Gauging the Impact of Statins Using Number Needed to Treat

The first evidence-based guidelines to treat coronary heart disease (CHD) with lipid-lowering drug therapy^{1 - 7} were largely based on findings from the Scandinavian Simvastatin Survival Study (4S),⁸ which involved secondary prevention in patients with CHD. Thus, intervention was indicated only in patients with (1) hypercholesterolemia and (2) increased risk of CHD. Conventionally, this entailed dietary manipulation followed, if necessary, by treatment with a statin drug. However, mounting evidence from several subsequently published, large-scale, randomized clinical trials involving long-term use of statins has important additional implications on clinical practice. Thus, the cardiovascular benefits of long-term intervention with statins have now been amply demonstrated in: (1) men with hyperlipidemia but no known CHD,⁹ (2) patients with CHD but without hyperlipidemia,^{10 - 11} and (3) men and women with average total and low-density lipoprotein plasma cholesterol levels and no known CHD.¹² In summary, statins have been confirmed to confer such benefit in persons with and without CHD, and with and without hyperlipidemia. Choosing the least expensive alternative from among many "me too" statins might represent a narrow short-term view of cost-effectiveness. Though important, the vagaries of such selections will not be discussed here, as the prices of individual drugs vary from time to time, and from place to place. The broader strategic view of their usefulness is the focus of this article, and requires an understanding of how much benefit is likely to accrue under different circumstances over given periods of time.

The widespread practice of stressing important findings from randomized controlled trials in terms of relative risk reductions (RRRs) may not always be helpful for making judgments about whether to use statins and could possibly mislead. Another useful comparative term is the number (of persons) needed to treat (NNT) over a given period to prevent 1 person from experiencing an event calculated as the inverse of the absolute risk reduction. The NNT conveys an idea of an intervention's possible cost-effectiveness. This is because NNTs can enable taking account of (1) event-related reductions in longevity and quality of life (often inversely related to age), and (2) various consequential, direct, and less tangible treatment costs. Both of these assessments are presumed to be averted by the intervention. The NNT can also provide a means of appreciating patient numbers necessarily exposed to possible adverse effects and the inconvenience of long-term medication to prevent a single event. In a randomized clinical trial for example, if the 1-year event rate among persons taking placebo and active treatment were 2 per 1000 and 1 per 1000, respectively, the corresponding RRR due to active drug would be 50% and the NNT over 1 year would be 1000. By contrast, if corresponding event rates in those taking placebo and active drug were 20 per 1000 and 10 per 1000, respectively, the RRR would still be 50% but the NNT would decrease to 100, indicating a much better return for the effort put in. If not already reported, these parameters are simple to calculate¹³ from the results of individual published trials. In the West of Scotland Coronary Prevention (WOSCOP) study,⁹ among individuals randomized to placebo or statin, coronary event rates were 284 per 3293 and 174 per 3302, respectively. This gives an absolute risk reduction per person of 284/3293 − 174/3124 = 0.0226 over the 5-year follow-up period, which extrapolates to an NNT of 1/0.0226 or 44 persons to prevent a single event. Calculation of corresponding 95% confidence intervals is based on those for absolute risk reduction.

Though not necessarily strictly true, it is frequently assumed that RRR is independent of treatment duration. Similarly, it may be assumed that the NNT to prevent 1 event in a given number of years reflects an evenly distributed effect over time, such that an NNT per year can be readily extrapolated. On that basis, Table 1 provides a strategic overview of the impact of long-term statin use in a variety of circumstances. The 4 selected studies were the only large-scale, randomized, double-blind, placebo-controlled clinical trials involving the use of statins in which those recruited either had or did not have CHD or hypercholesterolemia. In the 4S trial, total mortality was the primary end point, whereas coronary death, myocardial infarction, and resuscitated cardiac arrest constituted secondary end points. In both the WOSCOP and Cholesterol and Recurrent Events (CARE) studies,^{10 ,14} the primary end points were death attributed to CHD or nonfatal myocardial infarction, and the Air Force/Texas Coronary Artery Prevention study (AFCAPS/TexCAPS) trial also included unstable angina or sudden cardiac death. On this basis, in all 4 trials it is of interest to compare (1) total (all-cause) mortality, (2) fatal and nonfatal coronary events (secondary end point in 4S and primary end points in the others), and (3) fatal and nonfatal strokes. Among those taking the respective statin in all 4 trials, both the hypolipidemic effects and the RRRs were of the same order. On the contrary, values for NNT (and NNT per year) differed dramatically. Thus, among 4S patients (with both CHD and hypercholesterolemia) in whom the overall risk of death was reduced by 28%, the NNT for 1 year is an estimated 63 persons to prevent a single coronary event. Among AFCAPS/TexCAPS participants (with neither CHD nor hypercholesterolemia), the NNT per year was 256. Among persons in the CARE study (CHD only) and WOSCOP study (hypercholesterolemia only), corresponding NNTs were intermediate.

Only AFCAPS/TexCAPS included unstable angina as a component of what was termed a fatal or nonfatal coronary event (the trial's primary end point). Since unstable angina per se was also a predefined secondary end point, exclusion of corresponding patient numbers from the count of those having a fatal or nonfatal coronary event would make for more similar coronary event comparisons in all 4 trials. On that basis and assuming that persons encountering unstable angina did not also suffer any of the other events constituting a primary end point, the recalculated NNT would be as high as 82 (or 429 each year). Furthermore, in the latter trial, overall (all-cause) mortality in the lovastatin group was 4% higher than in the placebo group, indicating that any reduction in coronary deaths due to active medication was cancelled out by increased mortality due to other causes. In the face of such mortality data among AFCAPS/TexCAPS participants, it becomes difficult to justify prescribing statins for persons lacking both CHD and hyperlipidemia, regardless of the 37% decrease in coronary events. The most cost-effective scenario for prescribing statins was in patients with both CHD and hypercholesterolemia (4S entry criteria).

Regarding NNTs as a means of gauging the value of statins to counteract cardiovascular disease, several other caveats are worth noting. First, an important advantage of quantifying benefits in terms of NNTs is that an intervention's relative impact on different facets of morbidity can be compared. Thus, in the trials where data were available, NNTs for preventing a stroke were much greater than for preventing a coronary event (Table 1). Corresponding RRRs for preventing a coronary event as opposed to a stroke were roughly of the same order or nearly 3-fold more (in WOSCOP). Respective NNTs (but not RRRs) indicate that statins have a more profound impact on the prevention of coronary events than strokes. A composite NNT could also be used to express the overall impact of these drugs in the avoidance of all cardiovascular events.

Second, it is commonly assumed that dosages giving rise to equivalent hypolipidemic effects confer comparable benefits in terms of cardiovascular outcome. While this tenet may be justified, it is not necessarily true, as such actions may depend on properties other than lipid-lowering activity.¹⁵ In terms of improving outcome rather than surrogate markers of outcome, benefits have been convincingly confirmed only for simvastatin, pravastatin, and lovastatin. However, the NNTs in Table 1 involved comparing equivalent dosages of different statins, based on surrogate markers only.

Third, it is also pertinent to examine the implications of using NNTs in predefined patient subgroups. The corresponding values are summarized in Table 2, but the resulting smaller patient numbers in each subgroup reduce the reliability of any inferences. Whether patients had diabetes appeared to have only a minor bearing on NNTs for coronary events in CARE, AFCAPS/TexCAPS, and Long-term Intervention with Pravastatin in Ischemic Disease trials. Similarly, there was no important interaction between the NNTs encountered and either gender or the presence of hypertension. Ironically, consistent and more substantial differences between contrasting subgroups only ensued with respect to smoking; smokers appearing to benefit considerably more than nonsmokers in terms of NNT.

Fourth, the impact of interventions with lipid-lowering agents relative to those of entirely different types of indicated drug interventions (eg, antihypertensive agents or hormone replacement therapy) could also be compared in terms of NNTs.

In contrast to antihypertensive drugs, statins are regarded as having: (1) a greater impact on heart disease (rather than stroke) and (2) fewer side effects (being better tolerated particularly in the elderly). Even in older persons, statins confer an advantage in terms of coronary events,^{8 ,10 - 12} whereas the possible cardiovascular benefits of hormone replacement therapy in postmenopausal women are still being debated.^{16 - 18}

All published, large-scale, placebo-controlled, randomized clinical trials evaluating cardiovascular outcome following long-term intervention with statins have yielded comparable RRRs (24%-37%). However, depending on the widely differing baseline absolute risks in those recruited to these diverse studies, respective NNTs were quite dissimilar (varying from 63 to 256 per year). Clearly NNTs provide an alternative, and possibly more sound basis than RRRs for establishing priorities, thus enabling more sensible deployment of finite and often limited resources.