When to start a statin is a preference-sensitive decision

In this issue of Circulation, Heller et al report results of a simulation model suggesting that the ACC-AHA primary prevention lipid treatment guidelines treat many more people with a statin but also save many more lives compared with ATP III.¹ These findings are consistent with previous reports,^2,3 but their results further suggest that starting a statin at age 40 in everyone regardless of CVD risk would extend statin treatment to over 28 million more Americans but would further and substantially improve the public’s health, but only if the disutility associated with pill burden is quite low.

Public vs. Individual Decisions

If starting a statin in all adults at age 40 would really save hundreds of thousands of QALYs, implementing such a policy would seem to be a public health priority. But there are several problems with this perspective, principally, taking a statin is an individual decision, not a public health decision like interventions to improve air quality or a decision with externalities, like treating contagious diseases. For an individual decision without externalities, an individual’s chance and magnitude of net benefit (absolute risk reduction minus absolute risk increase plus/minus uncertainties) is the only meaningful consideration.⁵ This may sound heretical, especially coming from a professor of public health, but the ethical and legal standards are clear. When counseling an individual patient about treatment, the most relevant question is almost always, “What is the magnitude of and uncertainty bounds for estimated net benefit for the considered treatment?” Heller et al present a very relevant estimate in this regard. If their base-case estimates are correct, adoption of the treat at age 40 policy, compared to AHA/ACC guidelines, would average 1 additional QALY gained for every additional 1,080 statin treatment-years (ie, NNT= 1 QALY gained per 108 patients treated for 10-years).

But this average NNT hides dramatic variation in the expected benefits among individuals in the additional 28.8 million Americans who would start a statin under a treat at age 40 strategy beyond those treated by the ACC/AHA guidelines. For a 40 y.o. women with a baseline risk of < 1% starting a statin would probably result in net harm even under the optimistic assumptions, and a 55 y.o. with a 7.4% 10-year CVD risk would have about the same estimated net benefit as a patient with a 7.6% 10-year risk. Personally, although an age criterion only is certainly quite simple, I think estimating an individual’s CVD risk, as proposed by the ACC/AHA guideline, is sufficiently simple to move the most important question to, “At what CVD risk should a statin be started?²

Preference-sensitive decisions

An appreciation of the individual perspective for individual decisions inevitably leads to acknowledging a continuum of benefit and the existence of grey zones—a range within which the right decision is almost completely determined by an individual patient’s personal preferences.^4–6 In other words, as the chance of benefit declines (NNT = 2, vs. 25 vs 100, vs 1000), a point must inevitably come when the best decision is highly sensitive to how the patient feels about pursing a small chance of benefit against the treatment’s hassles and potential harms, known and unknown. This phenomenon has been demonstrated for a variety of conditions and decisions, from intensifying glycemic control to taking a daily aspirin.^6,7 When the NNT per QALY is relatively low (NNT per QALY < 50), competing risks and treatment hassle and adverse effects tend to be relatively unimportant, because the benefits are so substantial. In contrast, as NNT per QALY further increases, treatment decisions soon become highly preference sensitive. How sensitive? Well, the 0.00384 treatment disutility that Heller et al found would neutralize the marginal benefits of the treat at age 40 strategy is equivalent to someone saying that in order to take a daily statin for the next 10 years, I expect to gain, on average, at least 2 weeks of high quality life. Personally, this is not the type of low therapeutic-window for which I’m willing to make strong recommendations to my patients, especially given a variety of unknowns that could dramatically lower the risk/benefit tradeoffs.⁸

Unknowable unknowns

Donald Rumsfeld famously referred to “…known unknowns. …. things that we now know we don’t know.” Such unknowns are the raison d’être for researchers, but as Sec. Rumsfeld was implying, decision-makers need to make decisions based of the information in hand (see Known Unknowns, below). What people appreciate less is that there are frequently unknowable’s in medicine. The null can never be proved, especially when it’s something as difficult to study, such as being on a statin for 20–25 years has no major long-term adverse effects. Could being on a statin for 25 years result in a 25% increase in decline in cognitive function or a 35% acceleration in the decline in muscle health associated with aging? Such effects are unknowable given any data source currently available.⁹ Since starting statins at age 40 would result in tens of millions of people being put on a daily, highly biologically-active medication 10–20 years earlier than under current guidelines, such questions are relevant. For example, if being on a daily statin for 25yrs had even a modest negative impact on common aspects of aging, it would not just reverse the marginal benefits of routinely starting a stain at age 40, it would be a public health disaster. How likely is it that such undetectable long-term harms exist? Who knows, since how can you estimate the frequency of something that is currently undetectable? I would propose that the unknowability of long-term harms is yet one more reason that when the chance of an individual benefiting is small (such as potentially 1 QALY gained per 1000 treatment years), that at most, shared decision-making without a strong recommendation, either for or against, is in order.

Known unknowns

Simulation models are powerful tools, but for informing policy, the known unknowns can have a large impact. Heller et al fully acknowledge that one highly influential assumption in their model is unknown—how much does a non-fatal CVD event causally increase a person’s future risk of CVD events and mortality. This may seem easily determined, but is epidemiologically complex. Heller et al assume that a non-fatal CVD event approximately doubles future risk, including mortality. This estimate has strong evidence when predicting a person’s future risk, but could be a dramatic overestimation in terms of the benefit of preventing a non-fatal event. Having a heart attack increases a person’s predicted risk of future events for both causal and non-causal reasons. Substantial tissue damage almost certainly causally increasing future mortality risk for an individual, but minor CVD events (which are a large proportion of all non-fatal events) increase future risk for the same reason that a high CAC score increases predicted risk—because it is a marker for having greater underlying vascular disease than expected. Since most of the estimated 1 QALY saved per 1080 statin-years is due to extrapolations of future impacts due to non-fatal CVD events, there is a chance that Heller’s et al’s estimated benefits for treating lower CVD risk individuals is a substantial over-estimate.

Another uncertain assumption bears mentioning in passing. Heller et al results are fairly favorable regarding the efficiency of the ATP treat to LDL target approach. In part, this is related to the assumption that a non-fatal CVD event causally doubles future CVD risks as discussed above, but it is probably also heavily influenced by the model’s assumption that the relative effect of a statin is directly related to baseline LDL level. In the absence of direct evidence supporting this assumption and some evidence suggesting that this is not the case,¹⁰ the ACC/AHA guideline committee dropped LDL targets and adopted risk-based approaches. It would be quite simple for the CTT group to test the relationship between baseline LDL and a statin’s relative effects using data they have in hand. Answering this important clinical question would let us determine whether an individual’s absolute risk reduction is best estimated by using their 10-yr CVD risk alone, or whether their baseline LDL level is also important.¹⁰ If I was forced to guess, my suspicion is that the impact of baseline LDL has a modest, but clinically important, impact on a statin’s relative effect on outcomes, but not as strong as that assumed by Heller et al. There is also a chance that the answer differs for primary and secondary prevention.⁹

Summary Thoughts

Heller et al rightly point out that if there are no unknown harms of long-term statin use that starting a statin much earlier would have a small chance of helping an individual but a large public health benefit since it would extend statin treatment to over 28 million more Americans. Since the chance of each individual benefiting is so small, even a small pill burden would eliminate this benefit and a modest long-term adverse effect of statins (like an exponentially increasing risk of diabetes over 20 years) would result in this guideline causing substantial net-harm. How one feels about pill burden and potential yet unknown or unknowable treatment harms is more an issue of personal viewpoints and preferences, which in my view, puts them in the realm of shared decision-making and not clinician recommendations.