The Minimum Clinically Important Difference and its Use in Medical Research

Author: Ian McCarthy

It has been called the smallest difference that clinicians and patients would care about. The concept of Minimum Clinically Important Difference (MCID) has been around for over 20 years, but the literature still can’t quite decide what to make of it.
It is an attempt to address that terrible old medical non-joke ‘the operation was a success but the patient died’ – to give clinicians a handle on what really matters - the outcome from the patient’s point of view.
For the patient, if you are ill, and you are given a treatment, at the end of it do you feel that the treatment has helped? For the clinician to work it out, he basically has to ask the patient. A treatment given to an asthma patient, for example, might increase his lung capacity by a statistically-wonderful 5 percent; but if the patient feels no better the statistic is pretty meaningless – and the treatment has failed to achieve MCID.

Conceptually, therefore, MCID represents a threshold for some clinically meaningful improvement in a patient’s health-related quality-of-life (HRQoL), as measured by what the patient reports. There are standard questionnaires, such as the Short Form 36-item questionnaire (SF-36) or the EuroQol 5D (EQ-5D), which the patient can fill in; and from these a numerical score can be derived that gives some indication of his or her quality of life. The key word here is ‘quality’. The concept of MCID avoids the use of “statistically significant” results that are qualitatively meaningless, and dovetails with the general distinction between practical relevance and statistical significance.88

Despite its intuitive appeal, the appropriate estimation and use of MCID remains unclear. At least two general estimation strategies have been espoused in the literature: 1) the anchor-based approach; and 2) the distributional approach. The anchor-based approach uses some external responses, such as whether the patient reported feeling “better” or “worse” after treatment, and calculates the MCID by comparing the HRQoL of patients in the “better” group to those of the “worse” group. Conversely, the distributional approach calculates MCID based on the within-sample change in HRQoL scores relative to some measure of variability in the distribution. The distributional approach therefore bases its calculation of MCID solely on existing HRQoL responses and does not rely on an external criterion.

Each approach carries with it some fundamental flaws. For example, in practice the external criterion used in the anchor-based approach is often some other subjective questionnaire. Researchers are essentially converting one subjective measure to match another. If this is the way to go, why not simply drop the second questionnaire and analyze just the external criterion?

While the anchor-based approach suffers from multiple measures of subjectivity, the distributional approach suffers from a different flaw. The point of the MCID is to avoid reliance on statistical results with no qualitative value, but the distributional approach places a statistical framework around the calculation of the MCID. For example, some authors estimate MCID as one standard error above the mean score. Far from an “alternative” to statistical significance, this approach amounts to simply changing the threshold for statistical significance.

Perhaps more importantly, the general concept of MCID is somewhat lacking. We’re essentially assigning an objective threshold to a subjective metric. The point of incorporating HRQoL into the evaluation of health care programs was to acknowledge the importance of a patient’s self-reported health assessment and to distinguish self-reported outcomes from objective clinical measures. Efforts to morph the prior into the latter seem misplaced.

Bookmark and Share

Comment on this article

Submit your comment
  1. Image of unique ID

Comments

Mark Perry

For continuous outcomes such as blood pressure, it is quite easy to conceptualise a minimally important difference, or, to use another term, a clinically important difference. We can perform simple studies where we look at the associations between individuals’ changes in that variable and how that change makes them feel, and we can thus derive a threshold for a level of difference that is meaningful to patients.

 

However for categorical variables, it is not as simple. For categorical variables, such as mortality, we are not dealing with effect measures that have any meaning to individual patients. For example, if a new treatment leads to 2/1000 patients dying whilst the established treatment leads to 3/1000 patients dying, there is no way that any patient, clinician or guideline development group member can tell us if the absolute difference in mortality – 1/1000 – is clinically important to them or not. This is because the effect measure relates to the population, not the individual.

 

Despite this, there is a general consensus that it is correct to think of a minimal important difference for categorical effect measures. For example, an absolute difference of 1/1000 in terms of the risk of mortality may be regarded as not clinically important, but an absolute difference of 100/1000 may be. Does this really make any sense? To the single person out of a thousand surviving because of the new treatment being used instead of the old one the difference is almost infinite; indeed, the difference between life and death. That is to say nothing of the effects on that person’s relatives, friends or employers. It is utterly meaningless, therefore, to say it is clinically unimportant. Consider for a moment that you are an expert in aircraft safety. You have invented a safety strategy that will eliminate all aircraft crashes on earth for the rest of time. And it will only cost £50 to implement. It sounds great, but then you find out that only 350 people die in aircraft crashes per year, which is only 0.000016% of all passengers. So you decide not to impose the safety strategy, even though it has negligible costs, because the proportion of lives you’d save seems so small it is almost negligible. This is the paradox of population measures, where tragic catastrophes and calamities can be almost wiped out of existence in our minds as a result of the mere fact that populations are big things. This sounds absurd, but it is precisely the same as not using a new treatment that could save one life.

 

The notion that effects on only a small proportion of the population are ‘unimportant’ probably goes back to considerations of cost. If it costs £500,000 to save one life, then this might be some justification to stick with the old drug and not use the new drug. But by stating that an absolute difference of 1/1000 is not clinically important without consideration of economic factors, which is exactly what we are accustomed to doing in GDGs, we are jumping the gun, and assuming lack of cost effectiveness before we have even calculated the basic costs.

 

What is important with categorical variables is to know whether the effects from the studies you are considering are a true reflection of the whole population. Perhaps the small absolute difference of 1% is just sampling error. In such a case, we can definitely be justified in dismissing it. Dismissing small differences because they are not real is akin to not bothering to check under our beds at night for imaginary monsters – a sensible and rational approach. If the small differences ARE likely to be real reflections of the true population effect we can then go on to consider their cost, safe in the knowledge that the new treatment really does make a REAL difference, albeit to a small number of people. It is only at that point that we can consider whether those real effects are feasible in terms of being affordable. Note the use of the term ‘feasible’ rather than ‘important’.

 

reply to this comment

C K Tam

MCID is a qualitative measure of patient's subjective feelings towards clinical intervention and is difficult if not impossible to measure without due precautions. It is a psychosocial entity.

reply to this comment

Adrian Lambourne

I must disagree with Ian's definition of an MCID.

Firstly how the patient feels about their outcome is not the objective: it is, as its name says, the "Clinically Important" difference. While a diabetic may not "feel" any better from having his HBA1C lowered by 0.5% in absolute terms, if it can be shown that this reduction will result in fewer complications or his overall health and wellbeing deteriorating less rapidly, then it can be argued that this reduction is "Clinically important".

Secondly, we need to use a difference that can be tested as to what minimal level does generate some improvement - on average - across the patients to whom the new treatment would be applied. While this may be subjective  - as different experts may have different views based on thier own experience and knowledge - it is far less subjective than what patients think about their own outcome.

To my mind, the point of haivng an MCID is to be able to say whether a change really is clinically meaningful i.e. it does improve the average patient's outcome, rather than just whether the change is different from zero which is all that could be implied from a rejection of a simple null hypotheses.

reply to this comment

Dave Holtz

In clinical practice, patients voice widely different goals for therapy (some clinically unobtainable). It seems that a mathmatical MCID will always breakdown due to an inability to agree on the clinical goal against which we measure the incrimental improvement.

reply to this comment

Tom King

I have heard of an odds ratio of only 2 being dismissed as not clinically significant. Is this a common threshold?

reply to this comment

Skip to Main Site Navigation / Login

Site Search Form

Site Search