Population and sub-groups

The model considered a cohort of patients who had had a recent MI. The trial evidence that the model is based on included relatively few older (>65) or black patients, so the results may not be reliable for these groups.

Interventions compared

The analysis assessed the costs and effects of CR compared with no CR. Additionally it assessed using the output from the CR model, the cost effectiveness of two methods of increasing uptake and adherence compared to ‘current practice/usual care’, i.e. current uptake of CR. These two methods were firstly the use of phone calls together with home visits by a HCP, in which the HCP was assumed to make contact over the phone four times, each followed by a home visit and secondly the use of two consecutive invitation letters to a CR programme over a period of 6 weeks.

Outcomes

The treatment effects were measured in terms of reduction of CVD events: non-fatal MI, revascularisation CVD-related deaths and other deaths. Health outcomes for the cost-effectiveness analysis are summarised in the form of Quality Adjusted Life Years (QALYs), where one QALY represents one year of healthy life.

Model structure and assumptions

A Markov model was developed to evaluate the incremental costs and effects of lifetime intervention with CR in secondary prevention of cardiovascular disease (CVD) events in post MI patients from a UK NHS perspective.

In a Markov model there are a finite number of health states. It is assumed that at any point in time, all patients must be in one and only one of the states. The model then replicates how a hypothetical cohort of people moves between the states. Figure 1 shows a schematic representation of the patients’ pathways. All patients start in the event-free health state. During each six-month cycle of the model, a proportion of patients enter one of the qualifying event health states (MI, revascularisation and death) while the remainder stay in the event free state. Patients can experience more than one non-fatal event in subsequent periods of the model.

Figure 1

Economic Model Structure.

The rate at which people move through the model is regulated by transition probabilities, which describe the likelihood of moving between states over each model cycle (six months). For illustration, the equivalent annual transition probabilities for a 65-year-old patient receiving no-CR are shown in Table 1. The probabilities are derived from the placebo arms of the meta-analysis of CR trials.

Table 1

Probabilities for a 65-year-old man without Cardiac rehabilitation.

The model was run first assuming that the cohort was to receive no CR. The model was then re-run assuming that the cohort all received CR and complied 100%. Transition probabilities were adjusted to reflect the expected reduction in CVD events and revascularisations. Health care costs and QALYs were then estimated for each option by weighting the time spent in the various states by mean costs and ‘utilities’ (health-related quality of life) of the health states. The cost and utility data used in the model are described below.

The time horizon modelled is lifetime, with an assumed upper age of 100, by which time most of the cohort have died.

Cost data

The NICE reference case specifies that costs should be measured from an NHS and personal social services perspective. These should include the direct cost of drug treatment and also potential savings from avoided treatments due to reduced incidence of CVD and hospitalisations. Costs were calculated using cost weights for each of the states of the model, multiplied by the time spent in each state. Costs are at 2005 prices. As per current NICE guidance, an annual discount rate of 3.5% was used for both costs and health benefits (National Institute for Health & Clinical Excellence. 2006b).

The costs of health states used in the model are shown in Table 4. Costs for revascularization which includes hospitalisation were taken from the NHS reference cost 2005 (Department of Health Reference Costs 2005 2005). It was assumed that 67% of patients will have PCI and 33% will have CABG and the costs were weighted to reflect this (expert opinion). The cost of the well states was assumed to be the outpatient cost which includes the costs of medication and monitoring costs, these were taken from the NICE hypertension guideline 2006 (National Institute for Health and Clinical Excellence. 2006). The cost of CR was taken from a review (Beswick, A. D., Rees, K., Griebsch, I. et al 2004) and included staff costs, equipment, and that of recruiting patients to CR. Costs of acute MI (non-fatal reinfaction) were assumed to be the same as those of patients treated with thrombolysis, which includes the cost of hospitalisation (Hartwell, D. et al 2005). The cost of death was zero.

Table 4

Costs of health states.

The costs of each strategy used to increase uptake, invitation letters or phone call contacts followed by home visits, were calculated from resource use identified in the HTA (Beswick, A. D., Rees, K., Griebsch, I. et al 2004). The actual unit costs were taken from the Personal Social Services Research Unit PSSRU (PSSRU 2005).

The cost of invitation letters were calculated assuming that letters inviting participants to a CR programme were sent twice, soon after discharge and 3 weeks later. It was assumed that the letters were sent by a medical secretary, and also that 30 minutes work was required to type and send each letter.

For the HCP and phone calls (Beswick, A. D., Rees, K., Griebsch, I. et al 2004) estimated there would be about four visits and a phone call made before each visit. Contact by the HCP was assumed to last 30 minutes and the phone call about 11 minutes. Duration for the phone call and staff costs were taken from the PSSRU (PSSRU 2005)

Quality of life (Utility)

In the NICE reference case, the value of health outcomes – including beneficial and harmful impacts of treatment on mortality and morbidity – is estimated using the Quality Adjusted Life Year (QALY) approach. This requires estimates of survival and quality of life associated with each health state included in the model.(National Institute for Health & Clinical Excellence. 2006b)

The utility values used in the model are shown in Tables 6 and 7. The values were taken from literature or the Harvard cost effectiveness registry database (Harvard CEA Registry 1997)

Table 6

Health state utility weights.

Table 7

Utility weight by age.

Utilities were adjusted to reflect the fact that health related quality of life in the general population decreases with age (i.e. multiply the disease utility weight by age utility weight). Age utility weights were taken from the Department of Health, Health Survey for England (Department of Health 1998)

One study (Oldridge, N., Furlong, W., Feeny, D. et al 1993), found that there was a difference of 0.052 QALYs between patients who participated in CR and those who did not using the time trade off method. This factor was applied to all the well states in the CR arm to take account of this difference in quality of life in sensitivity analysis. The weight attached to death was zero

Cost effectiveness

The results of cost-effectiveness analysis are usually presented as Incremental Cost-Effectiveness Ratios (ICERs), which determine the additional cost of CR per additional QALY gained compared with no CR

ICERs = (cost of CR - cost of no CR)/(QALY of CR - QALY of no CR)

Where more than two interventions are being compared, the ICERs are calculated using the following process:

1. The drugs are ranked in terms of cost (from the cheapest to the most expensive).
2. If a drug is more expensive and less effective than the previous one, then it is said to be ‘dominated’ and is excluded from further analysis.
3. ICERs are calculated for each drug compared with the next most expensive non-dominated option. If the ICER for a drug is higher than that of the next more effective strategy, then it is ruled out by ‘extended dominance’. This means that there is some mixture of two other strategies that is more effective and less expensive.
4. ICERs are recalculated excluding any drugs subject to extended dominance. (Palmer, S., Sculpher, M., and Philips, Z. 2004)

Sensitivity analysis

The model includes a base case analysis supplemented with both deterministic and probabilistic sensitivity analysis. In the probabilistic sensitivity analysis all parameters in the model were allowed to vary simultaneously according to an assumed distribution reflecting the degree of uncertainty over the parameter value.

a. Cost effectiveness of Cardiac Rehabiliation vs. no Cardiac Rehabiliation

The base case results are presented in table 8a and b for 65-year-old men and women respectively. This suggests that CR is cost-effective for this population. The ICER of CR compared with no CR is about £7,860 and £8,360 per QALY gained for men and women respectively, which is below the level usually considered to be affordable in the NHS (about £20,000 to £30,000 per QALY).

Figure 2Cost effectiveness plane, CR compared to no CR in post MI patients

b. The comparative cost effectiveness of the methods used to increase uptake of CR after an MI

None of the strategies were ruled out on the basis of dominance. The base case model shows that the strategy of sending letters compared to usual care to increase uptake of CR is about £ 8,000/QALY. The strategy of using phone calls and home visits by a HCP compared to sending letters is about £ 8,400/QALY gained which is below the level usually considered to be affordable in the NHS (about £20,000 to £30,000 per QALY).

While the results of the ICER can be used to determine the optimal decision based on a comparison of mean costs and QALYs, they do not incorporate the uncertainty surrounding this decision. Figure 3 presents the base-case results in the form of cost effectiveness acceptability curves (CEACs) for CR versus no CR. Figure 4 shows the comparison between the three strategies of increasing uptake of CR. These curves detail the probability that each strategy is cost effective over a range of potential maximum values that the NHS is prepared to pay for an additional QALY.

Figure 3

Cost effectiveness acceptability curve, Cardiac rehabilitation versus no cardiac rehabilitation.

Figure 4

Cost effectiveness acceptability curve, Usual care vs letters vs telephone + HCP.

The CEACs demonstrate that CR is highly cost effective. The probability that CR is cost effective increases as the willingness to pay increases. If the NHS is willing to pay upto £10,000 for an additional QALY, the probability that CR is cost effective is around 60%, increasing to 71% if the maximum willingness to pay is £20,000.

In figure 4, the CEACs demonstrate that either the strategy of phone calls plus home visits by a HCP or the strategies of sending letters are cost effective. However by comparison, the strategy of using phone calls plus home visits by a HCP is the optimal strategy. If the NHS is willing to pay upto £10,000 for an additional QALY, the probability that phone calls plus home visits by a HCP is cost effective is around 57%, increasing to 69% if the maximum willingness to pay is £20,000.

Sensitivity analysis for relative risk of non-CVD death

Interpretation

The model assumed that patients after MI have a two fold increased risk of dying from any other causes than the general population. We tested this assumption in sensitivity analysis. When we assumed that there was no difference in non CVD mortality between the general population and the post MI patients the ICERs increased slightly to £8,980/QALY. Overall the model was robust to this assumption

Sensitivity analysis for Age and sex

Interpretation

Age and sex did not affect the results. However it should be acknowledged that the efficacy data available is mainly for middle aged men usually aged upto 65. Only mortality data was available a by age and sex in our model.

Sensitivity analysis for efficacy of CR

Interpretation

The efficacy of CR was tested using the upper and lower confidence intervals. When the lower confidence interval is used, the ICERs are expected to improve and when the upper confidence interval the ICERs is expected to worsen. The model remained robust when both upper and lower confidence intervals were used. Impact of CR on mortality appears to have a bigger impact on the ICERs. When the lower CI is used the ICERs fall to about £5,410/QALY and when the upper CI is used the ICERs rise to about £19,730/QALY.

Sensitivity analysis for reduction in quality of life due to CR

Interpretation

The impact of cardiac rehabilitation on quality of life was tested. It was assumed that that cardiac rehabilitation will result in disutility or a loss in quality of life. Arbitrary figures were used ranging between 1–4%. CR will cease to be cost effective at £20,000/QALY threshold if it resulted in quality of loss of more than 3.5%. This is an unlikely scenario unless CR is provided to very high risk patients whose health is made worse by participating in CR.

Sensitivity analysis for additional QALYs due to CR

Interpretation

The model is not sensitive to changes in additional QALYs as a result of CR. The base case model assumed that there was no difference in QALYs between those who participate in CR and those that do not. We assumed there would be an increase in QALY due to CR ranging from 1% to 10%. The estimated ICERs ranged from £7,060 per QALY gained.for a 1% increase in QALY to about £3,680 for a 10% increase in QALY due to CR

Sensitivity analysis for compliance to CR

Interpretation

The base case model assumed that patients will adhere to CR 100% However studies have shown that compliance rates are high in the first year and fall in subsequent years. The average for the first year is between 60 to 70% in the first 12 months, falling to between 45% to 70% after 3 years. We tested for adherence in our model. The model appears to be slightly sensitive to this assumption since compliance rates below 40% are not cost effective. For instance 40% compliance has an estimated ICER of about £20,000

Sensitivity analysis for Cost of CR

Interpretation

The results were sensitive to changes in the cost of rehabilitation but remained robust. The ICER ranged from about £2,320/QALY if the lowest cost per patient per year of £140 cited by Taylor et al is used to about £12,890/QALY when the cost is assumed to be about £800/patient per year. As the cost of rehabilitation increases the ICER become less favourable

Sensitivity analysis for Cost of CVD events and procedures

Interpretation

The model is not sensitive to outcome costs. The mean costs were reduced by 50% and increased by 100% and the results remained robust, all below £8,000/QALY.

Sensitivity analysis for discounting

Interpretation

The impact of the discounting was also explored. Assuming that there was no discounting, the results of the model remain robust with an estimated ICER of about £6,780/QALY. If the discount rate was raised to 6%, the ICERs slightly increased to £8,680/QALY. Thus the model was not sensitive to this parameter

Sensitivity analysis for baseline uptake of CR

Interpretation

The model was not sensitive to assumptions about baseline uptake of cardiac rehabilitation. Studies have shown that participation rates ranges between 14–50%. In The base case model assumed a 40% participation rate. We varied the participation rate between 14% to 85%. The ICERs were below £10,000/QALY for all comparisons thus the model remained robust to this assumption

Sensitivity analysis for efficacy of letters

Interpretation

When assumptions about the efficacy of letters were changed the model remained robust. When letters were assumed to result in a modest 1% increase in uptake, the ICERs compared to usual care increased to about £15,000/QALY. The ICERs of phone call plus home visit by a HCP improved as the efficacy of letters worsens, and worsens as the efficacy of letters improves. For instance when the efficacy of letters was assumed to result in a 100% increase in the uptake of CR letters the ICERs for letters compared to HCP + phone increased to about £11,000/QALY

Sensitivity analysis for efficacy of phone plus HCP

Interpretation

In the base case model phone calls plus home visits by a HCP resulted in 111% increase in uptake of CR. We did a threshold analysis to find the point at which this intervention ceases to be cost effective. The model estimates that when the efficacy of phone call plus home visits by a HCP was assumed to result in an increase in uptake of CR of less than 55% then phone calls plus home visits by a HCP will not be cost effective at a willingness to pay value of £20,000/QALY. For instance if the strategy of HCP + phone resulted in increase of 50% in uptake of CR, letters will dominate them. Thus the model is sensitive to this assumption, but the analysis is speculative since the ranges used in sensitivity analysis are arbitrary.

Population and sub-groups

The model considered a cohort of low risk post MI patients with preserved left ventricular dysfunction. Low risk is defined as the population who met the inclusion criteria of the meta-analysis (Al-Mallah, M. H., Tleyjeh, I M., Abdel-Latif, A. A. et al 2006) seen in primary and secondary care.

Interventions compared

The analysis assessed lifetime costs and effects of ACE inhibitors compared with placebo.

Outcomes

The treatment effects were measured in terms of reduction of cardiovascular events: non-fatal MI, revascularisation, unstable angina, heart failure, cardiovascular -related deaths and other deaths. Health outcomes for the cost-effectiveness analysis are summarised in the form of Quality Adjusted Life Years (QALYs).

Model structure and assumptions

A Markov model was developed to evaluate the incremental costs and effects of lifetime treatment with ACE inhibitors in secondary prevention of CVD events in low risk post MI patients from a UK NHS perspective.

In a Markov model there are a finite number of health states. It is assumed that at any point in time, all patients must be in one and only one of the states. The model then replicates how a hypothetical cohort of people moves between the states. Figure 1 shows a schematic representation of the patients’ pathways. All patients start in the event-free health state. During each six-month cycle of the model, a proportion of patients enter one of the qualifying event health states (MI, heart failure, unstable angina, revascularisation and death) while the remainder stay in the event free state. Patients can experience more than one non-fatal event in subsequent periods of the model.

Figure 1

Model structure for the cost effectiveness of ACE inhibitors in low risk patients with preserved LVDF compared to placebo.

The rate at which people move through the model is regulated by transition probabilities, which describe the likelihood of moving between states over each model cycle (six months). For illustration, the equivalent annual transition probabilities for a 65-year-old patient on placebo are shown in Table 1. The probabilities are derived from the placebo arm of the meta-analysis (Al-Mallah, M. H., Tleyjeh, I M., Abdel-Latif, A. A. et al 2006).

Table 1

Annual probabilities for an untreated 65 year old men.

The model was run first assuming that the cohort was to receive placebo. The model was then re-run assuming that the cohort all received ACE inhibitors and complied 100% with transition probabilities adjusted to reflect the expected reduction in CVD events and revascularisations. Health care costs and QALYs were then estimated for each option by weighting the time spent in the various states by mean costs and ‘utilities’ (health-related quality of life) of the health states. The cost and utility data used in the model are described below.

The time horizon modelled is lifetime, with an assumed upper age of 100, by which time most of the cohort have died.

Treatment effects

The effectiveness of ACE inhibitors defined as the reduction in relative risks of mortality, heart failure, revascularisation and non fatal reinfaction was obtained from the meta-analysis (Al-Mallah, M. H., Tleyjeh, I M., Abdel-Latif, A. A. et al 2006).

Cost data

The NICE reference case specifies that costs should be measured from an NHS and personal social services perspective. These should include the direct cost of drug treatment and also potential savings from avoided treatments due to reduced incidence of CVD and hospitalisations. Costs were calculated using cost weights for each of the states of the model, multiplied by the time spent in each state. Costs are at 2005 prices. As per current NICE guidance, an annual discount rate of 3.5% was used for both costs and health benefits (National Institute for Health & Clinical Excellence. 2006b).

The costs of health states used in the model are shown in Table 2c. Costs for revascularization which includes hospitalisation were taken from the NHS reference cost 2005.(Department of Health Reference Costs 2005 2005) It was assumed that 67% of patients will have PCI and 33% will have CABG and the costs were weighted to reflect this (expert opinion). The cost of the well states was assumed to be the outpatient cost which includes the costs of medication and monitoring costs were taken from the Statin HTA (National Institute for Health & Clinical Excellence. 2006a). The subsequent costs of MI and unstable angina were assumed to be the same and were taken from the NICE hypertension guideline 2006 (National Institute for Health and Clinical Excellence. 2006). Costs of acute MI (non-fatal reinfaction) were assumed to be the same as those of patients on thrombolysis, which includes the cost of hospitalisation, (Hartwell, D., Colquitt, J., Loveman, E. et al 2005). The cost of death was zero. Costs of drugs were taken from the drug tariff (Prescription Pricing Authority 2006)

Table c

Costs of health states.

Cost of heart failure was taken from the NHS reference cost 2005,(Department of Health Reference Costs 2005 2005) and subsequent costs after heart failure were assumed to be the same as those seen in subsequent MI patients (expert opinion). Costs of events were reduced by 50% and doubled in sensitivity analysis.

Quality of life (Utility)

In the NICE reference case, the value of health outcomes – including beneficial and harmful impacts of treatment on mortality and morbidity – is estimated using the Quality Adjusted Life Year (QALY) approach. This requires estimates of survival and quality of life associated with each health state included in the model.(National Institute for Health & Clinical Excellence. 2006b)

The utility values used in the model are shown in Table 2d and Table 2e. The values were taken from literature or the Harvard cost effectiveness registry database (Harvard CEA Registry 1997)

Table d

Health state utility weights.

Table e

Utility weight by age.

Utilities were adjusted to reflect the fact that health related quality of life in the general population decreases with age (i.e. multiply the disease utility weight by age utility weight). Age utility weights were taken from the Department of Health, Health Survey for England (Department of Health 1998)

Sensitivity analysis

The model includes a base case analysis supplemented with both deterministic and probabilistic sensitivity analysis. The impact of utility, costs of revascularisation, cost of ACE inhibitors and baseline risks for mortality, revascularisation second MI, heart failure and unstable angina were assessed.

Probabilistic sensitivity analysis

A probabilistic sensitivity analysis was also done, where all parameters are assigned a distribution and are allowed to vary at the same time. The results are reported below in the form of cost effectiveness acceptability curves (CEACs). These curves detail the probability that each strategy is cost effective over a range of potential maximum willingness to pay values that the NHS can afford to pay for an additional QALY.

The CEACs demonstrate that ACE inhibitors are cost effective when compared to placebo. The probability that ACE inhibitors are cost effective is around 70% at £20,000/QALY threshold. As expected the probability that an intervention is cost effective improves as the willingness to pay increases. Thus for a threshold of £5,000/QALY the probability that ACE inhibitors are cost effective is 59%, while at £30,000/QALY the probability increases to 72%.

Figure 2Cost effectiveness acceptability curve, Cost effectiveness plane ACE Inhibitors versus placebo

Deterministic sensitivity analysis

A range of univariate sensitivity analyses were conducted to assess the impact of different input parameters on the base case results. Detailed results for all parameters are shown in the appendix.

Quality of life

The base case model assumed that the side effect profile of ACE inhibitors was the same as in the placebo arm. However when it was assumed that ACE inhibitors will result in loss of quality due to side effects of more than 2.1%, then ACE inhibitors would no longer be cost effective at £20,000/QALY threshold. For instance if the loss in quality of due to side effects are assumed to be about 2.5% the estimated ICERs is about £230,200/QALY. If the loss was 3% ACE inhibitors are dominated by placebo. Overall the result is sensitive to loss in quality of life due side effects of treatment.

Health state utilities were arbitrarily reduced and increased by 0.2. This did not affect the base case conclusions suggesting the model was is not sensitive to changes in health state utilities. The ICERs ranged between £3,370 to about £3,480/QALY.

Costs

Cost of events (cost of treating MI, heart failure, revascularizations, and unstable angina) were increased by 100% and reduced by 50%. The model remained robust with ICERs remaining ranging between about £3,300/QALY and £3,400/QALY in all cases examined.

Worse case scenario

A worse case scenario was examined were the cost of events were doubled, and treatment effects were set at their upper limit of the 95% confidence interval. In this case the ICERs increased to about £8,400/QALY. This is still within acceptable limits of what is usually considered affordable by the NHS. Thus the model is robust to the worse case scenario assumption.

Efficacy

Assumptions about the efficacy of treatment were tested using the 95% confidence interval. The model was robust in all cases when either the lower or the upper 95% confidence interval was used. When the upper 95% CI was used, the ICERs increased to about £6,100/QALY but were still within the range considered affordable by NHS.

RR of non CVD death, age and sex

Then model was robust to assumptions about the relative risk of death from other causes between the post MI cohort and the general population. The base case assumed a relative risk of 2. When it was assumed that there was no difference in mortality from other causes between the general population and the post MI cohort, the ICERs slightly increased to about £4,100/QALY.

The model was also robust to assumptions about age and sex. The estimated ICERs ranged between about £3,000/QALY for a 85 year old men to about and about £5,000/QALY for a 55 year old men. For women it ranged between £4,000/QALY for an 85 year old to about £5,500/QALY for 55 year old women. There was no big difference between sexes; ICERs were more favourable to men than women.

Sensitivity analysis; quality of life loss due to side effects

Sensitivity analysis; health state utilities ± 0.2

Sensitivity analysis cost of CVD events/health state costs

Sensitivity analysis; worse case scenario 1, doubling the cost of events and using upper confidence limit of the 95% CI

Sensitivity analysis; efficacy of ACE inhibitors treatment

Sensitivity analysis; relative risk of non CVD death

Sensitivity analysis; age and sex

Population and sub-groups

The model considered post MI patients with left ventricular dysfunction who met the inclusion criteria of the Carvedilol Post Infarct Survival Control in left ventricular Dysfunction (CAPRICORN) trial (Dargie, H. J. 2001). The model was run separately for different cohorts, defined by age (65, 75 and 85) and sex. The base case analysis is presented for 65-year-old men and women. However the trial evidence that the model this is based on included relatively few women (27%) or black patients, so the results may not be reliable for these sub-groups.

Interventions compared

The analysis assessed the costs and effects of carvedilol compared with placebo.

Outcomes

The treatment effects were measured in terms of prevention of cardiovascular events: non-fatal MI, hospital admission for heart failure, and cardiovascular-related deaths. Other cardiovascular events, including onset of stable or unstable angina, stroke, and peripheral vascular disease were not modelled, as they were not reported in the trial. Health outcomes for the cost-effectiveness analysis are summarised in the form of Quality Adjusted Life Years (QALYs), where one QALY represents one year of healthy life.

Model structure and assumptions

A Markov model was developed to evaluate the incremental costs and effects of lifetime treatment with third generation beta blockers for post MI patients with left ventricular dysfunction seen in primary care from a UK NHS perspective.

In a Markov model there are a finite number of health states. It is assumed that at any point in time, all patients must be in one and only one of the states. The model then replicates how a hypothetical cohort of people moves between the states.

shows a schematic representation of the patients’ pathways. All patients start in the event-free health state. During each six-month cycle of the model, a proportion of patients enter one of the qualifying event health states (MI, heart failure, or death) while the remainder remains in the event free state. Patients can experience more than one non-fatal event in subsequent periods of the model.

Figure 2Model structure for third generation beta blockers

The rate at which people move through the model is regulated by transition probabilities, which describe the likelihood of moving between states over each model cycle (six months). These transition probabilities are adjusted for each subgroup by age and sex. For illustration, the equivalent annual transition probabilities for untreated 65-year-old men are shown in Table 1.

Table 1

Probabilities for a 65-year-old untreated man.

The model was run first assuming that the cohort received no intervention (placebo). The model was then re-run for the treatment arm with transition probabilities adjusted to reflect the expected reduction in CVD events from the clinical trial data. Health care costs and QALYs are then estimated for each option by weighting the time spent in the various states by mean costs and ‘utilities’ (health-related quality of life) of the health states. The cost and utility data used in the model are described below.

The time horizon modelled is lifetime, with an assumed upper age limit of 100, by which time most of the cohort have died.

Treatment effects

The relative treatment effects of third generation beta blockers were taken from the CAPRICON trial (Dargie, H. J. 2001).

Cost data

The NICE reference case specifies that costs should be measured from an NHS and personal social services perspective. These should include the direct cost of drug treatment and also potential savings from avoided treatments due to reduced incidence of CVD disease. Costs were calculated using cost weights for each of the states of the model, multiplied by the time spent in each state. Costs are at 2005/06 prices. As per current NICE guidance, an annual discount rate of 3.5% was used for both costs and health benefits. (National Institute for Health & Clinical Excellence. 2006b)

The cost of health states used in the model are shown in Table 4. Costs of acute MI (non-fatal reinfaction) were assumed to be the same as those of patients on thrombolysis, which includes the cost of hospitalisation Hartwell 2005 (Hartwell, D., Colquitt, J., Loveman, E. et al 2005). Costs of heart failure were taken from NHS reference costs. Subsequent MI costs were taken from NHS hypertension guideline 2006.(National Institute for Health and Clinical Excellence. 2006) Subsequent heart failure costs were assumed to be the same as those of MI (expert opinion)

Table 4

Costs of health states.

Drug costs were taken from the prices quoted in the Drug Tariff, (Prescription Pricing Authority 2006) based on the usual dose for post MI patients. In the base case model a conservative approach was taken, using the most expensive dose of carvedilol 25mg and the use of the smaller dose of 6.25mg was tested in sensitivity analysis.

Cost effectiveness

The results of cost-effectiveness analysis are usually presented as Incremental Cost-Effectiveness Ratios (ICERs), which determine the additional cost of using one drug (X) per additional QALY gained compared with no intervention or another drug (Y):

ICERs = (cost of X - cost of Y)/(QALY of X - QALY of Y)

Probabilistic sensitivity analysis

A probabilistic sensitivity analysis was also done, where all parameters are allowed to vary at the same time. The results are reported below in the form of cost effectiveness acceptability curves (CEACs). These curves detail the probability that each strategy is cost effective over a range of potential maximum willingness to pay values that the NHS can afford to pay for an additional QALY.

The CEACs demonstrate that beta blockers are cost effective when compared to placebo. The probability that beta blockers are cost effective is around 93% at £20,000/QALY threshold. Even at lower thresholds such as £5,000/QALY beta blockers are still highly cost effective with a 90% probability of being cost effective. This suggest that beta blocker treatment in patients with left ventricular dysfunction is value for money.

Figure 3–4Cost effectiveness acceptability curve

Deterministic sensitivity analysis

A range of univariate sensitivity analyses were conducted to assess the impact of different input parameters on the base case results. Detailed results for all parameters are shown in the appendix. The following parameters were tested costs of drugs, cost of events, discount rate, utility, age, and relative risk of non-CVD deaths and efficacy of treatment.

Efficacy of treatment

The results are not sensitive to uncertainty over the magnitude of treatment effects estimated from the CAPRICON trial (Dargie, H. J. 2001). When the relative risks of carvedilol compared with no intervention were increased to their upper 95% confidence limits and reduced to their lower 95% confidence limits the results remained robust. The ICERs ranged between about £800/QALY to about £1500/QALY when both lower and upper confidence intervals are used.

Relative risk of non CVD death

Relative risk of non CVD mortality does not affect the conclusions of the model. If its assumed that patients post MI have the same risk of dying from non circulatory causes as the general population, the ICERs increase by £20 to £1110. If it was assumed that post MI patients have a six fold increase in risk of dying from non circulatory causes, the ICERs slightly fell by £60 to £1030/QALY. This suggests that the model is robust to this assumption.

Quality of life loss due to treatment side effects

The model is robust assumptions about loss of quality of life as a result of treatment side effects. If it was assumed that beta blocker treatment would result in a 1% loss in quality of life, the estimated ICERs would be about £1090/QALY and if the loss was assumed to be as big as 10%, the ICERs will increase four fold to £4360/QALY, still within the range considered affordable by NHS.

Health state utilities

The results are not sensitive to assumptions about the health state utilities used in the base case model. When the observed health state utilities were arbitrarily reduced by 0.2, the model remained robust. When they increased by 0.2 the results did not change. The ICERs ranged between about £1100 to about £1200/QALY.

Cost of health states

The model is not sensitive to assumptions about the health state costs. When they were doubled or reduced by 50% the ICERs ranged between about £1,000/QALY to about £1,300/QALY.

Age and sex

The model is robust to assumptions about age and sex. However it should be noted that efficacy data by age and sex is not available except for baseline mortality. The trial data mainly had male population aged between 60–65 years. For ages below 55 and above 70 and to females the results need to be interpreted with caution.

Worse case scenarios

A worse case scenario was examined were the cost of events were doubled, and treatment effect was set at its upper 95% confidence interval. In this case the ICERs increased by ten fold, to about £11,000/QALY. This however was still within the range of what is considered affordable by the NHS. Thus in this worse case scenario, the model remained robust.

Another scenario was tested where in addition to doubling the costs of events and using the upper 95% confidence interval for treatment effect, it was also assumed that ACE inhibitors will result in a 1% loss in quality of life due to side effects of treatment. The ICER increased to about £55,000/QALY. The model was not robust to this worse case scenario. It is however important to note that this is an unlikely scenario since the cost of events/ health states are not as high as suggested in this assumption and the efficacy of beta-blockers is not as low as again suggested in this assumption.

All sensitivity analysis applies to 65 year old men

Population

Patients who have had an MI within 3 months and who are unable to eat sufficient oily fish of 2–4 portions per week to meet the recommended intake of approximately 3.5g eicosapentaenoic acid (EPA) and 2.5g decosahexaenoic acid (DHA) per week. Subgroup analysis was performed to estimate the cost-effectiveness of supplementation for people who partially comply with the recommended dietary intake of oily fish.

Intervention

The analysis compared increased intake of omega-3 fatty acids from supplemental sources compared with no supplementation. The supplements considered were:

• 1g per day omega-3-acid ethyl esters (460mg EPA, 380mg DHA per capsule) (Omacor)
• 3g per day omega-3-marine triglycerides (170mg EPA, 115mg DHA per capsule) (Maxepa)

Both of the above options provide the recommended levels of EPA and DHA, assuming no dietary intake. The cost-effectiveness of supplements for patients who partially meet the recommended dietary intake of oily fish was estimated by assuming that patients use half the above doses: one capsule every other day for Omacor; or for Maxepa, instead of taking the supplements twice a day they will take them once a day.

The use of other over-the-counter supplements was not considered due to potential concerns about contamination in unlicensed products. The Cochrane review (Hooper, L., Griffiths, E., Abrahams, B. et al 2004) discusses potential concerns over cancers and neurological deficits that could possibly be increased due to dioxin and PCB contamination of fish oils. Although they found no direct evidence for this in the RCT or cohort data, they note that there is a lack of data on important outcomes. They conclude “independent analysis of the levels of toxins in named brands of fish oil supplements and oily fish sold for food should be more widely available” (p20 Hooper et al 2004).

Source of effectiveness data

In the absence of evidence of a difference in effect between dietary and supplemental sources of omega-3 fats, we assume equivalence (at equivalent doses of EPA/DHA). Thus evidence from trials of dietary or supplemental sources was pooled where relevant. Thus estimates of effectiveness for the base case analysis were taken from a meta-analysis of the results of DART 1 (Burr, M. L., Fehily, A. M., Gilbert, J. F. et al 1989) and GISSI-P trials.(GISSI Prevenzione Investigators. 1999) In addition data from GISSI-P and DART 1 was considered in the sensitivity analysis. This is based on the hypothesis that the effect observed in these trials is specific to patients who have recently had an MI, and that the results of the other main trial DART2 (Burr, M. L., shfield-Watt, P. A., Dunstan, F. D. et al 2003) are not relevant to this population.

The outcomes considered were total deaths, non-fatal MI, non-fatal stroke and revascularisations. Other outcomes such as heart failure, peripheral artery disease were not considered because very few events were recorded in both trials.

The company submission used the results of a four way analysis from the GISSI-P trial, and for our base model, we used the same results from the meta-analysis and did sensitivity analysis to estimate the impact of uncertainty over the treatment effects (as reflected in the 95% confidence intervals estimated from the meta-analysis) and various other model parameters.

The model was not adjusted for non-compliance. We assumed compliance issues were accounted for in the intention-to-treat results. Assuming 100% compliance tend to over-estimate costs, thus our model is conservative biasing the results against treatment.

The Cochrane review (Hooper, L., Griffiths, E., Abrahams, B. et al 2004) found no evidence of long-term side effects of omega-3. However, they did find that increased omega-3 intake was significantly associated with drop outs due to side effects and gastrointestinal (GI) side effects. The model included a loss of quality-of-life due to GI side effects, based on the estimated incidence in the Cochrane review, pooling results across all levels of cardiovascular risk. This assumes that the rates of such side effects do not differ for the post-MI population.

Estimation of costs and effects

The costs and effects of treatment were estimated over a lifetime horizon using a cohort Markov modelling approach. We used a twelve-month model cycle length. This period was deemed sufficiently short to ensure that it is unlikely that patients would experience two events within the same cycle.

Baseline non-CVD mortality rates in the absence of additional omega-3 were estimated by age from population data for England and Wales. That is Governments Actuary’s Department and Office for National Statistics (Government Actuaries Department 2006) (Office for National Statistics 2006) and adjusted for assumed increased risks following a first MI. Incidence rates for CVD mortality, non-fatal MI and stroke following a first MI, in the absence of treatment, were estimated from the observed rates in the trial control groups. Estimates for sensitivity analysis were also taken from a cohort study by Kaplan 2002 (Kaplan RC, Heckbert SR Furberg CD Psaty BM. 2002), in which 2677 patients were followed up for an average of 3.4 years after an MI. The Kaplan data was also used to estimate the distribution of CVD death, non-fatal MI and non-fatal stroke by age, and the proportion of these events that occurred in the first year after an initial MI. Risks of stroke following a first stroke were estimated from a cohort study by Hardie et al 2004 (Hardie K, Hankey GJ Jamrozik K Broadhurst RJ Anderson C. 2004). The incidence of revascularisation by age was estimated from Johansen et al 1998 (Johansen H, Nair C Taylor G. 1998).

It has been reported that the survival benefits of omega-3 following myocardial infarction appear early and do not persist in the longer term Ness et al 2002 (Ness, A. R. et al 2002). In the base case analysis we assumed that treatment effects do not persist beyond the longest trial period (3.5 years for GISSI-P), and that supplements are only used for this time. We tested these assumptions in sensitivity analysis.

Costs were estimated from the perspective of the NHS and discounted at an annual rate of 3.5% in accordance with NICE guidance (National Institute for Health & Clinical Excellence. 2006b).. The cost of omega-3 supplements were taken from the BNF (British National Formulary 2007). The cost for non-fatal MI and strokes were based on those reported in the NICE technology appraisal of statins Ward et al 2005 (National Institute for Health & Clinical Excellence. 2006a), adjusted for inflation.

Outcomes were estimated in the form of quality adjusted life years (QALYs). The quality of life (‘utility’) associated with various health states in the post-MI population was also taken from the NICE statin technology appraisal (National Institute for Health & Clinical Excellence. 2006a) Utility was adjusted for age, using estimates from a representative general population sample in the Health survey of England 1996 (Department of Health 1998). QALYs were discounted at 3.5% per annum.

Cost effectiveness

The results of cost-effectiveness analysis are presented as Incremental Cost-Effectiveness Ratios (ICERs), which estimate the additional cost per additional QALY gained using Omega 3 supplements compared with no supplements.

Sensitivity Analysis

The model includes a base case analysis supplemented with univariate deterministic sensitivity analyses to test the impact of uncertainty over various model parameters and assumptions.

Duration of treatment costs and effects

In the base model we assumed that the benefits and costs of supplementation persist for 3.5 years, as this was the longest duration of demonstrated effectiveness (GISSI Prevenzione Investigators. 1999). The duration of the other included trial (Burr, M. L., Fehily, A. M., Gilbert, J. F. et al 1989) was two years, and a long-term follow-up study found that treatment effects did not persist beyond two years (Ness, A. R., Hughes, J., Elwood, P. C. et al 2002) If we assume that treatment costs and effects only last for two years, omega 3 supplementation appears to be less cost-effective (Table 4). At age 55 the estimated ICER is £23,400, which is above the £20,000 threshold. Conversely, if we assume that treatment effects and costs persist for life, supplementation appears to be more cost-effective: £6,600 per QALY for 55 year olds.

Table 4

Sensitivity to duration of treatment costs and effects.

Clearly if the benefit of omega 3 is of limited duration, it will not be cost-effective to continue using, and paying for, supplements beyond this period. If we assume that the benefits will cease at 3.5 years and costs will persist for life – that is that people continue to receive omega 3 supplements but they do not derive any benefit from them - omega-3-acid ethyl esters supplements would only appear to be cost-effective for older patients (age 65 and over).

Source of effectiveness data

The base model used estimates of treatment effects from a meta-analysis of DART 1 and GISSI-P trials. When we considered results of the GISSI-P trial alone, the ICERs increased slightly (Table 5). Treatment remained cost-effective with the GISSI-P data, except for the 45 year old group, for whom the estimated ICER was above the £20,000 per QALY threshold. The results using the DART1 data alone were very similar to those using the pooled data.

Table 5

Sensitivity to source of effectiveness estimates.

Uncertainty over the size of treatment effects

The robustness of the results to uncertainty over the size of treatment effects was assessed using the upper and lower 95% confidence intervals from the meta-analysis (Table 6). The model results remained robust when the treatment effects were improved (set to their lower confidence interval) and worsened (upper confidence interval) for all outcomes except for all cause mortality. When the upper 95% confidence limits were used for all cause mortality, omega-3-acid ethyl esters supplements were no longer cost effective at the £20,000/QALY threshold. The estimated ICER was about £130,700 per QALY for a person aged 55 years, £37,800 for a 65 year old and £20,400 for a 75 year old.

Table 6

Sensitivity to upper and lower confidence limits of treatment effects: age 55.

Outcomes included in model

In the meta-analysis, only all cause mortality was found to be significantly different between treatment groups. If we assume no effect for the other outcomes (non-fatal MI, non-fatal stroke and revascularisation) and model mortality alone, the results become slightly more favourable (Table 7). Thus the model results are largely driven by the treatment effect on all cause mortality.

Table 7

Sensitivity to inclusion of non-fatal outcomes (MI, stroke and revascularisation).

Source of baseline event rates

In the base case model, the annual risks of cardiovascular disease events (non-fatal MI, non-fatal stroke, revascularisation and cardiovascular disease-related mortality) in the absence of supplements was taken from the rates observed in the control groups of the included trials (Burr, M. L., Fehily, A. M., Gilbert, J. F. et al 1989) and (GISSI Prevenzione Investigators. 1999) These included 6,676 patients followed up for an average of 3.3 years. Estimates from a cohort study (Kaplan RC, Heckbert SR Furberg CD Psaty BM. 2002) gave rather higher ICERs (less cost-effective) (Table 8). Supplements still appeared to be cost-effective for people aged 55 and older, but not for 45 year olds (based on the £20,000 per QALY cost-effectiveness threshold).

Table 8

Sensitivity to source of baseline cardiovascular disease risks (MI, stroke, revascularisation, cardiovascular disease-related death).

Relative Risk of non cardiovascular disease mortality

Packham C et al (Packham C Gray D Silcocks P Hampton J 2000) and Robinson M et al {Robinson, 2005 5309 /id} estimated that the relative risk of dying from non cardiovascular disease in a cohort of patients with coronary heart disease compared with the general population lies between 2 and 8. We used a conservative estimate of 2 for the base case model. If we assume that there is no difference in non cardiovascular mortality between the general population and those with coronary heart diseases, the ICERs rise and treatment no longer appears to be cost-effective for younger patients (Table 9). When the risk is assumed to be around 4, the ICERs fall.

Table 9

Sensitivity to assumed relative risk of non-cardiovascular disease mortality.

Cost of supplements

The base model used Maxepa, which was used in the DART1 trial and costs about £150 a year. We also tested the use of the supplement used in the GISSI-P trial (Omacor), which is slightly more expensive (£181 a year). The model results remained cost effective, although for younger patients the estimated ICER was very close to the £20,000 per QALY threshold (Table 10). If we assume clinical equivalence between these supplements, it will not be cost-effective to use the more expensive product.

Table 10

Sensitivity to price of supplements.

Impact of treatment side effects

The base model assumes that there is a 0.07% loss in quality of life due to treatment side effects. This is probably an over-estimate, as it is based on the assumption that the additional 5% of patients who reported gastrointestinal (GI) side effects in omega-3-acid ethyl esters supplement trials (Hooper et al 2004) all experience a permanent loss of quality of life of 1.4%, which is an estimate for of the quality of life loss due to “nausea, vomiting or diarrhoea for 5 days” Anderson 1985 (Anderson JP, Moser RJ. 1985). In reality many of these patients would have only experienced transient effects, and those with longer lasting or more serious effects would have been likely to stop taking the supplements.

The Hooper review (Hooper, L., Griffiths, E., Abrahams, B. et al 2004) found that there were significantly more drop outs due to side effects with omega 3 than in the control groups, although overall drop out rates were not significantly different. Any loss of effectiveness due to drop outs is included in the model through the intention-to-treat estimates of treatment effects. No adjustment is made for the reduced cost of supplements due to drop outs, but if anything this will tend to bias the model against supplementation.

In addition to the quality of life loss, the base case model assumes that an additional GP visit (cost £24) is required per patient per year to treat side effects. This is also likely to be conservative, since only a minority of patients report side effects and these appear to be relatively minor in nature.

Assumptions about the cost and quality of life loss due to side effects made little difference to the base case results, which still showed that omega-3-acid ethyl esters supplements were cost effective, except possibly for younger patients with particularly high treatment costs for side effects (2 or 3 additional GP visits for every patient each year) – see Table 11.

Table 11

Sensitivity to cost and quality of life loss due to side effects.

Discounting

NICE (National Institute for Health & Clinical Excellence. 2006b) recommends we discount both cost and benefits at 3.5% per annum. We tested three different scenarios: no discounting, 3.5% for effects and 6% for costs, and 6% for both costs and effects. The model was not sensitive to assumptions (Table 12).

Table 12

Sensitivity to discount rates for costs and effects (QALYs).

Health state utilities

The health state utilities used in the model were obtained from the literature. We tested the assumption that the mean health state utilities were 0.2 less or more than the ones we got from the literature. The model was not sensitive to this assumption.

Costs of cardiovascular disease events

The costs of cardiovascular disease events do not affect the model results. When the costs were increased by 100% or reduced by 50%, the ICERs changed very little (Table 14).

Table 14

Sensitivity to cost of cardiovascular disease events.