The European Journal of Health Economics

, Volume 10, Issue 1, pp 111–119

An economic evaluation of a perindopril-based blood pressure lowering regimen for patients who have suffered a cerebrovascular event


    • School of ManagementUniversity of St Andrews
  • Neil Pumford
    • Nurseplus Ltd
  • Mark Woodward
    • Mount Sinai Medical School
  • Alex Doney
    • Ninewells Hospital and Medical School
  • John Chalmers
    • The George InstituteUniversity of Sydney
  • Stephen MacMahon
    • The George InstituteUniversity of Sydney
  • Ronald MacWalter
    • Ninewells Hospital and Medical School
Original paper

DOI: 10.1007/s10198-008-0108-3

Cite this article as:
Tavakoli, M., Pumford, N., Woodward, M. et al. Eur J Health Econ (2009) 10: 111. doi:10.1007/s10198-008-0108-3



Cerebrovascular disease (or stroke) is one of the main causes of long-term disability and the second leading cause of death worldwide. The economic impact of stroke is clearly seen, as it is the largest single cause of bed occupancy in hospitals in England and accounts for 6% of hospital costs. This analysis is the first to quantify the economic consequences of a blood pressure lowering regimen based on the PROGRESS study (perindopril-based regimen), for reducing future cardiovascular events.


A Markov decision analytical model was used to estimate the cost per quality adjusted life year (QALY) of blood pressure lowering in the treatment of patients presenting with a cerebrovascular event. The health states are based upon Barthel indices for which resource utilisation and health benefits have previously been estimated.


The participants for the economic analysis were obtained from the PROGRESS study database. 6,105 clinical study participants were recruited through both primary and secondary care centres.


The mean age was 64 years; 70% were male in the original study.


In the PROGRESS study, blood pressure lowering by a perindopril-based regimen was compared to standard care.

Main outcome measures

Cost per quality adjusted life year for the duration of the study (4 years) and for a time span of 20 years.


Using only direct hospital medical costs, the cost per QALY for a perindopril based regimen is £6,927 for the base study period and £10,133 for a 20-year time period. These results are sensitive to the cost of perindopril, the cost of the stroke unit, length of stay, and to a lesser extent, the cost of indapamide.


This analysis demonstrates a cost-effective treatment for patients suffering a cerebrovascular event with a blood pressure lowering regimen. The findings of this study are in line with current decisions and guidance by the national institute for health and clinical excellence (NICE) in England.


Blood pressure loweringCerebrovascularMarkovCost-effectiveness


Cerebrovascular disease is the second leading cause of death worldwide [1, 2], and one of the main causes of long-term disability in developed countries [3, 4]. Worldwide, about 5 million people die per year from stroke [1] while a further 15 million will have non-fatal strokes, of whom a third will be disabled [1, 4]. The risk of further strokes remains high among stroke survivors; one in six can expect to have another stroke within a 5-year period [4].

Circulatory disorders, including cerebrovascular disease (stroke and transient ishaemic attack [TIA]), coronary heart disease (myocardial infarction [MI] and angina), congestive heart failure (including left ventricular dysfunction) and hypertension, account for 10% of gross NHS expenditure in the UK. They are, by far, the leading cause of morbidity and mortality [5, 6]. In the UK, stroke is the third most common cause of death. In Britain, over 123,000 people have a stroke each year [7, 8], with over 56,000 of these are fatal [7], and cerebrovascular disease is the single most important cause of severe disability [9]. Financially, it is estimated that stroke accounts for over 4% of NHS expenditure and 6% of hospital costs [10, 11]. It is also estimated that 30% of patients die after a stroke within the first month and of those who survive for 1 year: 35% are significantly disabled, and about 5% will be admitted to long-term residential facilities [12], and at any one time there are 25–35 stroke patients in an average general hospital [13].

Stroke patients often suffer from coronary heart disease (CHD), and vice versa, reflecting the underlying atherosclerotic process so that the reduction in one is often accompanied by a reduction in the other [14]. Furthermore, about one quarter of strokes are recurrent [15] and potentially preventable. Therefore, given the large impact of stroke associated with high prevalence, its link with CHD, hospitalisation rate, morbidity and mortality have resulted in a number of interventions in order to reduce the incidence of the disease [1618].

It has been shown that blood pressure (BP) level is a strong predictor of both first and recurrent cerebrovascular events; lower BP is associated with a lower risk of stroke amongst both hypertensive and non-hypertensive patients [1924]. The perindopril protection against recurrent stroke study (PROGRESS) was the first to demonstrate that BP lowering after a cerebrovascular event reduced future vascular events, regardless of initial BP level [22].


The primary objective was to assess the economic impact of the perindopril-based treatment used in PROGRESS, compared to standard care, when applied to patients in the UK who have recently experienced a cerebrovascular event. Two time periods were examined: (1) a 4-year period (that is, within the average duration of the study); (2) the lifetime of patients, assumed not to exceed 20 years. The criterion used was the incremental cost per quality adjusted life year (QALY) gained. Only direct UK NHS hospital costs were considered. Indirect costs, such as loss of productivity, were not included in this study as there were no suitable data available. As such, the analysis is conducted primarily from the tax-payers’ (UK NHS) point of view including mainly hospital (acute and rehabilitation) costs as other NHS costs such as GPs' fees were excluded due to lack of suitable data. Other ACE inhibitor comparators could not be included in the model because no comparable head-to-head patient data were available.

Methods and data


PROGRESS has been described in detail elsewhere [19]. Briefly, 6,105 patients with a recent history of stroke or TIA from 172 centres in Asia, Australasia, and Europe were recruited and randomly assigned to active treatment (n = 3,051) or standard care (n = 3,054). Active treatment comprised a flexible regimen based on perindopril (4 mg daily), with the addition of the diuretic, indapamide (2.5 mg/d; 2 mg/d in Japan), at the discretion of physicians; 58% of patients were assigned combination therapy [22]. Randomised treatment was administered “on top of” standard care, including the use of other blood pressure lowering agents, cholesterol lowering drugs and aspirin, at the discretion of the responsible physician. Follow-up lasted for an average of 4.1 years. The mean age of patients was 64 years and 70% were male. The primary outcome was total stroke (fatal and non-fatal); secondary outcomes included major vascular events, MI and heart failure. TIA was not included among the outcomes, although it was defined among the inclusion criteria.

Active treatment gave a relative risk reduction of 28% (95% CI 17–38%, p < 0.0001) for recurrent stroke versus standard care, with similar reductions in stroke rates among both hypertensive and non-hypertensive sub-groups [22]. Benefit was also found for other cardiovascular outcomes [22, 25].

The economic model

The objective was to assess the cost-effectiveness (CE) of active treatment versus standard care. A comprehensive economic evaluation of stroke prevention will require that all appropriate health benefits and costs including hospitalisation, rehabilitation, and long-term care following strokes and recurrent stroke be considered and estimated as fully as possible over time. This would require a model that could incorporate both time varying transition parameters and costs. A widespread technique is the use of decision-analytical models of patient prognosis (subsequent to the choice of a management strategy) to estimate the cost-effectiveness of healthcare interventions. A common outcome measure of health improvement in CE is quality-adjusted life years (QALYs). This measure combines mortality and health-related quality of life (utility) gains by measuring the number of years of life saved adjusted for quality. A Markov decision analytical model was used [26], which are statistical representations of recurrent events over time within the decision analysis. Markov models are appropriate for use when the decision problem has an ongoing risk over time (for example, the risk of haemorrhage while on a specific therapy), when the timing of events is important, and when important events happen more than once as is the case in this paper. The ability of the Markov model to capture repetitive events and the time dependence of probabilities, costs and utilities provides a more accurate representation of clinical settings [26, 27]. Other studies have used Markov models in similar settings [28, 29]. Heterogeneity between patients is addressed through disaggregation, thus taking into account the patient’s group risk profiles. An important feature of the model is the incorporation of patient risk profiles depending on their initial conditions and their respective mortality rates. The decision tree ultimately tries to show what may occur in clinical reality and is compatible with the PROGRESS trial clinical path. It is assumed that patients in a given current health state can be treated as homogeneous groups, to be assigned the same resource utilisation, health benefit and utility value. In order to identify homogenous groups for the Markov Model patients were classified according to their health states (see Fig. 1b) at the end of each period, and were also identified by factors such as events experienced and the order of those events. Thus, individual patients were assigned to cohorts based upon (a) allocation to active/standard care, (b) suffering the following: stroke, myocardial infarction (MI), angina, percutaneous transluminal coronary angioplasty (PTCA), other hospitalisation, and (c) the order in which they experienced these events. This defined 42 patient cohorts. A heart failure group was also added for completion although this accounted for less than 1% of the events seen.
Fig. 1

a Decision tree diagram demonstrating clinical cohort. TIA was included among the entry criteria as a qualifying cerebrovascular event but was not included among the study outcomes [22]. b Markov model based upon Barthel index states and death. State 1 81–100, State 2 61–80, State 3 41–60, State 4 <40. Each cycle lasts for 1 year, the model is run for 4 years over the trial period and 20 years for life time

Markov models are state transition models that evolve over time. Patients are assumed to be in only one state at any one time, and transitions between states were calculated over a specified period of time known as the Markov cycle, here taken to be a year that was dictated by the availability of the data. However, Markov models suffer from what is called ‘Markovian’ or memoryless assumption. That is, the movement between states is only dependent on the current state and not the previous history before entering that state and so it is assumed that patients in a given state can be treated as a homogeneous group. This problem can, however, be overcome to some extent by adding extra states to the model [26, 30]. Furthermore, while more than one event can occur in between measurements, the Barthel indices [31] were only recorded at the end each year (Markov cycle). This mirrored the data collection in the PROGRESS study which after the first year of follow-up patients were seen at annual visits. The Barthel index was used as surrogate for quality of life and mapped to utility values. The Barthel index captures ten activities of daily living and on each the patient is assessed as to their functional ability (e.g., feeding, walking, and bathing). Activities are rated as 0, 5, 10, 15 depending on the activity, 0 being worst functional state. The original scale has a maximum of 100. Whilst the Barthel index has critics, it is widely used in stroke trials, is disease sensitive, and in mapping to utilities had a correlation of 0.59, which appears reasonable. It also measures parameters important to the patient.

Outcome measures/endpoints

The Markov model is based on four health states defined by Barthel index scores. The literature review showed no consistency between authors and their research in the choice of cut off for Barthel indices, for example, one study might use categories of <80, 80–99, 100, and another 45–80, 81–100, etc. As no one set of categories had been widely accepted, we relied on the literature to give us an indication of meaningful changes and this was refined following discussions with clinical experts. It is reported and concluded [31] that a 20-point difference is highly likely to represent a genuine, clinically meaningful change. On the basis of this and other similar studies [32, 33], and after discussions with a team of consultants, a 20-point difference between Barthel indices were assumed to represent true differences in functionality and disease severity levels.

Figure 1a shows the process and the structure of the decision tree assumed. It shows the clinical pathway that a patient could potentially follow. All patients start with a stroke or TIA. This is represented by the square on the left hand side of the diagram. Based on the low frequency of more than two such events in PROGRESS, only the first two events were used to identify the cohorts. The patient cohorts were then allocated to four Health States based upon the Barthel index: <40, 41–60, 61–80, 81–100, and death. From year to year, individuals were allowed to move between these states, except death, which is obviously an absorbing state (Fig. 1b). While there were some differences in the strength (but not direction) of the effect of the perindopril-based BP reduction among participants with haemorrhagic and ischaemic strokes, including embolic strokes, in PROGRESS [22, 34], the numbers were too small for reliable analyses by stroke subtype in the present context.

Transition probabilities

The probability of moving from one state to another is called the transition probability. In practice, these are measured by the proportion of patients moving from one health state to another (or remaining in the same state) within each patient group and for each Markov cycle. Transition probabilities were obtained from PROGRESS yielding 840 observed probabilities for the 4 years of observation. Beyond this 4-year trial period, each transition probability, except those for transitions to death, was assumed to follow a beta distribution whose parameters were estimated from PROGRESS. When there were not enough observations to estimate the standard deviation of the beta distribution, triangular or uniform distributions were used instead [35].

Mortality data

Transition probabilities to the death state within the first 4 years were obtained from PROGRESS. Few studies provide information on trends in the long-term mortality of stroke [3640]. However, all these studies report mortality by cause(s) of death of patients after a first nonfatal stroke and not by their risk profiles. To estimate mortality rates taking into account patients’ risk profiles, standardised mortality ratios (SMRs) were constructed for those with recurrent strokes, those with other events (e.g., MI) and those who had no further events beyond their initial stroke. Mortality rates for each of these three groups, separately for each sex, were estimated from the Dundee Stroke Unit Database [41]—(mean patient age of 64 years, as in PROGRESS) over a 10-year period, and were matched as closely as possible for age, sex, and illness (risk profiles) with that of PROGRESS and then combined with England and Wales’ deaths per 1,000 [42]. The sample population was assumed to follow the average age in the PROGRESS trial of 64. To do this, the patients were ordered by their age at the time of their first stroke and then were selected one by one to the population while maintaining the average age closely to 64. Having selected the patients based on their risk profiles, age and gender mortality rates were estimated. These rates were then used to calculate the number of deaths which would be expected from the general population. For each pair of gender and age represented by the patients the chance of such a person dying is looked up in the mortality rates of England and Wales [42]. The number returned is multiplied by the number of patients in the population with that gender and age. These are summed to give the number of expected deaths. Dividing the number of actual deaths by the number of expected deaths gave the standardised mortality rate (SMR). SMRs were then averaged from year 4 onwards to provide an estimate beyond the sample period: 4.8 for those who had recurrent stroke, 3.3 for those with other events and 1.7 for those with no further events, relative to the general population. Table 1 gives standardised mortality rates for different risk profiles after stroke and their range. These are broadly consistent with other studies [3739]. The SMRs were then used to weight the chances of death beyond 4 years of survival that would be expected, for each year of age and for each sex [42].
Table 1

Standardised mortality rates for different risk profiles after stroke, average age of 64 years




Recurrent stroke






No further events



Source: SMRs were estimated using Dundee Stroke Unit Database, Ninewells Hospital, Dundee, and Office for National Statistics, mortality per 1,000 for England and Wales [42]

However, as with many other statistics, SMRs have to be interpreted with caution. They can vary from sample to sample and small sample sizes hence small number of deaths could make interpretation difficult. Another inherent problem with SMRs is that they cannot be used to compare different geographical areas, as each area’s population profile weights the age specific death rates differently. Finally, the cause of death reported in the case report may not reflect the underlying risk profile of patients and the SMRs used in this paper will go some way to address this problem.


Resource utilisation included cost of initial hospitalisation, number of bed days in hospital, type of unit, and prescribing (or pharmacotherapy) costs. No costs from outside the hospital were considered.

Typical resource utilisation for the initial hospitalisation was obtained from the Dundee Stroke Unit Database and a Delphi [43] panel of a team of consultants who specialise in stroke in Ninewells Hospital, Dundee (Table 2).
Table 2

Resource utilisation and costs


Cost per procedure

Laboratory tests


CT scanb


Carotid duplex Dopplerb


ECG (1 for stroke and 2 for MI)a


Echocardiogram (needed in 30% in stroke and 80% in MI of cases)a


MRI scan (needed in 8% of cases)b


Blood tests on admission (biochem (£80) haematology (£30))a


Other direct costsa


AHP (Allied Health Professionals)a


X-ray and treadmill test (67% of MI)c


Source: Dundee Stroke Unit database, Ninewells Hospital, Dundee and expert advice

aCommon to both

bStroke patients only

cMI patients only

Prices were obtained from Scottish Health Service Costs [44], and the British National Formulary (2005) [45]. The Dundee Stroke Unit Database also provided the number of days spent in the hospital for every health state based on the Barthel index, events experienced and the order of the events. Direct medical costs for each health state were based on hospitalisation according to the type of event (stroke/TIA or MI/angina) and length and place of stay (stroke unit, rehabilitation unit, geriatric unit (long-term stay) in hospital over a 10-year time period. Direct costs for each health state were then calculated using resource usages from Dundee Stroke Register (Stroke Unit Database), prices from the NHS official statistics, and Scottish Health Service Costs (2004). All costs are expressed in 2005 values. As far as the authors are aware, this is the first study that the length of stay has been measured over time as this allows measuring the costs more accurately (cost data are not reported here due to their large volume). These data and the information noted above from the Delphi panel were combined to estimate direct medical costs of each health state.

Table 3 shows the cost per day in a stroke, rehabilitation, and geriatric unit. The published daily cost for geriatric units was £171 [44]. These costs include the cost of the bed and direct physician/ nurse time. It was assumed that after 40 days in the stroke unit a patient would be moved to a geriatric unit. As no costs per day for stroke units were available, we used the cost per day in general medicine and adjusted it to reflect the higher costs of stroke units (estimated as 20% more expensive) [44]. Thus, estimated costs of in-patient care per day in stroke units was £355. This is broadly in line with other studies [46]. The costs per day for rehabilitation and long-term care were unadjusted.
Table 3

Costs per day in a stroke unit


Scottish health service costs, 2004

Stroke unit


Rehabilitation unit


Geriatric unit (long-term stay)


a20% over the cost per day in general medicine

The initial costs of hospitalisation (investigations and indirect staff) were estimated at £571 and £559 for stroke/TIA and MI/Angina patients, respectively (see Table 2). Costs of medication were obtained from the British National Formulary (2005): perindopril £10.95 and Indapamide £2.82 each for 30-tablet packs, respectively. It was assumed that indapamide would be allocated within the active treatment group to the same proportion of patients as in PROGRESS (58%). For the lifetime analysis, triangular distributions were assumed for costs in the extrapolations.


The European health-related quality of life 5 dimensions (EQ-5D) questionnaire was used to assign quality of life to health states. The Sheffield Health Economics Group provided data from a study of three residential care homes involving more than 3,000 people, which related EQ-5D to Barthel scores (correlation of 0.59). Table 4 shows summary statistics. These data were then used to construct beta distributions for each health state as they are bounded on 0–1 interval.
Table 4

Utility values for each health state

Barthel index


Standard deviation













Sensitivity analyses

Probabilistic sensitivity analysis is used to assess the uncertainty of the cost-effectiveness of the model and each parameter/variable/transition probability had to be assigned a probability distribution in order to conduct this analysis. Depending on the availability and the nature of each variable, different distributional assumptions were made. Triangular distributions were used for cost data as the data was generally skewed and only a small number of observations were available in some cases. Hence we could not use log normal or gamma distributions. For the utility parameters and transition probabilities beta distributions were used because they are bounded on the interval 0–1 [47]. Finally, triangular distributions were used as proxies for beta distributions, if there were not enough observations [34, 48].


An annual discount rate of 3.5% has been used for both costs and benefits [49], as recommended by the UK Treasury and supported by the National Institute of Health and Clinical Excellence. Discounting is used to bring future costs and benefits into their equivalent present values for comparison purposes.


Four-year analysis

The discounted incremental cost per QALY gained for perindopril-based treatment over the 4-year period was estimated to be £6,927. Table 5 shows the breakdown into the separate components.
Table 5

Disaggregated cost-effectiveness analysis


Cost (£)

Incr cost (£)

Efficacy (QALY)

Incr Eff (QALY)

C/E (£)

Incr C/E (£)

Standard care














Simple sensitivity analyses showed how this result depended upon cost assumptions. When the cost of perindopril was reduced by 10% (to £9.86), the incremental cost per QALY dropped to £3,485, and if the price of perindopril was reduced to £8.65 (a 21% reduction) the treatment became cost-saving. When the cost of care in the stroke unit was increased to £440, active treatment also became cost-saving (dominating). Finally, if the cost of indapamide is to increase by 30% to £3.67, the cost per QALY gained for perindopril-based treatment increases to £8,484.

Life time analysis

Costs, transitional probabilities and utilities were simultaneously allowed to vary in 1,000 Monte-Carlo simulations. The median cost-effectiveness was thus estimated to be £9,455 per QALY with mean £10,133 and standard deviation of £3,344. Over 99% of simulations had a cost per QALY below £25,000 (Fig. 2) a figure accepted by technology assessors such as NICE, as being cost effective [50].
Fig. 2

Incremental cost-effectiveness ratios (ICER)


Our findings suggest that based on a 4-year period, the incremental cost per QALY for a perindopril based regimen versus standard care is £6,927. The life time incremental cost per QALY for BP lowering with a perindopril based regimen is £10,133 and this is well below notational boundaries accepted in recent appraisals by NICE.

The analysis was performed on active treatment versus standard care. It should be noted that many patients were already on anti-hypertensive medications and many more will have commenced these during the study. In the study, individual clinicians were allowed to decide whether one or two drugs were used in a patient. This was then matched with single or double placebo. The percentage on two agents and one agent was 58 and 42%, respectively.

The analysis was performed on the overall results for two reasons; firstly this is the randomised population (with the use of intention to treat data) and thus any internal bias is reduced and secondly the overall population captures real-life practice where a person with a lower blood pressure might be given only one agent and one with a higher starting BP given two agents. Hence the analysis should be valid for extrapolating to a population similar to that of the PROGRESS study.

Whilst the order of the events was felt to be important, it proved too complex to match the timing of the event, i.e., event 1 in year 1 and event 2 in year 4, to the resource utilisation. The timing of the event to some extent will be reflected in the Barthel index as this should be lower soon after a stroke and slowly improve. Thus the year in which the stroke took place would be associated with a lower Barthel index and greater resource utilisation.

Direct comparisons are problematic as different authors use different models, different assumptions, and different outcomes. However, the cost per QALY calculated for PROGRESS is similar to other health economic analyses of major cardiovascular intervention studies. For example, a health economic analysis put a value of £5,502 per life saved on lipid lowering strategies from Scandinavian Simvastatin Survival Study (4S) [51]. In the UK it is estimated the cost to the NHS for every patient experiencing a stroke is £15,306 over 5 years, and when informal care costs are included this figure increases to £29,405 [52], thus suggesting that total costs are nearly twice the NHS hospital costs.

The original study was not designed to decide what agents should be used to lower blood pressure and therefore the question remains unanswered as to whether other anti hypertensive agents would deliver the same cost-effectiveness ratio. It does demonstrate the cost-effectiveness of blood pressure lowering in a population at risk of future events.

Whilst the analysis uses UK resource utilisation and costs, these results can be broadly generalised since the clinical benefits were obtained from a wide geographical population and the clinical benefits in PROGRESS were broadly consistent by age, sex, and geographic region, with no evidence of significant heterogeneity [53]. However, the pattern of healthcare provision may vary from country to country, which may have differing impacts on resource utilization (e.g., number of days in hospital, investigations, etc) and thus could result in differing overall cost-effectiveness. Furthermore, the UK is usually a low user of health care resources in comparison to many countries. Other countries with greater resource use/ costs might find a lower cost-effectiveness value.


This work was funded from an unrestricted research grant from Servier Laboratories Ltd (UK) and a program grant from the Australian National Health and Medical Research Council. We thank Sam Colman for statistical support from The George Institute, University of Sydney, Australia.

Conflict of interest

Neil Pumford was until recently an employee of Servier Laboratories (the manufacturer of Coversyl/perindopril and Natrilix/indapamide), Manouchehr Tavakoli has received a grant in relation to the economic modeling from Servier. John Chalmers and Stephem MacMahon have received grants from Servier as Chief Investigators for PROGRESS and ADVANCE administered in Sydney. John Chalmers, Stephen MacMahon, and Mark Woodward have all received honoraria from Servier for presentations regarding the studies at scientific meetings. Alex Doney received some financial support from Servier to attend the European Stroke Conference 2006. Ronald MacWalter has received honoraria from Servier for presentations and reimbursement for attending International symposia.

Copyright information

© Springer-Verlag 2008