I completely agree with david. There are some works to determine the clinical significance of some functional tests but it will depend on your initial design.

The following link is from an study of clinical significance in Fugl Meyer scale. You can find similar papers for other tests of your interest that could help you...

http://www.ncbi.nlm.nih.gov/pubmed/22282773

Hi there

This is a question that comes up often with the clinicians that I assist with research. Too often people focus purely on the statistical significance and lose sight of the clinical relevance. To my mind they are both important- how often do we see statistically significant findings with an effect size so small as to be clinically worthless?

In an ideal world clinical relevance should be built in to a study from the start. Your minimally clinically important difference should be what you aim to detect, which will then influence your sample size. That difference may be grounded in the literature, and each of those tools should have something that guides that decision, but more often than not it is a clinical judgement. Statistical significance is at the end of the process and helps you understand whether that difference was found or not.

In practical terms clinical relevance is best judged by clinicians, which is why we should be empowering and enabling clinicians to be involved in the research process.

Hi David,

Thanks for the input. Imagine a study with 2 groups, 15 participants each and assuming that at the end of the study , both the groups were equally effective statistically and if post-hoc shows one group better than the other, how can I identify the clinical effectiveness?

I completely agree with david. There are some works to determine the clinical significance of some functional tests but it will depend on your initial design.

The following link is from an study of clinical significance in Fugl Meyer scale. You can find similar papers for other tests of your interest that could help you...

http://www.ncbi.nlm.nih.gov/pubmed/22282773

Hi Shankar , an other way to calculate MCID or MDC (minimal detectible change) comes from the formula MCID=1.96 * square root 2 * SEM, where SEM= standard error of measurement. This is the way we calculate it in our papers (http://www.ncbi.nlm.nih.gov/pubmed/23231785) (http://www.ncbi.nlm.nih.gov/pubmed/24480626)

Hello All

One of the reasons I think there is so much difficulty establishing milestone minimal clinical important difference scores is we have to decide to whom the difference is important (e.g. the therapist? the patient? a 3rd party payer?). Each might have a different perspective on whether the magnitude of change observed is "worth" the cost/burden/time/effort required to achieve that change.

An editorial is the Australian Journal of Physiotherapy reviews this nicely (attached).

Also, I highly recommend a little (~150pgs) textbook that discusses these issues clearly from a literature consumer's perspective (rather than the literature producer's): Riddle & Stratford. Is this change real? Interpreting patient outcomes in physical Therapy. FA Davis Company. 2013. isbn: 978-0-8036-2957-8. It's the only stats book I've read cover to cover and has helped me in developing an understanding of what various research designs in this area can tell us/not tell us.

Dear Vaseleios,

Thanks for the formula. Unfortunately, I do not have full access to those articles. I have a query though; when we use that formula what is the cutoff value to say the results are clinically significant?

Dear Vaseleios,

Also can you comment on which SEM to use?

Dear Shankar,

Ries AL (2005) calculated in his study MCID for for dyspnea measurement tools like VAS, Borg and SOBQ . In this study an ES (effect size) approach, which is a kind of distribution based approach, was used to determine the MCID for dyspnea measurement tools.

SEM and SRD are another type of distribution based approach measurement to determine the cutoff value.

If your study is an interventional study, and you are interested to know whether the intervention you provide is clinically significant and not only statistically significant, you should calculate the reproducibility of your measurement methods (VAS, NRS and etc.).In order to measure the reproducibility (SEM+SRD) - you should take a test-retest measurements (usually 1 week apart) of the variables you are interested in. After taking test-retest - use SEM formula: SEM=SD*√(1-rxx). To determine the cutoff value - calculate SRD (smallest real difference): SRD=SEM*2.77. After running the intervention - check for each patient whether the absolute change in VAS/NRS is bigger than SRD. If the answer is yes - you can assume that the change was not due to error measurements but due to the intervention.

Irena

You may find useful information by connecting to Bruno Falissard (he has a Research Gate account). He is a renowned clinician AND statistician, and recently posted publications on this topic, for example : "The minimal clinically important difference determined using item response theory models: an attempt to solve the issue of the association with baseline score".

Dear colleagues,

To my understanding – the MCID and SEM are similar in terms of statistical estimation but have different applications. As described by Irena Paiuk the SEM and SRD (absolute reliability measures) will apply if we are interested in looking at the psychometrics of a measure – where the SEM is the ‘estimated’ random error score that is inherent with the measurement procedure or something like that. If somebody uses a particular measure in a clinical set up, for which SEM has been determined, as a measure to monitor a client over time – any change over and above the SEM can be deemed as a change due to something other than the measurement error. But that is not the end of the story – we then have the SRD which is more or less a 95% confidence interval for the SEM, again an ‘estimated’ measure. I say ‘estimated’ because the estimations can be enhanced or worsened by many factors such as number of participants in the study, type of participants (clinical/non-clinical/sub-clinical), level of control exercised with the measurement etc.

On the other hand MCID is a measure which is interpreted from a clinical intervention perspective. The estimated MCID is in itself a cut-off value for the clinicians. If I intend to use an intervention reported by somebody in a ‘particular’ clinical scenario, any change over and above the MCID of ‘that study’ can be taken as the effect of the intervention.

However the application of SEM/SRD and MCID cannot be mutually exclusive in all situations. For example – the SRD of timed up and go test reported in persons with amputation is 3.6secs (Resnik.L et al, 2011) and if I use it to study an intervention in amputees and find the MCID as 3secs or 4secs – I will not be sure if the difference obtained here is either due to the intervention or due to the measurement error in TUG. Perhaps, that’s what the theme emerging from the discussion above.

Hope this makes sense and I stand corrected if I haven’t got it right!

Thank you all for insight provided on MCID.

Just for clarification the formula: z-value*square root of 2*SEM is the formula for the minimal detectable difference (MDD) not the MCID!!! The MDD it tells you the smallest amount of change an instrument can accurately measure. However, if this is as well a clinical significant change is questionable. Unfortunately, I think for many measures and scales nobody really knows the MCID. To sum up, I would say the MDD is definitely a good starting point, but has to be interpreted with caution.

http://www.rehabmeasures.org/Lists/RehabMeasures/PrintView.aspx?ID=1114

I joined this discussion with many years of delay, but I have a very simple question: Why is the root of 2 used in the MDD formula? Is it not enough z-value * SEM?

thanks