period of 14 days. 59 The majority of subjects recruited within this study had been previously vaccinated for influenza. Prior vaccination does not necessarily confound the analysis of the efficacy of prophylaxis; however, it is likely that the presence of effective vaccination would reduce the statistical power of the trial comparison (as a result of lower attack rates in both prophylaxis and placebo groups). Efficacy estimates within the outbreak control setting were assumed to be similar to those for amantadine when used as post-exposure prophylaxis, as the duration of prophylaxis is similar (assuming post-exposure prophylaxis using amantadine would be taken for a duration of 10 days). Based on the event data reported in the clinical trial publication, the RR of amantadine versus placebo was estimated to be 0.10 (95% CI 0.03–0.34). Owing to a lack of any alternative evidence, this RR was applied to all subgroups in the model.

The model assumes that a proportion of patients develop amantadine-resistant disease; these patients are assumed to derive no prophylactic benefit from amantadine. Surveillance data (provided as academic-in-confidence) were provided by the HPA regarding the proportion of H1N1 and H3N2 isolates that were resistant to amantadine during the years 2004–7. Resistance may occur in either strain; recent data suggest that amantadine resistance is considerably higher in the H3N2 strain. Based on the data for the 2006–7 influenza season, the model assumes that 37% of influenza A cases are resistant to amantadine. This proportion is a crude estimate based on the experience over a single influenza season and may vary considerably as resistance levels and the ratio of H3N2 and H1N1 strains vary from year to year.

Preventative efficacy of prophylaxis using oseltamivir

The systematic review of clinical effectiveness presented in Chapter 3 identified a more substantial evidence base relating to the effectiveness of oseltamivir in the prophylaxis of influenza. Two studies of seasonal prophylaxis using oseltamivir were identified;64,66 one study66 recruited healthy adults (unvaccinated), while the other trial recruited at-risk elderly subjects in a residential home (> 80% of subjects vaccinated in intervention and control groups). 64 The study reported by Hayden et al. 66 was applied to the otherwise healthy and at-risk paediatric and working-age adult populations, while preventative efficacy estimates from the study reported by Peters et al. 64 were applied to the otherwise healthy and at-risk elderly populations. Based on event data reported by Hayden et al. ,66 the RR of developing influenza was estimated to be 0.24 (95% CI 0.10–0.58). Analysis of event data reported by Peters et al. 64 suggested an RR of developing influenza of 0.08 (95% CI 0.01–0.63). It is unclear whether the difference between efficacy rates from these two trials is a result of differences in terms of study population, underlying risk or another unknown source of heterogeneity.

Two studies were identified which evaluated the preventative efficacy of oseltamivir in the post-exposure prophylaxis of influenza. 48,49 The preventative efficacy of oseltamivir for the healthy adult group was based on a random-effects meta-analysis of these two studies; the mean RR used in the model was estimated to be 0.19 (95% CI 0.08–0.45). Importantly, within the two trials included in the meta-analysis one trial included paediatric and adult subjects48 while the other included only adult subjects. 49 Owing to a paucity of alternative evidence, this RR was applied to all otherwise healthy and at-risk adult populations. In the paediatric population, the RR of developing influenza following oseltamivir post-exposure prophylaxis was modelled on the subgroup analysis reported by Hayden et al. ;48 the mean RR of developing influenza for children was 0.36 (95% CI 0.16–0.80). This RR was applied to both the otherwise healthy and at-risk paediatric subgroups.

Preventative efficacy of prophylaxis using zanamivir

The systematic review identified two clinical trials relating to the benefit of zanamivir for the seasonal prophylaxis of influenza. 70,75 The study reported by Monto et al. 70 recruited healthy adults, the majority of whom were unvaccinated. The study reported by La Force et al. 74 recruited at-risk adults; subjects recruited into this study had a higher level of vaccination. Based on the event data reported in the clinical trial paper, the RR of developing influenza in healthy adults was estimated to be 0.32 (95% CI 0.17–0.63). 70 This estimate was applied to the otherwise healthy and at-risk children subgroups as well as to the healthy adult subgroup. Similarly, the RR of developing influenza in at-risk adults was estimated to be 0.17 (95% CI 0.06–0.50); this RR was applied to the at-risk adult working age subgroup. 75 The RR for the elderly populations was based on a subgroup analysis reported by LaForce et al. ;75 this RR was estimated to be 0.20 (95% CI 0.02–1.72).

The review identified three trials which reported the clinical efficacy of zanamivir versus placebo for the post-exposure prophylaxis of influenza in adults72 and children and adults. 46,47 The RR of developing influenza in all subgroups receiving zanamivir was estimated using a random-effects meta-analysis of these three trials; the RR was estimated to be 0.21 (95% CI 0.13–0.33). One study did evaluate zanamivir as outbreak control in largely at-risk elderly subjects;76 the model does not use efficacy data from this study because of differences in the duration of prophylaxis. The use of the meta-analysis estimate for zanamivir in post-exposure prophylaxis in households represents a bias in favour of zanamivir in this subgroup.

Relative risks and 95% CIs (shown in parentheses) used in the model are summarised in Table 32. The footnotes detail whether each RR is based on trial evidence relating exclusively to the model subgroup, trial evidence that includes the subgroup or trial evidence relating to other subgroups.

It should be noted that the evidence surrounding the effectiveness of amantadine, oseltamivir and zanamivir within specific subgroups is not ideal, and decisions regarding the appropriate inclusion of specific preventative efficacy estimates are not straightforward. For the most part, preventative efficacy is assumed to be the same across a number of age and risk subgroups (even those where there is no trial evidence relating to the subgroup under consideration, e.g. amantadine post-exposure prophylaxis in the elderly). In other instances, where multiple sources exist, there are known heterogeneities between study populations (age, risk status, level of prior vaccination), methods of end-point measurement and duration of prophylaxis. It is unclear whether differences observed in these preventative efficacy estimates are a result of one or a combination of these known heterogeneities or some other underlying differences between the studies. The uncertainty surrounding all RRs of developing influenza for vaccination and prophylaxis was modelled using lognormal distributions; estimates of preventative efficacy were sampled from a normal distribution characterised by the logmean RR and the standard error of the log of the RR. The reader should be aware that there is likely to be a greater level of uncertainty surrounding these effectiveness estimates than the uncertainty reflected in data from the studies included in the systematic review.

Probability of an individual with ILI presenting symptomatically

There is considerable uncertainty surrounding the probability that an individual with ILI will consult a health-care professional in either primary or secondary care. The model reported by Turner et al. 10 used evidence from a study of the excess GP consultations reported by Fleming. 112 The use of these data implies the assumption that all excess GP consultations over the influenza season compared with the baseline rate are due to influenza. The validity of this assumption is questionable,122 as other ILIs such as RSV are often more prevalent during the influenza season, thus accounting for an unknown proportion of excess cases between the influenza season and baseline periods. Instead, the ILI consultation rate was based on a European ILI surveillance study reported by van Noort et al. 104 This study used an internet-based approach to monitoring ILI symptoms and consultations in the general population in the Netherlands, Belgium and Portugal. The study reported highly variable consultation rates for individuals with ILI ranging from 25% to 67%. The model assumes that the true probability that an individual with symptomatic ILI will present is likely to be at the lower end of this range. The model assumes a central estimate of 0.25; uncertainty surrounding this parameter value was modelled using a beta distribution assuming a subjectively large standard error (alpha = 5, beta = 15). The probability of presentation with ILI is assumed to be the same for all subgroups included in the model. The impact of this assumption is explored in the sensitivity analysis (see One-way/multiway sensitivity analysis and scenario analysis, p. 90).

Probability of an individual presenting within 48 hours of symptomatic onset of ILI

Treatment using oseltamivir is currently recommended only for those individuals who are considered to be at high risk of developing complications who present within 48 hours of symptomatic onset. 116 The probability of an individual presenting with ILI within 48 hours of onset was derived from a study reported by Ross et al. 123 The model assumes that half of those presenting on day 2 would be within the 48-hour cut-off; this assumption is in line with the previous model reported by Turner et al. 10 In this study, the probability of presentation within 48 hours was reported to be 52%, 16% and 11% in the paediatric, working-age adult and elderly populations respectively. These probabilities are assumed to be the same in otherwise healthy and at-risk populations. The uncertainty surrounding these parameters was modelled using beta distributions based on the empirical data reported by Ross et al. 123

Probability that an individual presenting within 48 hours is prescribed an NI for the treatment of ILI

In line with current recommendations from NICE concerning the use of NIs for the treatment of influenza and other ILI, the model assumes that oseltamivir and zanamivir are prescribed only for patients who are at risk of secondary complications of ILI (including elderly patients over 65 years of age). For the paediatric population who are eligible for treatment, the model assumes that patients are treated using oseltamivir. For at-risk adult populations, the model assumes that 89% of patients receive oseltamivir, based on data reported within the submission to NICE by Roche. 20 The remaining 11% of patients are assumed to receive treatment using zanamivir.

Probability of developing complications due to influenza and other ILI

The incidence of complications associated with influenza and ILI is not reported in detail within clinical trials of influenza prophylaxis (see Chapter 3). Instead, the probability of developing a complication of influenza or other ILI was taken directly from a large UK-based observational study reported by Meier et al. 12 This study collected and analysed data concerning the incidence, risk factors, clinical complications and drug utilisation associated with influenza and ILI using data collected in the GPRD in the period 1991–6. A total of 141,293 patients in the database were reported to have one or more diagnoses of influenza or ILI. Data concerning the incidence of specific complications, including exacerbations of underlying diseases and death due to influenza, were reported by age group (1–14 years, 15–49 years, 50–64 years, and > 65 years) and by presence of pre-existing chronic diseases. The rates of specific complications reported by Meier et al. 12 are shown in Table 33.

Data concerning complication rates for the predisposed group were assumed to represent the at-risk populations within the model. Complication rates among the 15–49 year age group and the 50–64 year age group were combined to represent the working-age adult model populations. Uncertainty surrounding the probability of experiencing a complication of influenza within each population group was modelled using beta distributions, while the multinomial probabilities of experiencing specific complications were modelled using Dirichlet distributions with minimally informative priors based on the methods reported by Briggs et al. 124 The model assumes that the risk of developing complications is the same for influenza and other ILI.

It should be noted that the use of this study is flawed in that many of the ILIs reported by Meier et al. will be caused by viruses other than influenza. This problem is compounded further as the study reported ILI complications over the whole year rather than the influenza season, hence the proportion of episodes caused by other ILIs is likely to be higher than that for the period when influenza is known to be circulating. A limitation of these data, and their use in the model, is that the rates of complications resulting from other ILIs which are not influenza may not reflect complication rates due to influenza infection.

Effectiveness of NIs for the treatment of symptomatic influenza

The efficacy and safety of oseltamivir and zanamivir for the treatment of influenza and other ILI is beyond the scope of this assessment and is scheduled for reappraisal in 2008. However, both zanamivir and oseltamivir are currently recommended for treatment of symptomatic influenza and ILI in at-risk individuals. 116 Evidence concerning the safety and efficacy of the NIs for the treatment of ILI was derived from the earlier HTA report by Turner et al. 10 The model assumes that oseltamivir and zanamivir reduce the probability of experiencing complications due to influenza and lead to a modest reduction in the impact of influenza on quality of life compared with best symptomatic care alone. The model assumes an odds ratio for all complications for zanamivir versus no treatment of 0.49 (95% CI 0.23–1.04) in all at-risk populations, while the odds ratio for complications for oseltamivir versus no treatment is assumed to be 0.65 (95% CI 0.43, 0.97) in the at-risk paediatric population and 0.40 (95% CI 0.16–0.93) in at-risk adult and elderly populations. 10 The model assumes that the NIs are not effective in reducing complications due to other ILIs which are not influenza. The odds ratios derived from Turner et al. 10 relate to reductions in complications requiring antibiotics; owing to the high rates of antibiotic use for the treatment of ILI-related complications,12 and the absence of alternative evidence, the model assumes that these efficacy rates are applied to all ILI-related complications. It is possible that this may overstate the benefit of zanamivir and oseltamivir in terms of reducing complications due to influenza and other ILI. A summary of treatment efficacy values assumed within the model is shown in Table 34.

As noted in Model structure (above), the model assumes that the use of neuraminidase inhibitors for the treatment of symptomatic influenza is independent of the prophylactic strategy and requires a further prescription (any remaining NI prophylaxis at the point of infection cannot be used as treatment at a higher dose). The impact of this assumption is explored in the sensitivity analysis by excluding the possibility of NI treatment for patients who develop symptomatic ILI.

Probability of receiving antibiotics

The model assumes that antibiotics may be prescribed for both patients who present with uncomplicated ILI and those who present with complicated ILI. The probability that an individual with or without complications is prescribed antibiotics was derived from the study reported by Meier et al. 12 The probability that a patient with uncomplicated influenza or ILI receives antibiotics was estimated to be 0.28, 0.42 and 0.55 in the paediatric, adult and elderly populations respectively. The probability that a patient with complicated influenza or ILI receives antibiotics was estimated to be 0.71, 0.80 and 0.74 in the paediatric, adult and elderly populations respectively. Owing to a lack of evidence to the contrary, these values are assumed to be the same for both the otherwise healthy and the at-risk populations. Uncertainty surrounding these probabilities was modelled using beta distributions based on the empirical data reported by Meier et al. ,12 as shown in Table 35.

Probability of hospitalisation due to ILI-related complications

The model assumes that patients who experience ILI-related complications may require hospitalisation. As noted above, the clinical trials of influenza prophylaxis do not consistently report the incidence of complications and as such do not provide any information regarding the probability that an individual requires hospitalisation. Furthermore, data relating to hospitalisation rates were not available from the study by Meier et al. 12 Instead, the probability of hospitalisation was derived from hospitalisation rates for lower RTIs reported within a meta-analysis of 10 trials of oseltamivir for the treatment of symptomatic influenza reported by Kaiser et al. 99 The probability of hospitalisation for individuals with influenza-related complications was estimated from the placebo arm data across the 10 included studies; this probability was estimated to be 0.11 (5/46) in the otherwise healthy children and working-age adult subgroups and 0.16 (15/95) in the at-risk subgroups (including otherwise healthy elderly). 99 The data presented in the study publication did not allow for these estimates to be subdivided further according to age group; this is unfortunate as age is likely to affect the risk of hospitalisation. Uncertainty surrounding the probability of hospitalisation was modelled using beta distributions based on the empirical data reported by Kaiser et al. 99

A proportion of patients who are hospitalised may require ITU care with or without mechanical ventilation. The previous model reported by Turner et al. 10 assumed that 4.9% (22/453) of patients undergo mechanical ventilation. No alternative evidence could be identified, hence the model uses these same parameter values. A beta distribution was used to characterise the uncertainty surrounding this parameter.

Probability of death due to ILI-related complications

The probability of death due to ILI-related complications was taken from the population-based study reported by Meier et al. 12 The probability of death due to influenza complications was observed to be very low in the paediatric and adult populations (< 1%); this probability was observed to be considerably higher in the elderly patients represented within the database (10–11%). The risk of death due to complications of ILI was observed to be slightly elevated in the predisposed populations compared with the otherwise healthy patients. As complications may be a result of true influenza or other ILI, the model assumes that the probability of death is the same for those patients who develop complications due to influenza and for those patients who develop complications due to other ILI. Uncertainty surrounding this parameter was modelled using beta distributions based on the empirical data reported by Meier et al. ,12 as shown in Table 36.

Costs of prophylaxis and treatment using amantadine, oseltamivir and zanamivir

Prophylaxis and treatment were costed according to BNF list prices at the time of the assessment. The number of doses of prophylaxis required using amantadine, oseltamivir and zanamivir was calculated based on the dosages and durations in line with licensed indications (see Table 27). The model assumes that seasonal prophylaxis using amantadine is given for a period of 6 weeks (42 days) for patients who have not been previously vaccinated, and 3 weeks (21 days) for patients who have been previously vaccinated. The model assumes that seasonal prophylaxis using oseltamivir is given for a period of 6 weeks (42 days). Seasonal prophylaxis using zanamivir is assumed to be given for a period of 4 weeks (28 days). The model assumes that post-exposure prophylaxis using amantadine, oseltamivir and zanamivir is given for a period of 10 days. The duration of treatment of symptomatic ILI using oseltamivir and zanamivir is assumed to be 5 days. In line with licensed indications, the daily dosage of amantadine prophylaxis and zanamivir prophylaxis is assumed to be 100 mg and 10 mg respectively for all populations. The cost of prophylaxis and treatment using oseltamivir for children assumes a mean body mass of between 23 kg and 40 kg. Unit costs for amantadine, oseltamivir and zanamivir were taken from the BNF No. 54. 14 Amantadine is available in both capsule and syrup form, and oseltamivir is available as capsules and as a suspension for reconstitution with water. The model assumes that prophylaxis for adults is administered using capsules rather than syrup or suspension, as this allows for more reliable dosing (Dr Andrew Ross, RCGP, personal communication). The cost of each prophylaxis course and treatment course includes the possibility of wastage. Where multiple products were available, the least expensive is assumed. The costs of prophylaxis used in the model are presented in Table 37.

In the base-case analysis, the model assumes that each prescription of prophylaxis requires a GP consultation. The model assumes also that the administration of vaccination and the prescription of antiviral prophylaxis require separate consultations. The impact of prescribing multiple courses of prophylaxis for contact cases is explored in the sensitivity analysis (see One-way/multiway sensitivity analysis and scenario analysis, p. 90).

Cost of vaccination

The cost of influenza vaccination was estimated from list prices derived from BNF 54. 14 Current unit costs for influenza vaccine products range from £4.98 to £6.59, including both proprietary and non-proprietary products (Table 38). Recommended influenza vaccines vary between influenza seasons; the mean vaccine price was assumed within the model (£5.63). The model assumes that influenza vaccination is administered by a GP; the cost of vaccination is assumed to include the cost of a GP consultation based on costs reported by Curtis and Netten. 102 A GP visit is assumed to cost £25. As these costs are common to all patients receiving vaccination, these parameters have no impact on the incremental cost-effectiveness of influenza prophylaxis.

Cost of ILI presentation

The model assumes that patients present with symptomatic ILI either to their GP (in the surgery or at home) or at an A&E department. The probability that a patient with influenza or other ILI requires a home visit was derived from the study reported by Ross et al. 123 Counts of patients with ILI who had home visits were reported in aggregate form for patients aged under 75 and those aged over 75. Further data regarding the proportion of consultations which took place at home within each age group were provided by the lead author of this study (Dr Andrew Ross, RCGP, personal communication). The proportion of home visits was low in the paediatric and adult populations (5% and 8% respectively); the proportion was considerably higher in the elderly population (38%). Beta distributions were used to characterise the uncertainty surrounding this parameter based on the empirical data provided by Dr Ross of the RCGP. The proportion of all ILI presentations that take place in A&E departments was based on clinical opinion (Professor Robert Read, University of Sheffield, personal communication); the model assumes that 3% of patients present to A&E (range 1–5%). A beta distribution was used to capture the uncertainty surrounding this quantity. This parameter was assumed to be the same for otherwise healthy and at-risk paediatric, adult and elderly populations.

Unit costs for GP surgery consultations and home visits were derived from the PSSRU102 while the cost of an A&E consultation was derived from the NHS reference costs. 125 The model assumes that a GP surgery consultation costs £25,102 a home visit costs £69102 and an A&E attendance costs £95.56 (first attendance data code 180F). 125 The unit costs for A&E attendances are assumed to include the costs of diagnostic tests (e.g. blood and sputum tests, lung function tests, etc.). Based on the information reported above, the model assumes a mean cost of presentation with symptomatic ILI of £29.52 for children, £30.73 for working-age adults and £43.20 for elderly individuals.

Cost of antibiotics for the treatment of ILI-related complications

The model assumes that antibiotics are prescribed for individuals presenting with uncomplicated ILI as well as those presenting with influenza and ILI-related complications. The precise antibiotic prescribed depends on the type of complication; for simplicity, the model assumes that the preferred antibiotic for the treatment of symptomatic ILI and related complications is co-amoxiclav. In its non-proprietary tablet form, the unit cost for co-amoxiclav is £6.80 for a 21-tablet course. 14

Cost of managing adverse events resulting from vaccination and prophylaxis

The model assumes that adverse events resulting from vaccination and prophylaxis (amantadine only) incur additional costs due to additional GP attendances. As noted above, the cost of a GP attendance was assumed to be £25. 102 It should be noted that not all patients who experience adverse events will consult their GP, hence it is possible that the costs of managing adverse events is overestimated in the model, although the impact of this bias on cost-effectiveness outcomes is minor. The model assumes that adverse events due to oseltamivir and zanamivir are mild, self-limiting and incur no additional medical costs.

Cost of hospitalisation due to serious complications of influenza and other ILI

The cost of hospitalisation for serious complications was taken from the NHS reference costs 2005–2006. 125 The unit cost for lobar, atypical or viral pneumonia (D14) without complications was assumed; this was divided by the mean length of stay to derive an estimate of the daily cost of hospitalisation. The standard error for this parameter was estimated by dividing the interquartile range by 1.349 and dividing this by the square root of the number of submissions. This cost was then multiplied by the assumed duration of inpatient stay within each population group reported by Turner et al. 10

Mean lengths of hospitalisation stay due to ILI-related complications were taken from Turner et al. ;10 these are assumed to differ substantially between the paediatric, adult and elderly population subgroups. Turner et al. 10 did not include any uncertainty surrounding these estimates, hence the degree of uncertainty surrounding these mean values has been subjectively modelled using gamma distributions. These data are shown in Table 39.

A proportion of patients with particularly severe complications may require ITU care and mechanical ventilation; Turner et al. 10 note that the proportion of cases requiring mechanical ventilation is not known. The model uses the same value reported by Turner et al. 10 (probability of ITU care = 0.05). The typical duration of intensive care required for severely complicated cases was derived from a descriptive study of pneumonia management in the US reported by Oliveira et al. 126 Oliveira et al. report a mean duration of intensive care unit (ICU) stay of 28 days ± 26 days (10 patients). It should be noted, however, that this study may not reflect UK treatment patterns. Uncertainty surrounding this parameter was modelled using a lognormal distribution. The cost per intensive care day was taken from the NHS reference costs 2005–2006. 125 The cost per critical care day was assumed to be £1345.39 (Critical care level 2 code TCCS CC1L2).

QALYs lost due to influenza and ILI episodes

Previous evaluations of influenza and its prevention and treatment have used health utility scores derived using the EQ-5D127 or the Health Utilities Index, mark III (HUI3)128 based on general population valuations or retrospective valuations from individuals with a history of virologically-confirmed influenza. These studies were based on small numbers of subjects (n < 25). The study reported by Griffin127 reported an extreme value for the utility associated with influenza infection which is valued as a state worse than death (utility = –0.066). 127 It is likely that the impact of influenza on quality of life will be greatest when the illness is at its peak, and that it will have a lesser impact in the first and last days of illness.

The methodology reported by Turner et al. 10 was used to generate QALY loss estimates for cases of influenza and other ILI (see Review of exisiting economic evaluation studies, p. 43). The expected QALY loss due to an episode of influenza was estimated using data collected in five clinical trials of oseltamivir for the treatment of influenza in healthy adults and at-risk and elderly populations. Within these studies, a 10-point Likert scale was completed daily for up to 21 days by patients receiving oseltamivir treatment and patients receiving placebo. The scale employed was similar to a VAS, using a lower anchor which had a score of 0 describing ‘worst possible health’ and an upper anchor which had a score of 10 describing ‘normal health for someone your age’. As the upper anchor on the rating scale did not describe a notional state of ‘best possible health’, Turner et al. 10 recalibrated the upper anchor to represent mean utility scores for each age group using data from the Measurement and Valuation of Health (MVH) study. 105 The VAS equivalent data were then converted into TTO utility scores based on a VAS–TTO transformation algorithm reported by the MVH group. 105 Turner et al. 10 assumed that missing values resulted from the respondent having returned to normal health; missing values were therefore imputed as ‘normal health’ utility scores. The number of QALYs gained over the 21-day period was estimated for the healthy adult and at-risk and elderly populations for oseltamivir and placebo. The number of QALYs lost due to an influenza episode was calculated as the expected QALYs gained in the non-influenza population over 21 days minus the QALYs lost due to influenza over 21 days. For example, assuming a baseline utility of 0.90 without influenza, and a mean 21-day QALY loss of 0.041 with influenza, the number of QALYs lost due to influenza is calculated as (0.90 × 21) – (0.041 × 365)/365.

As equivalent data were not available from the zanamivir trials, the model assumes that the impact of zanamivir treatment on HRQoL is equivalent to that for oseltamivir. Data were not available for the paediatric population; therefore, the model assumes the same QALY loss as in the healthy adult population. The model also assumes that the QALY loss for an uncomplicated influenza episode is the same as that for an uncomplicated ILI episode. Mean QALY gains over 21 days used in the model are presented in Table 40. In their earlier report, Turner et al. 10 modelled the uncertainty in the data, but did not account for additional uncertainty resulting from the process of mapping from Likert data collected in the trials to a VAS and subsequently to TTO utilities. In order to better reflect this uncertainty, the model uses the mean QALY scores and an assumed level of additional uncertainty (subjectively assigned). These parameters were modelled using beta distributions.

QALYs lost due to adverse events due to prophylaxis

The model assumes that adverse events due to amantadine impact upon a patient’s health-related quality of life. The model assumes a utility decrement of 0.20 for a mean duration of 5 days based on the previous work reported by Turner et al. 10 Uncertainty surrounding the disutility of adverse events was modelled using a beta distribution, whilst uncertainty surrounding the duration of adverse events was modelled using a gamma distribution, assuming a standard error of 1 day.

QALYs lost due to ILI-related complications

In principle, the Likert scale data collected within the oseltamivir trials should have included quality of life valuations for individuals who experienced serious complications of influenza (or at least those occurring within the 21-day evaluation period). However, it should be noted that beyond the first 7 days, the number of respondents in the treatment and placebo groups declined considerably. The model assumes that serious complications such as respiratory illness and the exacerbation of underlying health problems are not captured within these valuations, and that such complications result in a further reduction in a patient’s HRQoL.

Systematic searches were undertaken to identify studies reporting preference-based valuations of the impact of influenza, ILI and related complications on HRQoL (see Appendix 1). The searches did not identify any published studies that reported preference-based valuations of the impact of the range of ILI complications associated with influenza and ILI (bronchitis, pneumonia, otitis media and exacerbation of an underlying condition, e.g. asthma). Instead, health utility decrements for secondary complications were derived from a modelling study of vaccination against a variety of diseases. 129 Within this study, committee HUI (mark II) scores were derived for a number of health states associated with influenza and ILI (Table 41). These utility estimates represent the consensus of the committee who undertook the valuation exercise and as such do not include any estimates of uncertainty. Wide standard errors were assumed within the probabilistic sensitivity analysis based on lognormal distributions.

The duration over which these utility decrements are applied was based on clinical trial data presented within the Roche submission,20 sourced from clinical trials of oseltamivir. The duration of each illness was derived simply by calculating the number of days between the onset of the complication and its resolution (Gavin Lewis, Roche, personal communication). The submission contained data relating to the duration of pneumonia, bronchitis and otitis media in children, healthy adults and at-risk groups. The mean duration of disutility for any respiratory complication was estimated by weighting the durations observed in the clinical trials by the ratio of pneumonia : bronchitis in each age group, as reported by Meier et al. 12 In the absence of any alternative evidence, the duration of other respiratory complications was assumed to follow this same pattern. The uncertainty analysis assumes a large standard error of 3 days for each subgroup; uncertainty surrounding these quantities was modelled using gamma distributions. Owing to a lack of alternative evidence, the duration of other non-respiratory complications is assumed to be the same as that for respiratory complications. Table 42 shows the assumed durations for these reductions in HRQoL.

QALYs lost due to premature death resulting from ILI complications

The expected number of QALYs lost due to premature death resulting from secondary complications of ILI was also based on the methods reported by Turner et al. 10 Crude estimates of the mean age of death due to influenza for the paediatric, adult and elderly populations were derived from data reported by the Office for National Statistics (ONS; DH2). 130 Interim life tables for England and Wales were then used to calculate the expected number of life-years lost due to premature death for each age group based on the mean age of death. Life-years lost were weighted by general population utility scores derived from Kind et al. 131 to generate estimates of the number of QALYs lost within each age group. Expected QALYs lost were discounted at a rate of 3.5%. It should be noted that while the risk of death due to ILI complications is higher in the at-risk groups, the estimate of the number of QALYs lost is assumed to be the same for the healthy and at-risk populations; this assumption may be biased in favour of prophylaxis within the at-risk population subgroups. Table 43 shows the modelled estimates of the expected discounted QALYs for each population group. 96

Sensitivity analysis 1: proposed price reduction for zanamivir

In November 2007, the manufacturer of zanamivir (GSK) applied to the Department of Health for a price modulation of two of their drugs, one of which was zanamivir. The current list price for zanamivir is £24.55 (five disks, four blisters per disk); the new proposed price for zanamivir is £16.36 (Toni Maslen, Health Outcomes Programme Leader, GSK, personal communication). This proposed price reduction for zanamivir was approved by the Department of Health with effect from 1 February 2008 but was not listed in the BNF (No. 54)14 at the time of submission. This scenario analysis presents the central estimates of cost-effectiveness of influenza prophylaxis including this proposed price reduction for zanamivir. All other parameter values and assumptions in this analysis are the same as those in the base-case analysis presented in Central estimates of cost-effectiveness (see below). The reader should note that where zanamivir remains dominated by another prophylaxis option despite the price change, the slight differences in the cost-effectiveness of the remaining prophylactic options from the base case results are due to sampling errors in the stochastic model.

Sensitivity analysis 2: deterministic estimates of cost-effectiveness

The base-case health economic analysis is based upon the expected (mean) costs and health outcomes for each prophylactic option, drawn from the stochastic model. The second scenario presents the cost-effectiveness results based on the deterministic model.

Sensitivity analysis 3: cost-effectiveness of oseltamivir given in suspension form

The base-case analysis assumes that seasonal prophylaxis using oseltamivir is prescribed in capsule form to all adult populations, as this is likely to ensure more accurate dosing. However, in principle, oseltamivir given as suspension may allow for less wastage than in capsule form, thus leading to a reduction in the cost of the drug. A 56-cap pack of oseltamivir provides 10×75 mg tablets providing 750 mg of the drug (10 doses) while a 75 ml bottle (60 mg/5 ml) of oseltamivir in suspension form provides a total of 900 mg of the drug (12 doses of 75 mg). While both products cost £16.36 per unit, the use of suspension could, in principle, offer savings over oseltamivir capsules.

Sensitivity analysis 4: multiple prescriptions

The base-case model assumes that each prescription of prophylaxis requires a GP consultation; for vaccinated patients, the model assumes that prophylaxis can be given during the same consultation as the influenza vaccine. The Roche model assumed that four prescriptions of prophylaxis could be obtained per GP attendance. This scenario analysis assumes that four prescriptions may be obtained per individual, resulting in a reduction in the cost of GP attendances for unvaccinated patients. 20

Sensitivity analyses 5 and 6: reduced vaccine efficacy

Sensitivity analysis was undertaken assuming a lower efficacy rate for vaccination to capture the potential impact of a mismatch between vaccine and circulating strains of influenza. Scenario 5 assumes an RR for vaccination of 0.50, while scenario 6 assumes an RR of 0.75.

Sensitivity analysis 7: protection over entire influenza season

The base-case analysis assumes that patients receiving seasonal prophylaxis are at risk of infection when they stop taking the drug. This scenario assumes that the patient is protected over the entire influenza season.

Sensitivity analysis 8: no antiviral treatment for symptomatic influenza

This sensitivity analysis assumes that patients who develop symptomatic ILI do not receive antiviral treatment using oseltamivir or zanamivir.

Sensitivity analysis 9: equivalent efficacy for oseltamivir and zanamivir prophylaxis

There is uncertainty surrounding the relative efficacy of oseltamivir and zanamivir for the prophylaxis of influenza. The Roche model assumed that oseltamivir and zanamivir had equivalent efficacy. This scenario assumes that oseltamivir and zanamivir are equivalent, and uses the most favourable efficacy estimate for NIs within the model subgroup under evaluation.

Sensitivity analysis 10: no adverse events

There is uncertainty regarding the cost and health impact of adverse events associated with influenza prophylaxis. The base-case model assumes that individuals receiving prophylaxis may experience adverse events that may lead to additional medical care costs and a further loss of quality of life for amantadine. This scenario explores the impact of assuming no costs or health impacts associated with adverse events.

Sensitivity analysis 11: no withdrawals from prophylaxis

The model assumes that a proportion of patients withdraw from prophylaxis, and that patients who withdraw gain no protective benefit against influenza. This scenario assumes a withdrawal probability of 0.

Sensitivity analyses 12–16: resistance against oseltamivir

The base-case model assumes that resistance to oseltamivir is 0. These scenarios explore the impact of oseltamivir resistance on resulting cost-effectiveness estimates. Levels of resistance against amantadine are assumed to be the same as the base-case value for each scenario.

Sensitivity analysis 17: lower attack rates

Previous models of influenza prophylaxis have reported that cost-effectiveness estimates are highly sensitive to the true influenza attack rate. This scenario assumes that the attack rate is half that of the base case in each model subgroup.

Sensitivity analysis 18: higher attack rates

This scenario assumes that the attack rate is double that of the base case in each model subgroup.

Sensitivity analysis 19: use of a higher threshold for influenza activity

The base-case analysis assumes that seasonal prophylaxis will be used when the GP consultation rate for ILI is in excess of 30 per 100,000 population. 8 This scenario analysis examines the potential impact of using the previous influenza threshold of 50 consultations per 100,000 population on the cost-effectiveness of prophylaxis. This analysis draws on parameter values reported by Turner et al. 10 which was undertaken when the previous influenza threshold was implemented.

Sensitivity analysis 20: lower GP consultation rate

The base-case model assumes that the probability that an individual with symptomatic ILI consults a health-care professional is 0.25; however, this is based on a single survey and is associated with considerable uncertainty. This sensitivity analysis assumes that the probability that an individual with symptomatic ILI consults their GP is half the base-case value.

Sensitivity analysis 21: higher GP consultation rate

This sensitivity analysis assumes that the probability that an individual with symptomatic ILI consults their GP is double the base-case value.

Sensitivity analysis 22: alternative mapping function for influenza QALY loss

The base-case model uses rating scale data from clinical trials, mapped to a VAS, and subsequently mapped to TTO to generate QALY losses for the period in which an individual has influenza. This sensitivity analysis uses an alternative mapping function, converting VAS data into EQ-5D utilities.

Sensitivity analysis 23: lower QALY losses for at-risk groups

The base-case model assumes that the likely reduction in expected QALYs lost due to premature death as a result of influenza complications is the same in healthy and at-risk populations. This analysis assumes that the expected QALY loss in the at-risk group is half the value used in the base case.

Sensitivity analysis 24: complication utility decrements halved

The evidence concerning the impact of ILI complications on health outcomes is scarce and subject to considerable uncertainty. This analysis assumes a 50% reduction in utility decrements associated with ILI complications.

Sensitivity analysis 25: impact of assumptions regarding hospitalisation in uncomplicated cases

The base-case model assumes that uncomplicated ILI cases do not result in hospitalisation or death. Scenario 25 assumes that 10% of uncomplicated cases result in hospitalisation.

Sensitivity analysis 26: undiscounted cost-effectiveness estimates

Within the base-case model analysis, health outcomes were discounted at a rate of 3.5%. This analysis presents cost-effectiveness estimates without discounting.

Group 1: healthy children

The model results presented in Table 44 suggest that the most effective seasonal prophylaxis option for healthy children is oseltamivir, irrespective of vaccination status. Oseltamivir is expected to produce a small improvement in terms of QALY losses avoided compared with the other prophylactic strategies; however, this is not the most expensive prophylactic option. Zanamivir is less effective and more expensive than oseltamivir, irrespective of vaccination status, hence it is ruled out by simple dominance and is not included in this analysis. For healthy children who have not been previously vaccinated against influenza, amantadine is expected to be ruled out by extended dominance, as oseltamivir has a more favourable incremental cost-effectiveness ratio. For healthy children who have been previously vaccinated, amantadine is expected to be dominated by no prophylaxis. For unvaccinated children, the incremental cost-effectiveness of oseltamivir versus no prophylaxis is expected to be around £44,000 per QALY gained. For healthy children who have received prior vaccination, the incremental cost-effectiveness of oseltamivir compared with no prophylaxis is estimated to be approximately £129,000 per QALY gained.

Group 2: at-risk children

The model results presented in Table 45 suggest that the most effective seasonal prophylaxis option for at-risk children is oseltamivir irrespective of whether or not they have been previously vaccinated. Again, zanamivir is expected to be less effective and more expensive than oseltamivir, hence it is ruled out of the analysis by simple dominance. Amantadine is expected to be ruled out of the analysis by extended dominance (again oseltamivir has a more favourable cost-effectiveness ratio). The incremental cost-effectiveness of oseltamivir compared with no prophylaxis is estimated to be approximately £17,000 per QALY gained in unvaccinated at-risk children and £51,000 per QALY gained in at-risk children who have previously been vaccinated against influenza.

Group 3: healthy adults

The results presented in Table 46 suggest that oseltamivir is expected to be the most effective option for seasonal prophylaxis of influenza in healthy adults. This analysis suggests that zanamivir is expected to be slightly less expensive than oseltamivir, but is ruled out by extended dominance. For unvaccinated healthy adults amantadine is ruled out of the analysis by extended dominance, while for vaccinated healthy adults amantadine is expected to be dominated by no prophylaxis. The incremental cost-effectiveness of oseltamivir compared with no prophylaxis is estimated to be approximately £148,000 per QALY gained in unvaccinated healthy adults and £427,000 per QALY gained in healthy adults who have previously been vaccinated.

Group 4: at-risk adults

Table 47 suggests that oseltamivir is expected to be the most effective option for seasonal prophylaxis in at-risk adults. As with the healthy adult model, zanamivir is expected to be ruled out by extended dominance as oseltamivir has a lower incremental cost-effectiveness ratio. For unvaccinated at-risk adults, amantadine is expected to be ruled out by extended dominance, while for vaccinated individuals, amantadine is expected to be less effective and more expensive than a policy of no prophylaxis. The incremental cost-effectiveness of oseltamivir compared with no prophylaxis is estimated to be approximately £64,000 per QALY gained in unvaccinated individuals and £187,000 per QALY gained in at-risk adults who have previously been vaccinated against influenza.

Group 5: healthy elderly

The cost-effectiveness results presented in Table 48 suggest that oseltamivir is expected to be the most effective seasonal prophylaxis option for elderly adults who are otherwise healthy. As with the working-age adult models, zanamivir is expected to be ruled out by extended dominance. Amantadine is expected to be ruled out by extended dominance for unvaccinated individuals, and is dominated by no prophylaxis in vaccinated populations. The incremental cost-effectiveness of oseltamivir compared with no prophylaxis is estimated to be around £50,000 per QALY gained in unvaccinated healthy elderly adults and around £122,000 per QALY gained in healthy elderly adults who have previously been vaccinated.

Group 6: at-risk elderly

The results presented in Table 49 suggest that oseltamivir is expected to be the most effective seasonal prophylaxis option for at-risk elderly adults. Zanamivir and amantadine are both ruled out of the analysis by extended dominance. The incremental cost-effectiveness of oseltamivir compared with amantadine is estimated to be around £38,000 per QALY gained in unvaccinated at-risk elderly individuals and £94,000 per QALY gained in at-risk elderly adults who have previously been vaccinated.

Group 1: healthy children

The model results presented in Table 50 suggest that zanamivir is expected to be the most effective option for the post-exposure prophylaxis of influenza in otherwise healthy children. In this instance, oseltamivir and amantadine are ruled out of the analysis by extended dominance. The incremental cost-effectiveness of zanamivir versus no prophylaxis is estimated to be £23,000 per QALY gained for unvaccinated healthy children and around £72,000 in vaccinated healthy children.

The reader should note that oseltamivir is the only licensed prophylactic in children under the age of 5 years; the incremental cost-effectiveness ratio for oseltamivir versus no prophylaxis is expected to be around £24,000 per QALY gained and £74,000 per QALY gained in unvaccinated and vaccinated groups respectively.

Group 2: at-risk children

The cost-effectiveness results presented in Table 51 suggest that zanamivir is expected to be the most effective option for the post-exposure prophylaxis of influenza in at-risk children. Oseltamivir and amantadine are expected to be ruled out of the analysis by extended dominance for unvaccinated and vaccinated subgroups. The incremental cost-effectiveness of zanamivir versus no prophylaxis is estimated to be around £8000 per QALY gained in unvaccinated at-risk children and approximately £28,000 per QALY gained in vaccinated at-risk children.

For at-risk children under the age of 5 years, the incremental cost-effectiveness of oseltamivir versus no prophylaxis is expected to be around £9000 per QALY gained for unvaccinated at-risk children and around £29,000 per QALY gained for vaccinated at-risk children.

Group 3: healthy adults

The cost-effectiveness estimates presented in Table 52 suggest that oseltamivir is expected to be the most effective option for the post-exposure prophylaxis of influenza in healthy adults. Within this subgroup, zanamivir is expected to be dominated by oseltamivir irrespective of vaccination status. Amantadine is expected to be ruled out by extended dominance. The incremental cost-effectiveness of oseltamivir versus no prophylaxis is estimated to be £34,000 per QALY gained for vaccinated healthy adults and around £104,000 per QALY gained for unvaccinated healthy adults.

Group 4: at-risk adults

Table 53 suggests that oseltamivir is expected to be the most effective option for the post-exposure prophylaxis of influenza in at-risk adults. Again, zanamivir is expected to be dominated by oseltamivir irrespective of vaccination status. Amantadine is again expected to be ruled out by extended dominance. The incremental cost-effectiveness of oseltamivir versus no prophylaxis is estimated to be around £13,000 per QALY gained for unvaccinated at-risk adults and £44,000 per QALY gained for previously vaccinated at-risk adults.

Group 5: healthy elderly

Table 54 suggests that oseltamivir is expected to be the most effective option for the post-exposure prophylaxis of influenza in otherwise healthy elderly adults. Zanamivir is again expected to be dominated by oseltamivir and is hence ruled out of the analysis. Amantadine is expected to be ruled out of the analysis by extended dominance. The incremental cost-effectiveness of oseltamivir versus no prophylaxis is estimated to be around £11,000 per QALY gained for unvaccinated healthy elderly individuals and around £28,000 per QALY gained for at-risk elderly who have previously been vaccinated against influenza.

Group 6: at-risk elderly

The model results presented in Table 55 suggest that oseltamivir is expected to be the most effective option for the post-exposure prophylaxis of influenza in at-risk elderly individuals. Zanamivir is expected to be dominated by oseltamivir and is ruled out of the analysis. Amantadine is expected to be ruled out by extended dominance. The incremental cost-effectiveness of oseltamivir versus no prophylaxis is estimated to be around £8000 per QALY for vaccinated at-risk elderly individuals and around £22,000 per QALY gained for at-risk elderly individuals who have previously been vaccinated.

Seasonal prophylaxis

Results for seasonal prophylaxis are presented in Tables 56–61.

Post-exposure prophylaxis

Results for post-exposure prophylaxis are presented in Tables 62–67.

Table 68 summarises the ICERs presented in the base-case analysis and those including the proposed reduction in the price of zanamivir.

The summary of cost-effectiveness results presented in Table 68 shows that the proposed price reduction has no impact on the majority of economic comparisons presented in the base-case analysis. In terms of seasonal prophylaxis, the cost-effectiveness of zanamivir is no longer ruled out by extended dominance in at-risk adults; however, the incremental cost-effectiveness ratio for zanamivir versus no prophylaxis remains in excess of £50,000 per QALY gained for these comparisons. In terms of the post-exposure prophylaxis of influenza, the price reduction has no impact on the adult and elderly subgroup analyses, as zanamivir consistently remains dominated by oseltamivir. The proposed price reduction is, however, expected to lead to an improvement in the cost-effectiveness of zanamivir for otherwise healthy and at-risk children. For unvaccinated healthy children, the reduction in the price of zanamivir is expected to result in a reduction in the cost-effectiveness of zanamivir versus no prophylaxis from £23,000 per QALY gained to £19,000 per QALY gained. The cost-effectiveness of zanamivir in vaccinated, otherwise healthy children is expected to be in excess of £59,000 per QALY gained. For unvaccinated at-risk children, the lower price for zanamivir is expected to lead to an improvement in the cost-effectiveness of zanamivir versus no prophylaxis from £8000 per QALY gained to £6000 per QALY gained. For vaccinated at-risk children, the cost-effectiveness of zanamivir is improved from £28,000 per QALY gained to £23,000 per QALY gained.