The supposed ‘gold standard’ method of medical research is the ‘randomised controlled trial’. What this means in drug therapy research is that a group of individuals are randomly assigned to the treatment being studied or placebo, and the outcomes and effects compared. In the ‘best’ trials, neither the study participant not the investigators know who is taking what until the ‘code is cracked’ and the analyses are done. These studies are referred to as ‘double-blind’ studies.
‘Blinding’ is important because a lack of it can introduce bias into the study and corrupt its findings.
Let’s imagine, for example, that a study is being conducted in which a statin (cholesterol-lowering drug) is being pitted against a placebo. Imagine, also, that for whatever reason (we’ll discuss how, later) an investigator knows that a participant is taking the statin (not the placebo). He or she is then, perhaps, more likely to dismiss chest pain reported by the participant as ‘non-cardiac’.
The thinking (even unconsciously) can be that if this person is on the ‘correct’ treatment and their cholesterol is well-controlled, heart-related pain such as angina is unlikely. There might also be less tendency for such a patient to be hospitalised or even be subject to certain procedures (such as insertion of a stent into a coronary artery – an example of what is termed ‘revascularisation’).
These sorts of outcomes (angina, hospitalisation, revascularisation) are sometimes referred to as ‘soft outcomes’, because they are quite subjective, and depend quite a lot on the judgement of health professionals, This is not so true for ‘harder’ endpoints such as deaths due to heart attack or stroke.
The ultimate hard endpoint is ‘overall mortality’ (where the ‘diagnosis’ is not in doubt and it’s not open to interpretation). The other good thing, by the way, about overall mortality is that it is a measure that encompasses everything (deaths from all causes).
The thing is, though, some study investigators choose to include soft endpoints in analyses. The rationale is often that lumping a bunch of different endpoints together makes it more likely that the result will reach ‘statistical significance’. What this means in reality, often, is that overall benefits can be claimed for a drug that actually did nothing in terms of improving hard endpoints such fatal heart attacks and overall mortality.
A very interesting paper published recently in the BMJ makes the points above, but also draws our attention to how blinding may be lost, including in studies on statins [1].
Generally, when a patient is engaged in a study, they do not live in a bubble during the process. They still, say, have health issues and go and see their regular caregivers. When a doctor looks at someone’s health records, they may notice that since their patient entered a statin study, their LDL-cholesterol level dropped from 3.5 mmol/l to 1.5 mmol/l. It could be argued that it might be difficult for this doctor to avoid concluding that the patient is on the active drug (statin), and not placebo. They (as explained above) may be more likely, say, to dismiss chest pain reported by the patient as ‘non-cardiac’ in origin. They may be also less like to refer the patient for hospital admission and the chances of revascularisation go down too.
It occurs that it is not impossible for blinding to be lost for the participant too. Finding out that one’s LDL-cholesterol has dropped from 3.5 to 1.5 mmol/l could lead some to be more likely to dismiss, say, some chest pain, and be less likely to report it to their doctor.
The author of this piece points out that in the majority of statin trials, we have no idea how successfully blinding was ensured. This is not a minor issue, either – it introduces bias that might be corrupting the evidence base for statins (and other drugs).
What to do? The author has two suggestions:
1. One idea would be to enforce blinding strictly. However this would limit, say, access to health information (e.g. cholesterol levels) by regular caregivers.
2. Alternatively, we can accept that blinding can be compromised, and choose endpoints that are less prone to bias. In other words, we should concentrate on analyses of good, hard endpoints such as overall mortality.
As the author points out, the first suggestion might compromise care of patients. Regarding the latter, the author tells us that: “However, most deaths, especially in cohorts with low cardiovascular risk, may be non-cardiovascular and therefore not amenable to statin prevention.” This article is, I think, extremely important and well-argued, but my response to the author’s last point is “So what?”: If a ‘life-saving’ drug turns out to do no such thing, then perhaps we need to know.
References:
1. Nguyen PV. Electronic health records may threaten blinding in trials of statins. BMJ 2014;349:g5239
