Monday, April 12, 2010

Forget Blood Tests, Just Treat with Statins? More (Partial) Evidence

Warning: If you are up to here with my ragging on about statins and cholesterol, skip this post. The reason I am pursuing this is to try to get to the bottom of the influence of Pharma marketing over both medical science and actual physician prescribing behavior, and I think the statin/cholesterol/prevention story is a rich and multifaceted case study.


A study published in January in Annals of Internal Medicine would seem at first glance to be yet another important piece of evidence with regard to the "lipid hypothesis"--whether statins actually work solely or even primarily by their lipid lowering effects, or if they actually work by some other means such as anti-inflammatory. At a closer look, it does indeed add further evidence that the lipid hypothesis is shaky at best, and it makes a good point about practical strategies for managing heart disease risk. But ultimately it fails to be as important as it might be because it compares only options A and B, when the really best options are probably to be found among C, D, and E.

A team headed by Drs. Rodney Hayward of U-Michigan and Harlan Krumholz of Yale (IMHO, meaning the study is both high quality, and not bought off by Pharma) set out with funding from the VA to try to compare two strategies for preventing heart disease with statins. The "treat to target" strategy is what all the guidelines recommend--measure cholesterol levels, start a statin, keep checking the blood levels and adjust the medicine until you have reached a target level of cholesterol (or, specifically, "bad" cholesterol or LDL). (These guidelines, I hasten to add, were mostly written by folks whose pockets are deeply lined with industry cash, as we have previously discussed.) Hayward and colleagues wanted also to test a "tailored treatment" strategy, which consists of two doses of two different statins. On this strategy, you decide what the cardiac-disease risk level of the person is from some population measurement tool such a Framingham. If the person is medium risk you pick a lower dose of a generic statin. If they are at especially high risk you pick a higher dose of a more potent statin. In neither "tailored" case do you bother to measure cholesterol levels, now or later. The "tailored" strategy is, so far as I interpret the literature, more evidence-based than the "treat to target" strategy.


Cutting to the chase, Hayward and colleagues knew that to do this study as a randomized clinical trial would require huge numbers of subjects, about a decade, and many more bucks than they had on hand (being funded by the VA). So they chose instead to do the study as a mathematical modeling exercise. The things that make their model apparently quite robust were, first, that they made a set of assumptions that slanted in favor of the "treat to target" strategy; and second, that they did a whole slew of sensitivity analyses to see if changing any of many different assumptions changed the outcomes.


Their results were a resounding win for the "tailored treatment" strategy. They could not find a set of assumptions that made the "treat-to-target" strategy come out better in terms of patient benefit. (They did not consider cost in their study, probably simply taking it for granted that the strategy that did away with the need to chase cholesterol levels all over the place would be the cheaper strategy, especially when combined with maximum use of generic drugs.)


So two things seem to follow, at first glance, from this study. One: treating with statins based on risk classification, not based on cholesterol levels, is the better way to go. Two: as both treatment strategies resulted in saving lives and quality-adjusted life years over baseline, then statins are indeed good for primary prevention of heart disease, and the only question is how to use them best.


So: is the first glance view accurate? I think not.


Consider what the study, for all its sophistication, did not test. There was no comparison between trying to use statins for primary prevention of heart disease, using either strategy, and the best reported results of diet/exercise/lifestyle interventions. Also, there was no comparison between a single, low dose "tailored treatment" strategy and the authors' two-dose variation. If one is to believe the work of James Wright and John Abramson in particular, there should have been no difference.


So the question, then, is on what basis did Hayward and colleagues decide that primary prevention with statins works? The answer--they assumed it to be so. They admitted up front that as part of their plan to give up as much handicap as they could to the opposing strategy, they made two assumptions-- that reducing LDL cholesterol accounted for 100% of the statins' beneficial effects; and that reduction in LDL cholesterol is a 100% accurate accounting of how much benefit an individual receives from taking a statin. They immediately add that the first assumption is quite controversial and the second is demonstrably untrue.


They chose to make these assumptions because it suited their narrow purpose of demonstrating the superiority of their "tailored treatment" strategy, and for that purpose it was quite an effective way to design their study. For our purposes, however, those same assumptions mean that we cannot use this study to test out the more skeptical--and I think more evidence-based--positions on statin use in primary prevention.

When I ran a draft of this post by EBM-mayvin Jerry Hoffman at UCLA, he kindly added the observation that I needed to mention "number needed to treat" (NNT). The way Jerry put it, suppose that we located the most up-to-date statistics and presented to a bunch of higher-risk men the option to take statins for primary prevention of heart disease, or not, using the so-called "tailored treatment" approach now that we know it's superior. There is no guarantee that these men would choose the statin therapy even under these "superior" treatment conditions--if we told them frankly about how many of them would have to take a statin before one of them had benefit (such as avoiding a heart attack or death). If men knew that as many as several hundred would have to take statins (being exposed to all the risks of adverse reactions) so that one would get a benefit, they could very reasonably say "no thanks." So even the "superior" treatment strategy is potentially undesirable if the NNT statistics from many large clinical trials are valid.


It is nevertheless nice that we have one more argument to use against extensive screening for cholesteol and LDL and increasing statin treatment levels if the targets are not met. Finally, in conclusion, Hayward et al make an excellent point. Why, they ask, did the "treat to target" strategy ultimtely fail? Their conclusion-- it overemphasized one single heart-disease risk factor, LDL, at the expense of not taking into account the total picture of the individual patient's heart disease risk status. That led to some patients being overtreated, some being undertreated, and overall, less net benefit. This is yet another argument that if anything should guide the use of statin therapy, it's the patient's overall risk status and not any single blood test.

6 comments:

JPB said...

For most of us, our doctors are adamant about taking statins (even with very low risk factors). When is this information on the failure of the lipid hypothesis going to filter down to the average doc?

Michael Kirsch, M.D. said...

The public needs to be reminded that cholesterol levels are surrogates, and not actual disease states. The extent that an individual truly benefits from lowering cholesterol levels is much lower than the patient believes. The benefits are apparent in population studies. Most folks on statins exaggerate the benefits of the drugs. www.MDWhistleblower.blogspot.com

Marilyn Mann said...

Just a clarification -- you would need to measure baseline lipids in order to calculate the Framingham Risk Score, or whatever risk score you are using. The NHLBI version of FRS uses total cholesterol and HDL-cholesterol, along with age, gender, systolic blood pressure, smoking status and use of antihypertensives. There are other risk scores, but I'm fairly certain they all use some information on lipids (e.g., TC, HDL, LDL or some combination), along with other information. I agree that in the tailored treatment model you would not need to test lipids again, according to my understanding.

Anonymous said...

Thanks for sharing the information

http://drug4health.blogspot.com/

Victor said...

You may want to see here (http://archinte.ama-assn.org/cgi/content/abstract/167/10/1076) what patients in our study did when shown the risk and reduction afforded by statins given their estimated risk and assuming fixed-dose statin treatment (not treat to target). So no need to speculate, at least not among diabetes patients. As you can see, a substantial proportion chose not to take statins, even though they are considered high risk patients. The tool used can be downloaded here (http://mayoresearch.mayo.edu/mayo/research/ker_unit/form.cfm). None of this was funded by pharma, a key feature of our work at the KER UNIT at Mayo Clinic.

Howard Brody said...

"Victor's" study seems to be excellent and well worth reviewing. To my surprise, a few more patients in the intervention group, that got the sophisticated decision aid based on cardiovascular risk level, elected to begin statin therapy compared to the control group that just received a standard pamphlet (30% vs. 21%). However, the decision device was more effective in getting those people at higher risk to decide to use statins and vice versa.