I have no specific variables in mind, I am just wondering if anyone has a sense of what constitutes a clinically important change in a gait variable such as peak pronation or peak knee abduction.
ideally you would have a fundamental value such as O2 consumption or velocity showing improvement over baseline. AS for kinematics, improved measures this includesduration as well as progression, during single limb stance on the affected side.
The problem is a lack of large, longitudinal, intervention gait studies that collect PROMS in order to calculate MID. Also gait parameters are highly variable. I think there is a general trend towards understanding baseline biomechancial risk factors and responders to treatments.
This study (Barton 2011) does not report longitudinal changes in biomech variables but does use them as a clinical predictor:
C J Barton, H B Menz, P Levinger, K E Webster, K M Crossley Greater peak rearfoot eversion predicts foot orthoses efficacy in individuals with patellofemoral pain syndrome. Br J Sports Med 2011;45:697–701
Unfortunately, I don't know that there is a strong consensus on what constitutes "clinically important." Sample sizes are often calculated on either what an individual researcher considers a meaningful difference to be or by looking to match or exceed the differences seen in prior investigations on the same topic.
Pablo, that's a good point, but I have no particular pathology in mind. My question is indeed vague, and that is because I am asking it from a philosophical point-of-view. If someone (healthy, pathological, whatever) changes their gait by X degress, then something meaningful has happened. I am interested in everyone's thoughts on what X could or should be.
Jill, thanks for the response. I totally agree that there is a lack of study data which could address my questions. Thanks for the references; I think the use of a gait variable as a predictor could provide some insight into meaningful differences. Much appreciated.
The concept of "Clinically Important" is going to be very subjective and hard to nail down, I would think. I do know of a good recent reference on minimal *Detectable* change differences, which should provide a baseline:
"Reliability and minimal detectible change values for gait kinematics and kinetics in healthy adults"
Jason M. Wilken, Kelly M. Rodriguez, Melissa Brawner, Benjamin J. Darter
That is very helpful, Jonathan. I had a tough time searching for references to address my question, and I never came across this article. Great stuff, thanks.
that is a very interesting topic and unfortunately I can probably not give you a lot of references in the human gait world, however I am facing this question often when staring at the 'objective movement symmetry parameters' that we calculate from our clinical lameness cases (horses). Similar to humans we also have the problem that there are huge variations between individuals and compared to the human gait analysis world we also still do not have comparable numbers of 'patients'. We are talking if we are lucky a couple of hundred. But what I find when I look at the data of a horse that shows a certain gait abnormality (i.e. asymmetry), I will not overly emphasize a finding that is outside our current normal range (which by the way are based on a few tens of 'normal' horses) if it is only apparent in one particular movement condition. Apart from having the horses exercise on the straight we also make them go around in circles (left and right) and also on different surfaces and only if I see a consistent change in more then one exercise condition, then I am more convinced and will 'put my head on the block' and tell our clinicians that something is wrong with a particular limb.
In addition we also use 'joint or nerve blocks' (administering local analgesia) to eliminate the pain from certain structures to further localise the lesion and then this minimal clinically important difference becomes even more important. Yet again, I will then want to see a consistent change from the baseline condition(s) in more than one exercise condition, so I guess 'long story short' for me it is a bit less about the actual amount of change (although that of course also plays a role) but also the consistency of a change or a gait deficit.
This is all 'relatively easy' since we only record very basic gait parameters (only 5 landmarks, only vertical movement, currently, but working on more as we speak) and we use IMUs so we can move between different locations easily and can also record quite a few strides in relatively short amount of time (so we can also take SD into account in our decision).
Hope this helps somewhat, although I am well aware that this is probably a lot different in human gait analysis.
Thanks for your comments. Although your methods aren't directly applicable to what I am trying to do, they do raise an interesting question for human gait. Would it be more effective to examine a general movement pattern using IMUs instead of using markers to obtain specific joint angles? I like your approach to the detection of abnormal gait, and perhaps there is potential to apply similar methods in humans. For the moment, though, I am still stuck with cameras, markers and joint angles!
I recently found a paper related to this issue, though it used function indices and scales rather than actual gait data: Landorf & Radford 'Minimally important difference: Values for the Foot Health Status Questionnaire, Foot Function Index and Visual Analogue Scale' The Foot 18 (2008) 15-19.
Apart from that, I would see clinical importance predominantly in 2 parameters that are fairly straight forward: gait velocity and single limb support. I cannot give you hard data on that but if the freely-chosen gait velocity is below 1.2 to 1.5 m/s (my margin would be
The paper by Wilken looked at healthy controls, if you want a sense of MDC for a pathological group McDermott et al estimated it for numerous gait variables in the Cervical Spondylotic Population. The SEM was calculated for each variable so from this you should be able to estimate the MDC. Also there is a systematic review by McGinley et al on reliability of 3DGA which might be of interest to you also.
It's my favourite reference to Effect Size (ES) issues
Discussion about parctical / clinical significance should be based on statistical evaluation of the ES and more importantly on statistical evaluation of ES accuracy.
A very recent article observed that even in high-standing revues, ES is not adequately reported:
High Impact = High Statistical Standards? Not Necessarily So
Okay, for me there are now 2 interesting topics here: 1) speed (or velocity), and 2) statistical versus biological/clinical relevance.
1) I totally agree that speed might be an indicator of 'disease', for example we see this in neurological cases (again in particular if you attempt to change the exercise condition and eg blindfold them or elevate their head etc) ... neurological patients (horses!!!!) will walk slower. We have recently discussed this and in particular have then asked the question if we for example calculate more 'complicated' parameters like inter-limb coordination pattern or timing parameters like stance times etc, and see differences between groups: how much of this is down to the simple fact that the diseased group is slower? It is reasonably tricky to get the healthy group to slow down for a fair comparison. So now my question: do you use any sort of 'speed normalization' and/or 'size normalization' when you compare gait parameters?
2) if you find a very small difference eg before/after treatment (in our case a few mm of displacement) and you have the luxury to collect many strides so can reasonably well estimate variation, how do you then decide whether something that is statistically different is biologically significant
In many cases the reason for altered movement is pain. One develops certain strategies to avoid pain, which leads to abnormal motion patterns. This in turn changes the load distribution in joints and in the long term possibly to accelerated joint degeneration. So the clinical relevant parameters in such cases are joint kinematics, which can be detected and evaluated by motion analysis.
Thilo, question #2 is very interesting indeed. How much change in a joint angle does it take to qualify as "biologically significant"? I would love to know that answer, but it is very tricky to figure out. Most likely the only way to really know the biological effects of such a change would be a prospective study looking at either performance or injury outcomes.
Thanks again for the contributions. From everything that I have read through so far, it seems as though a good "rule-of-thumb" for clinical change in joint angles for an individual is anything greater than 5 degrees. For some measurements this can be less.
The 5 degree rule seems to be what is measured as the MDC based on errors in gait analysis. However, from a philosophical standpoint I am also interested in knowing opinions on Thilo's question. What kind of change in a joint angle might be "biologically significant"? Or in other words, could a change of less than 5 degrees be important even though we perhaps can't reliably detect it?
Thanks for the postings. I am unclear as to how you can arrive at practical significance by evaluating effect size. As an example, I performed biomechanical analysis of golf swings on a group of players in two conditions. I found differences that were, from an effect size standpoint, rather small. Despite this, I could show, mechanically, that the differences represented a large practical effect for the player in terms of the distance the golf ball would travel. I could very easily argue that, on average, players performed better in one condition than the other, despite the fact that my effect size could be called 'trivial'.
From my experience, I'm not sure I would be comfortable saying that something is "important" or "unimportant" based only on effect size.
I was thinking about this very interesting discussion all day and thought about this paper:
Turner D, Schunemann HJ, Griffith LE, Beaton DE, Griffiths AM,
Critch JN, et al. The minimal detectable change cannot reliably
replace the minimal important difference. J Clin Epidemiol
2010;63(1):28e36.
I think this is the "nub" of the discussion. Are you interested in measuring real change in biomechanical variables, or whether there is clinical/physiological benefit. Both valid. I am both a clinician and a scientist. I struggle with this.
From the example above, it seems to me that the study outcomes or end-points require some feasibility testing. What is most important how far the ball goes? then power the study for this?
In terms of biophysical vs biochemical response. That's a much harder question to answer. I would look towards finite element modelling, or include surrogate measures of physiology with your biomechanical variables. Implanted train gauges may be?? I read an in-vivo study the other day, would never get ethics in the UK.
Great comment, Jill. Your point about being clear regarding desired outcome from a study is absolutely right. From my end, I don't exactly have a specific aim other than to improve the quality of research data that are collected by ourselves and our partners. To this end, I am working on some mathematical tools that *could* be helpful to researchers who want to improve their ability to detect small biomechanical changes. Hence, I am asking what kind of changes everyone thinks are "important". There is no real answer, but in this case I think there is value in getting opinions from researchers involved in gait analysis.
glad I am not the only one who struggles with the differences between 'research and clinical decision making' (I am a computer scientist by training but now working in gait analysis and one of my main remits is to assist clinical decision making in lameness exams in horses through gait analysis). Love the article about minimal detectable change (MDC) and minimal important difference (MID). Just a shame that we don't have enough data (yet) for this for my application but at least there is some guidance on how to deal with some of the issues. So basically until we have conducted a similar study to determine the MID for all the various scenarios we can only go with estimates. Will have to think hard how to implement this into the current way I am presenting the data. So far I give mean values (or actually medians since often not normally distributed) and also give an idea of variation (SD or interquartile range) for the patient before and after intervention to give a little more guidance on how relevant an observed 'shift' in the mean (median) is. But I guess this really only approximates a t-test in a way since you are looking at differences between means and the variation around it. Maybe effect size would be a good additional measure to report but ultimately it will have to relate to the MID ... long way to go ...
does anybody have any comments on what to do about 'speed'? In our scenario (you can't just ask a horse to do a certain speed unless it is on treadmill) speed varies quite a bit in particular also before and after intervention when eg pain is reduced. So obviously this influences all sorts of gait variables and not necessarily each individual patient 'behaves' like all the others as a reaction to changes in speed (probably also influenced by the type of the lesion etc) so would people nevertheless try and correct for speed effects using the effects observed in the general population or rather not touch this?
"Thanks for the postings. I am unclear as to how you can arrive at practical significance by evaluating effect size. "
Estimating practical significance always has a substantial degree of subjectivity. But you must base your subjective judgment on objective evidences. Effect Size (ES) estimation (and the associated precision, expressed as CI's) is the only way to allow other investigators to make their own opinion about your results.
"Interpretation is essential if researchers are to extract meaning from their results. However, the interpretation of effect sizes is a subjective process. What is an important and meaningful effect to you may not be so important to someone else. Many researchers trained in the pseudo-objectivity of statistical significance testing are uncomfortable making these sorts of value judgments. Consequently research results often go uninterpreted [...]"
With regards to speed, we constrain our human patients to a given speed on a treadmill. From a theoretical perspective there are good arguments for allowing "self-selected" running speeds, or for having all patients run at the same speed. Fortunately for humans, changes in running speed actually don't seem to affect joint kinematics THAT much. By that, I mean certain joint angles change systematically with speed (and therefore could be corrected for speed), but most are only marginally affected. However, as we have discussed, it is difficult to say what is an important change and what is not.
For injured patients the question can be difficult. Some researchers have successfully used self-selected gait speed as a measure of disability with regards to injury or disease. So, it might be difficult to prescribe a speed for a standardized comparison. Our approach to injury uses norms collected on hundreds of patients to try to evaluate the biomechanical pattern of a new patient in terms of a normative cohort. This might be a feasible approach if you have a substantial amount of data from horses.
I appreciate what the author is trying to say. My point is that using effect size to estimate importance is perhaps just as flawed as using p-values. I think that ultimately, the only way to fully understand the effect of a change to a system is to prospectively examine the effect of that change on an outcome that is clearly understood.
Going back to the golf example, I understand what it means to drive the ball further and how important that is. I don't really understand what changes in swing kinematics mean, or how important they are. However, if I can relate those swing changes to ball flight, then I can assign importance to the swing kinematics themselves. The problem with this approach is that sometimes a clear outcome is very difficult to measure. This is certainly the case with injuries where expensive long-term studies are required.
I would also like to draw the attention of my colleagues to the fact that some times biomechanical variables during gait may not change significantly, while clinical evaluations and subjective reports by the pationts demonstrate improvement.
that's interesting ... could you please provide more details about the references. we definitely see influence on variability of gait parameters in neurological patients and I think I mentioned before that the consistency of gait parameters is also affected by orthopaedic conditions (lameness) in horses ... work by Peham and coworkers ... need to dig out the references .... don't have them in my head ...
I would guess that a gait deviation becomes significant when it results in a significant increase in the amount of energy required to achieve mobility. This will have large individual variation.
"that's interesting ... could you please provide more details about the references. we definitely see influence on variability of gait parameters in neurological patients"
The work of Hausdorff is worth to mention, but this is rather and "old-school" approach, because it is based on temporal variability only (stride-to-stride fluctuation of step time). In addition, it seems that the paradigm, which makes the link between "fractal dynamics" and disease is not quite generalizable.
By chance, a participant in the thread is a great specialist of non-linear gait dynamics: Pr Jonhatan Dingwell. I recommand you read his publications (he has a page on ResearchGate). In particular, this one was a quite outstanding study:
This is on of our papers on this topic. The reliability of gait variability (temporal and spatial) seems not very good in seniors during treadmill walking (large minimal detectable changes). Best wishes