For 40, 50, 60 and 70 year old men and women, generic CCB, thiazide, ACE inhibitor, ARB and beta blocker are all cost-effective and even cost-saving antihypertensives, even when two or all three are used in combination when this is indicated. CCB seems to be the most cost-effective alternative and consequently the first-line drug. If patients do not reach the treatment targets on CCB, thiazide is the most cost-effective add-on treatment. The sensitivity analyses, however, indicate considerable uncertainty in the ranking, and other factors such as side effects may well dictate the choice in the clinical setting.
The modelling showed lower incremental health benefit in younger than in older age groups. This may seem counterintuitive. To test this result, we performed validating analyses without discounting (Figure ). From these, we concluded that the reason for the counterintuitive results is discounting. When the model was run without discounting, the incremental benefits are greater in younger age groups. Discounting decreases life years gained more in younger than in older age groups. Because benefits from antihypertensives in terms of life years gained occur at a late stage in life, discounting leads to lower incremental health benefit in the younger compared to the older age groups.
We used a 4% discount rate throughout this study, because this is currently recommended in Norwegian guidelines. We are aware, however, of the ongoing discussion regarding whether costs and/or effect should be discounted, and at what percentage [48
]. If no discounting were applied, the age effects would be reversed in the sense that the ICERs would be more favourable for the young than the older age groups (Figure ). If only costs were discounted, the difference in cost-effectiveness between age groups would be relatively small (Figure ).
The estimated life year gains from taking CCB for the remainder of the lifetime compared to no antihypertensives is close to 0.5 (undiscounted). This may seem to be a small number when being 50 years old and expecting about 30 more years to live. Two points are worth mentioning, however. First, the 0.5 year is an average, and some patients may gain much more while others have no benefit. Some patients may die from other causes such as cancer or accidents and have no benefit from the antihypertensive treatment. Others may avoid a fatal stroke in their 60's and live until 80 because they started using antihypertensive drugs in their 50's. Second, 0.5 year is a considerable benefit in a public health perspective.
Models may be constructed in several different ways, and structural differences between them may in some cases result in considerably different results and alter the conclusions. Inclusion of heart failure and angina in the model made the results much more uncertain because the efficacy data are not consistent for these outcomes.
We have not identified other cost-effectiveness analyses of all these antihypertensives. However the UK guidelines from 2011 [50
], which are based on similar efficacy documentation as our analyses, propose ACE inhibitor, CCB or possibly a low-cost ARB as the first choice. These guidelines also suggest that that a combination of CCB and either ARB or ACE inhibitor is the recommended combination of two drugs. If three drugs are to be combined, a thiazide-like diuretic is to be combined with the two-drug combination. These recommendations seem to fit well with our results.
Our analyses are based on meta-analyses from a recent systematic review [2
]. The results are not very different from other meta-analyses [51
]. Hence we assume that our results would not be substantially different if they were based on other meta-analyses of antihypertensive drug trials.
In our estimates of lifetime costs (e.g. Table ), costs include consumption of all modelled resources. Hence, it is not possible to read drug prices directly from the undiscounted column of this table due to the complexity of NorCaD. This advantage with NorCaD helps avoid jumping to conclusions that the cheapest drug is the most cost-effective if no statistically significant evidence is available. For instance would a small decrease in rates of myocardial infarction more than outweigh a small increase in acquisition costs of a cheap generic drug.
Strengths and limitations
The NorCaD model is comprehensive in the sense that it captures more CVD events and health states than most previous models. It is also a strength that the model is based on country-specific data for some of the crucial input parameters. The NorCaD model is designed primarily for primary prevention strategies for cardiovascular disease, and is therefore useful not only for statins and other pharmaceutical interventions, but also non-pharmaceutical interventions such as dietary advice and exercise.
While several previously published models have estimated the risk of CVD events on the basis of risk equations (typically the Framingham risk equations), we used observed incidence rates in the population and adjusted these rates up or down depending on the presence or absence of risk factors. The advantage of this approach is first that we avoid bias introduced by uncertainties in risk equations, and second that we avoid uncertainties introduced by distance in time or distance in geography. Our approach, however, is not without problems. Most important in this context is that we use register data for incidence rates, and there may be limitations in the quality of these registers and there may be inconsistencies between the registries.
The validation process proved that the input to the model need to be somewhat adjusted to fit Norwegian mortality data. This is a limitation of the model, which might be more consistent if it were based on fewer data sources, such as Framingham data. However, we considered the use of old data from the US likely to generate more bias.
Most of the trials that form our evidence base had duration of less than five years [2
]. The life year gain generated through five years of treatment however is modest (usually less than 10% of the total gain). For the time beyond five years, we lack solid empirical data and simply assume that the relative effect stays constant. The assumption of "continued benefit" [53
] is not necessarily true, and may overestimate the effect of treatment.
We have not incorporated side effects into our model, mainly because we use life years gained as measure of effectiveness and then possible fatal side effects will be captured in the clinical trials. The costs of side effects, however, may not be captured. Thiazide for example may have diabetogenic effect. Hence, thiazide may have a smaller effect and higher costs over time than what is assumed in our analyses. Some side effects may also have positive impact, such as the diuretic effect of thiazides. Whether incorporation of side effects would change the results of these analyses is uncertain, however. It should be noted, however, that side effects in terms of mortality is captured in the model because we used intention-to-treat data from trials. Such side effects are only omitted from the model if they occur after end of the trials.
As mentioned in the methods section, we assumed a multiplicative relationship when modelling combination treatment. Cohort studies have demonstrated an exponential relationship between blood pressure and CVD mortality. Thus one may argue that the risk reduction is proportional to the reduction in blood pressure. If one assumes that one drug results in a given reduction in blood pressure, independent of level, a combination of drugs will result in a multiplicative relationship. Even though some combination trials have been undertaken [2
], these are too few to represent a basis for evaluating all clinical relevant combination therapies. Hence, the results with respect to combination therapy are more uncertain than those based with single drug comparisons.
The idea of a model is mimicking real life. All models are however to some extent a simplification of the clinical setting. In this model we chose not to include combinations of health states, such as for instance heart failure and stroke. In addition, we did not include all possible cardiovascular events, such as intermittent claudication. These simplifications would influence results to some extent, but we regarded the possible gain in accuracy to not be worth the hassle. This was mainly due to the fact that trustworthy data on occurrence and progression for these patients would be difficult to obtain.