A Predictive Model of the Health Benefits and Cost ...

A Predictive Model of the Health Benefits and Cost Effectiveness of Celiprolol and Atenolol in Primary Prevention of Cardiovascular Disease in Hypertensive Patients

Richard J Milne1, Stephen Vander Hoorn2, Rodney T Jackson3 1 Health Outcomes Associates Ltd, Auckland, New Zealand 2 Clinical Trials Research Unit, University of Auckland, New Zealand 3 Dept of Community Health, University of Auckland, New Zealand

Richard J Milne PhD, Health Outcomes Associates Ltd

1

Summary This study compares the antihypertensive and lipid modifying effects of 5 years of treatment of mild to moderate hypertension with celiprolol and atenolol, beta-blockers which are widely prescribed for treatment of hypertension in New Zealand. It also models the cardiovascular risk reduction and the cost effectiveness of monotherapy with either agent. The effects of celiprolol and atenolol on systolic blood pressure (SBP), total serum cholesterol (TC) and high density lipoprotein cholesterol (HDLc) were obtained from a pooled analysis of published studies of celiprolol and atenolol (n=243 and 494 respectively). Although celiprolol and atenolol had similar effects on SBP (14.4 vs 16.0 mm Hg reduction), celiprolol reduced the ratio of total serum cholesterol to high density lipoprotein cholesterol by 10.2% (95% confidence intervals -16.4%, -4.0%) but atenolol increased the ratio by 7.7% (3.4%,12.0%). The 5-year absolute risk of an initial coronary or cerebrovascular event or cardiovascular death were computed for hypothetical cohorts of patients treated with either agent or remaining untreated, using a non proportional hazards Weibull accelerated failure time (AFT) model based on first- and second-generation Framingham data . Inputs to the model were age, gender, smoking status, SBP, TC and HDLc. The relative and absolute risk reduction were computed using the changes in SBP, TC and HDLc obtained from the pooled analysis of clinical trials. Average life months gained by therapy were computed as differences between KaplanMeier survival curves estimated from the model plus differences in 5-year cardiovascular death rates combined with average life expectancy obtained from life tables. Direct medical costs included drug treatment and the costs of acute care for coronary and cerebrovascular events avoided by therapy over the 5-year treatment period. The model shows that in the lowest risk base case (male age 60, non-diabetic non-smoker with SBP = 160 mm Hg; 5-year absolute cardiovascular risk 12%) that celiprolol (271 mg/day) is about 2-fold more effective than atenolol (77.7 mg/day) in reducing coronary risk but equally effective in reducing cerebrovascular event risk. The number of individuals needed to treat for 5 years to avoid one coronary event is about 30 for celiprolol vs 60 for atenolol. Therapy with celiprolol yields more (undiscounted) life months and at current prices the cost per life year gained by therapy is slightly lower. Both drugs are cost effective by international standards in treatment of patients with 5-year absolute cardiovascular risk >10%. Taking into account hospitalisations avoided for cerebrovascular and coronary events, treatment for 5 years with celiprolol costs the same or slightly less than treatment with atenolol despite a 19% higher tablet price, at the dosages used in the clinical trials. Both drugs are more effective at prolonging life and more cost effective in patients at higher levels of absolute cardiovascular risk. These findings are sensitive to the drug dosages, tablet prices and the discount rate, but robust to the costs of treating coronary and cerebrovascular events and to other parameters. Based on epidemiological and clinical data, replacing atenolol with celiprolol in patients with mild to moderate hypertension but without overt cardiovascular disease is predicted to have similar effects on stroke risk but to be substantially more effective in reducing the risk of coronary events, at no additional direct medical cost over a 5-year treatment period.

Richard J Milne PhD, Health Outcomes Associates Ltd

2

Introduction As in other countries, beta blockers and diuretics are widely prescribed as first line

treatment of hypertension in New Zealand (NZ), consistent with the Core Health Services `Guidelines for the Management of Mildly Raised Blood Pressure in New Zealand’ (1995), US national guidelines (JNC-V 1993) and European guidelines (Pyorala et al. 1994). Atenolol and celiprolol are commonly prescribed for treatment of mild to moderate hypertension in NZ, although their relative clinical effectiveness in reducing cardiovascular risk, and their cost effectiveness, are unknown. Over 50% of hypertensive patients in NZ who are prescribed beta-blockers receive monotherapy. The role of beta blockers including atenolol in secondary prevention of cardiovascular disease is well established (Wadworth et al. 1991), however its role in primary prevention of coronary risk in patients without overt heart disease is less clear. Monotherapy with atenolol 50-100 mg/day reduced stroke risk but had no significant effect on coronary risk in a trial conducted in over 4000 hypertensive elderly adults without pre-existing heart disease (MRC Working Party 1992) although combination therapy with a diuretic plus atenolol reduced both stroke and coronary risk in elderly patients with isolated hypertension (SHEP 1991). Celiprolol reduces systolic and diastolic blood pressure effectively (Milne & Buckley 1991) but its effects on cardiovascular risk have not been established in long term trials. Meta-analyses of clinical trials have shown that commonly prescribed antihypertensive agents (mostly thiazide diuretics and beta-blockers) reduce cerebrovascular risk to the extent predicted by epidemiological data, but their effect on coronary risk is less than that predicted (Collins et al. 1990; Hebert et al. 1993; MacMahon & Rodgers 1993; Insua et al. 1984; Gueyffier et al. 1997). One contributing factor could be the unfavourable effects of older antihypertensive agents on serum lipids; this would be expected to partially offset any potential benefits for coronary risk, but have minimal effect on cerebrovascular risk which is less dependent on the serum lipid profile (Prospective Studies Collaboration 1995). Such meta-analyses cannot provide information about individual agents or classes of agents because they pool the effects of several different diuretics and beta blockers. A recent meta-analysis of 18 randomised clinical trials (Psaty et al. 1997) has confirmed that beta blockers reduce cerebrovascular event risk but was unable to show any effect on coronary risk. The clinical trials analysed included patients who received propanolol, atenolol, pindolol and metoprolol, but not celiprolol. Howes et al. (1996) showed from a pooled analysis of published studies reporting the effects of 5 antihypertensive agents on serum lipids that celiprolol and atenolol had equivalent effects on SBP but celiprolol reduced whereas atenolol increased the ratio of total serum cholesterol (TC) to HDL-cholesterol (HDLc) which is a prognostic indicator of cardiovascular risk (Kinosian et al. 1994). These authors proposed that then different effects of celiprolol could be mediated by its mild beta-2 adrenoceptor agonism. Using regression equations based on data from the Framingham Heart Study (Anderson et al. 1991a,b), these authors predicted a 3-fold greater reduction of relative coronary risk for celiprolol than for atenolol. The same model forms the basis of the Core Health Guidelines for the Management of Raised Blood Pressure in New Zealand (1995). The present study was designed to establish the effects of celiprolol and atenolol on SBP and serum lipids and to model their effects on cardiovascular risk as well as Richard J Milne PhD, Health Outcomes Associates Ltd

3

their cost effectiveness in treatment of patients with mild to moderate hypertension. It has 3 main parts. First, we determined the effects of monotherapy with celiprolol or atenolol on the serum lipid profile using a pooled analysis of 16 published trials in patients with mild to moderate hypertension. Next, we substituted the effects of each drug on SBP, TC and HDLc into regression equations based on data from the Framingham Heart Study (Anderson et al. 1990, 1991) to predict the relative risk of cardiovascular events and death, and the survival benefit in life months gained. Finally, we incorporated drug acquisition costs plus hospitalisation costs for coronary and cerebrovascular events into the same model, to predict the incremental cost of 5 years of monotherapy with celiprolol compared with atenolol and average costeffectiveness ratios for both drugs. We have used a marginal analysis which addresses the question of the investment in antihypertensive therapy that is required over a 5-year period to provide survival benefits both within and beyond this period (Johannesson & Jonsson 1992). This analysis was conducted from a limited societal perspective and therefore included public and private direct medical costs (Government, insurer and patient) although the costs of rest home care as well as non medical and indirect costs were excluded.

Methods Pooled Analysis of Comparative Studies Our pooled analysis consisted of 15 of the 23 studies of antihypertensive drugs used by earlier investigators (Howes et al. 1996) excluding those that did not study either celiprolol or atenolol. The original search criteria have been stated in this publication (Medline; period 1988-1994; English language; randomised; compared lipid lowering ability). These studies were supplemented by one further head-to-head comparative study of celiprolol vs atenolol in which effects on blood pressure but not serum lipids were reported (Silke et al. 1966). In all, 16 studies were included in our pooled analysis, including 243 patients who received celiprolol and 494 who received atenolol (table I). Although we considered confining our analysis to 5 studies that compared celiprolol with atenolol, this study did not provide sufficient power to meet the objectives. A pooled analysis was performed to determine the effects of celiprolol and atenolol on blood pressure and lipid levels. Variables of interest were systolic and diastolic blood pressure (SBP, DBP), total cholesterol (TC) and high-density lipoprotein cholesterol (HDLc). As the manner in which estimated changes in blood pressure and lipid levels was reported varied between the studies, for consistency in the pooled analysis these estimates were all reported as the mean change and standard error. Mean change in this instance was defined as the average change from the beginning to the end of the study for patients on a particular treatment who completed the study. The estimates from each study were subsequently pooled using the inverse variance weighted method (Nony et al. 1995) to give an overall average treatment effect. 95% confidence intervals (CIs) were constructed, and tests against untreated and between treatment groups were performed using t-tests. Since the inverse variance method relies on the fact that all estimated changes come from the same overall effect (i.e. a fixed effects model), tests of homogeneity (tests of whether the mean changes were statistically significant across studies), Richard J Milne PhD, Health Outcomes Associates Ltd

4

were performed. For parameters that showed heterogeneity, a random effects model was used (Greenland1994). In setting the 95% confidence limits on the ratio TC/HDLc, it was assumed that the effects of both drugs on TC and HDLc were uncorrelated. The mean change and standard error of TC/HDLc was therefore obtained assuming independent distributions of TC and HDLc. Average dosages for studies in which dosage was titrated to effect were calculated based on the relative proportion of patients in whom a dosage increase was considered necessary. The pooled treatment dosage was calculated by weighting each study according to sample size. Predictions from an Epidemiological Model The outcomes used were selected from the Framingham Study as follows: cerebrovascular events (cerebral infarction, transient ischaemic attack, embolus, haemorrhage); CHD (coronary heart disease) events including myocardial infarction and CHD death plus angina pectoris and cardiac insufficiency; and death from cardiovascular disease including all of the above plus congestive heart failure and peripheral vascular disease. The 5-year absolute risks for coronary events, myocardial infarction, cerebrovascular events including transient ischaemic attacks, and cardiovascular death (coronary, cerebrovascular events, congestive heart failure or peripheral vascular disease) were determined from a non proportional hazards accelerated failure time (AFT) model developed for the first- and second-generation Framingham cohorts containing 5573 subjects of both sexes (Anderson et al. 1990, 1991a,b). The absolute risk of cardiovascular events was defined as the sum of coronary events plus cerebrovascular events (Anderson et al. 1990). This method of calculating absolute risk differs slightly from the New Zealand Core Health Guidelines on Management of Hypertension (1995) and the National Heart Foundation Guidelines for Detection and Management of Dyslipidaemia (Mann et al. 1993) which used a Cox proportional hazards model based on the same data set (Wolf et al. 1991) to predict the absolute risk of cerebrovascular event. Compared with the AFT model, the Cox model predicts a 35 to 60% higher absolute risk of cerebrovascular events at ages >50 in the lowest risk base case, but a 40% lower relative risk reduction for stroke with antihypertensive treatment (unpublished observation). Use of the AFT model for cerebrovascular events suggests a slightly higher threshold for antihypertensive treatment but predicts a relative risk reduction with beta blockers that is closer to that reported recently (Insua et al. 1994; Psaty et al. 1997). The other advantage of the AFT model for cerebrovascular events is that it includes TC and HDLc as covariates, consistent with recent evidence that reduction of serum lipid levels with statins reduces stroke risk (Corti et al. 1997; Crouse et al. 1997, Simvastatin Scandinavian Survival Study 1994; Sacks et al. 1996). These regression equations from the AFT model were applied to hypothetical cohorts of non-diabetic male and female patients with the following characteristics: no prior history of cardiovascular disease; mild to moderate hypertension treated with either celiprolol or atenolol or no antihypertensive agent (`no treatment’); specified combinations of other risk factors. In the first base case analyses, total serum cholesterol and HDLc levels were set so Richard J Milne PhD, Health Outcomes Associates Ltd

5

that the absolute 5-year cardiovascular risk for a non smoking non diabetic male age 60 years was 12% (table II). Drug dosages were set at the weighted mean values obtained from our pooled analysis of 16 studies of celiprolol and/or atenolol (271 mg/day for celiprolol and 77.7 mg/day for atenolol). Base case analyses were also performed for male and female non smokers and smokers with SBP = 180 mm Hg and variable lipid profiles (absolute 5-year cardiovascular risk at age 60 = 15%, 18%, 21% and 24%). To accommodate possible delays between initiation of treatment and reductions in cardiovascular risk (Scandinavian Simvastatin Survival Study 1994; Sacks et al. 1996; Shepherd et al. 1995), any changes in risk due to alterations in lipid levels were considered to begin after a lag time of 1 year. A zero lag time was assumed for the effects of blood pressure reduction. Input parameters for the model were the age of commencement of treatment, gender, smoking status (having quit smoking prior to 1 year before treatment, or not, according to the Framingham data set) and presence of diabetes. The age range for commencement of treatment was limited to 40 to 70 years to correspond to the epidemiological data set (Anderson et al. 1990, 1991). Early cohort data such as the Framingham study underestimates the strength of the association between risk and events (`regression dilution bias’: MacMahon et al. 1990). We corrected for this effect by multiplying the effects on SBP by 1.6 and the effects on both TC and HDLc by 1.2 (Davis et al. 1990; Johanneson et al. 1996; Law et al. 1994; MacMahon et al. 1990; Martens 1992). The gain in life months conferred by each drug within a 5-year treatment period was estimated as the difference between `untreated’ and `treated’ Kaplan-Meier survival curves predicted by the AFT model of cardiovascular death. The additional survival benefit at the end of the treatment period was then estimated as the 5-year absolute risk reduction for each treated cohort obtained from the model, multiplied by the ageand gender-specific life expectancy obtained from 1990-92 NZ Life Tables (Statistics NZ 1992). This assumes conservatively that the survival risk returns immediately to that of the untreated cohort after the treatment period. The life years gained beyond the 5-year treatment period reflect differences in survival during the treatment period only. Life expectancy estimated from life tables may be optimistic for individuals who are at moderate levels of cardiovascular risk, therefore we reduced it by 33% in a sensitivity analysis. In tables and figures, life months gained are presented undiscounted, but costs and life years gained or QALYs were discounted at 5% per annum for cost-effectiveness ratios. To discount benefits, we assumed an exponential (DEALE) approximation of survival distribution applied to the age- and gender-specific life expectancy (Beck et al. 1982). Economic Model Drug costs during the 5-year treatment period included wholesaler and retailer markups plus a fixed monthly dispensing fee plus container cost [$2.93 less Goods and Services Tax (GST) which is considered a transfer cost]. We assumed that patients in all 3 cohorts (untreated; treated with either drug) would consult their general practitioner biannually (at a cost of $35 per consultation less GST) and Richard J Milne PhD, Health Outcomes Associates Ltd

6

would have no other diagnostic tests specifically related to hypertension. This assumption holds for an incremental analysis but different patterns of resource consumption will affect the average cost-effectiveness ratios slightly. We also assumed that all patients who remained alive during the treatment period would continue to take the antihypertensive agent, whether or not they experienced a cardiovascular event. Antihypertensive therapy reduces the incidence of coronary and cerebrovascular events. The avoided costs of acute therapy for coronary or cerebrovascular events prevented by antihypertensive drug therapy were obtained by estimating the acute costs of managing a coronary event and a cerebrovascular event, and multiplying each of these costs by the number of initial coronary or cerebrovascular events avoided by therapy during each year of the 5-year treatment period, estimated from the AFT model. Recurrent events were excluded. This is a conservative assumption because the risk of recurrence is higher after an initial event, and agents that defer or prevent initial events also defer or prevent recurrences. We also assumed conservatively that the cardiovascular risk would return to that of the untreated cohort immediately that treatment ceased, so that costs were not avoided after this period. Ongoing costs of care for patients who are institutionalised following a major cerebrovascular event were excluded because of their uncertainty. In calculating costs, no correction was made for the different numbers of survivors in each cohort because the difference in absolute risk between treated and untreated cohorts is small [e.g.