cerebral palsy

Why bother with gait classification?

There is a sector of our community that sees classification as the holy grail of gait analysis. If only we could divide our patients into neat little categories then our problems will be solved. But why do we want to classify patients?

The most obvious reason for categorising patients into groups would be if genuinely different groups existed. But do they? In an earlier blog I’ve written about the GMFCS and pointed out that there is absolutely no reason to believe that children with cerebral palsy fall into distinct categories of severity – they almost certainly lie on a continuous spectrum.  I suspect that the same is true of gait patterns. Indeed in work that Fiona Dobson completed for her PhD thesis but never published we used a technique called multi-dimensional scaling to look for evidence of clustering of gait patterns in children with hemiplegic cerebral palsy. We couldn’t find any. Whichever way we looked at the data it looked like a random scatterplot.

So if there is so little evidence that gait patterns fall into distinct categories why are we so obsessed by it?

I suspect that classification systems are assumed to help clinicians – particularly less specialist ones than are employed in our gait analysis services. If we can simplify understanding of gait to a small number of discrete categories then we’ll be able to teach people to recognise these. If we can describe management plans for the different categories then maybe we can even teach people to treat the patients.

There are problems though.  The first one is that the classification systems don’t seem to work – clinicians can’t agree on what category a gait pattern should be placed in. Fiona Dobson, in one of the few studies to assess reliability of a classification in a centre outside that in which it was developed (Dobson et al,2007), looked at the agreement of 16 clinicians in rating gait patterns of 34 children with hemiplegic cerebral palsy . On the basis of video data, only one pattern was put in the same category by all clinicians and in over a third there was less than 50% agreement. An important feature of this study was that the 16 clinicians were all specialists in gait analysis of children with cerebral palsy. What would the results have been like if the non-specialist clinicians, who presumably have most to gain from classification systems, had been recruited to the study? Results were a little better when gait analysis data was used as a basis for classification, but then the clinicians for whom classification is likely to be most useful don’t generally have access to gait analysis.

Another issue is that gait patterns are almost certainly a combination of characteristics of the child and of how they have been managed previously. The Winters, Gage and Hicks (1987) classification of hemiplegia intentionally excluded children who had had previous surgery but the classification of diplegia by Rodda et al. (2004) included children with prior calf surgery (because there were so many of them). We never quantified it but it appeared to us that fewer and fewer children in Melbourne fell into the Rodda “crouch” category over time which we attributed to more conservative early surgical management and the wider use of Botulinum toxin. If this is the case then classifications schemes will have to be sufficiently generalizable to cover the effects of different management practices or maybe we need a number of different classification schemes depending on different management practices in different parts of the world.

Developing and validating classification schemes is a considerable undertaking. Earlier schemes were generally based on the enlightened but essentially subjective opinions of leading clinicians. In an era of evidence-based practice this is no longer satisfactory and widespread consultative processes (such as a Delphi process) would now be needed to convince the clinical community of any new scheme.  Similarly were existing schemes have been validated this has tended to be on the basis of repeatability studies based within the teams who developed the scales and typically using gait analysis techniques that may not be available to those who are most likely to benefit from the classification. Convincing validation of new schemes will need to be more comprehensive. Such an investment of time and effort would be perfectly justified if we are certain that the schemes will be useful – but we do need to be certain.

Validation is made considerably more complex by the fact that the classification systems do not fit the underlying reality. How can you interpret reliability studies designed to test the competence of clinicians to ascribe gait patterns to categories when there is genuine ambiguity in whether any individual child actually fits a particular category? A similar issue arises with the usefulness of the scheme for less experienced clinicians. Many children will either be on the borderline of categories or may not fit cleanly into categories at all. Explaining how to deal with such exceptions to generalist staff may lead to more confusion not less.

So is there an alternative? Well I’d like to float the potential of what I call profiling. Rather than select a number of discrete groups into which gait patterns can be assigned we define aspects of the gait pattern that vary across the patient population. We can then score or grade the different aspects separately. This allows us to accept that there is a continuous distribution of gait patterns and that also that gait might vary independently across the different aspects rather than suggesting that there is necessarily grouping of the aspects as is required by classification schemes.

We presented a paper proposing this for the gait of children with hemiplegic cerebral palsy at the Sydney CP meeting in 2008 (Tirosh, Dobson et al.) but never took the ideas any further. A factor analysis based on movement analysis profile scores suggested that 85% of the variability could be explained by 5 factors which reflected:

  • distal sagittal plane features (knee flexion and ankle plantarflexion combined),
  • leg length discrepancy (anatomical and functional),
  • hip rotation,
  • hip flexion
  • foot alignment.

These factors all make clinical sense as does the idea that hemiplegic gait should be categorised by assessing how far each of these different aspects contributes to the gait pattern of an individual child rather than trying to assign the gait pattern to one of a small number of categories. Perhaps more importantly it appears to me (and I admit this is a subjective opinion) that the underlying model of gait patterns varying along a continuum in a number of different dimensions matches reality more closer than a model assuming that gait patterns fall into distinct and recognisable categories.


Dobson, F., Morris, M. E., Baker, R., & Graham, H. K. (2007). Gait classification in children with cerebral palsy: a systematic review. Gait and Posture, 25(1), 140-152.

Rodda, J. M., Graham, H. K., Carson, L., Galea, M. P., & Wolfe, R. (2004). Sagittal gait patterns in spastic diplegia. Journal of Bone and Joint Surgery. British Volume, 86(2), 251-258.

Winters, T. F., Jr., Gage, J. R., & Hicks, R. (1987). Gait patterns in spastic hemiplegia in children and young adults. Journal of bone and Joint Surgery – American, 69(3), 437-441.


The importance of objective outcome measurements

This post was stimulated by a presentation given to the GCMAS by Nancy Lennon of the A.I. Du Pont Hospital in Delaware. She presented data on measured activity levels over the year following major orthopaedic surgery for children with cerebral palsy. Her data came from a case study and showed how the patient’s activity levels fell markedly at 3 months after surgery before picking up through the rest of the post-operative year.

The graph below is an update of one I prepared for a lecture on Outcome Measures at the Melbourne Gait Courses last year. It puts together data from a range of sources to suggest a time history for the Gross Motor Function Measure (GMFM, Russell et al., 1989)  for a child with cerebral palsy who has single event multi-level surgery (SEMLS) at the age of 10. The data points are taken from Thomason et al (2013) and represent average GMFM scores for a cohort of children at baseline (blue) and 12, 24 and 60 months (green) following SEMLS.  Before surgery I’ve assumed that the data follows the latest GMFM curves (Hanna et al., 2009) to arrive at the baseline value.


The red point is invented. It is an estimate of the GMFM a child might record if assessed on coming round after surgery in a hospital bed with below knee casts. The actual value is not particularly important but seems reasonable when I glance through the GMFM manual. I’ve then extrapolated the curve from this point through 12, 24 and 60 month data points.  Having seen the videos of kids coming back for 3, 6 and 9 month follow-up after such surgery whilst in Melbourne I don’t think the time course over the first year is too far away from reality. I’ve finished off the curve assuming that it follows the known GMFM data (Hana et al., 2009).

First thing to point out is that average GMFM score at one year is almost exactly the same as at baseline and the maximum GMFM is recorded at two years following surgery suggesting that the one year follow-up may be a little early to assess outcomes.

The point I really want to make though is that if you look at this graph the biggest feature is not the improvement from pre-op to 12 or 24 month status. It is the drop in function immediately after surgery and the improvement back to baseline at 12 months. This has the potential to impact on patient, family and clinical perceptions of outcomes. If the dominant memory of the surgery is of the condition the child was in immediately afterwards, then the perception may well be of the change following surgery as being represented by the difference between the green points and the red point which might lead to a much more positive view of outcomes than a more scientific comparison with the blue point. Particular caution may have to be exercised in interpreting the results of subjective or semi-subjective assessments such as heath related quality of life questionnaires or informal assessment of outcomes.

Final point is that there are a multitude of reasons for performing such surgery and assessing outcomes on the basis of any one measure in isolation is inappropriate. I’ve plotted this data to make a particular point about the time course of recovery not to make any general conclusions about the effectiveness of the surgery. Gait Profile Scores (Baker et al., 2009) reflecting the quality of the gait pattern improved by over 30% in the same cohort for example.


Russell, D. J., Rosenbaum, P. L., Cadman, D. T., Gowland, C., Hardy, S., & Jarvis, S. (1989). The Gross Motor Function Measure: a means to evaluate the effects of physical therapy. Developmental Medicine and Child Neurology, 31(3), 341-352.

Thomason, P., Selber, P., & Graham, H. K. (2013). Single Event Multilevel Surgery in children with bilateral spastic cerebral palsy: a 5 year prospective cohort study. Gait Posture, 37(1), 23-28.

Hanna, S. E., Rosenbaum, P. L., Bartlett, D. J., Palisano, R. J., Walter, S. D., Avery, L., & Russell, D. J. (2009). Stability and decline in gross motor function among children and youth with cerebral palsy aged 2 to 21 years. Dev Med Child Neurol, 51(4), 295-302.

Baker, R., McGinley, J. L., Schwartz, M. H., Beynon, S., Rozumalski, A., Graham, H. K., & Tirosh, O. (2009). The gait profile score and movement analysis profile. Gait Posture, 30(3), 265-269.

Averaging up the profits

Here are two graphs. The first from very early in my career shows a parameter we called the “dynamic component” of gastrocnemius length. It plots the improvement in this after injection of botulinum toxin in children with cerebral palsy against the baseline score (Eames et al., 1999). I remember when Niall first showed me the graph. We’d captured a  whole load of data on these kids and were wondering what to plot to make sense of it. This was the first suggestion and I can still remember Niall’s excitement when it came up with such a strong relationship.


At the other end of my career here’s another graph from a paper that has only just been published electronically in Gait and Posture (Rutz et al. 2013). Here is the improvement in Gait Profile Score (GPS, Baker et al., 2009) for children with cerebral palsy plotted against baseline score (with GMFCS II and III children plotted separately). Again there is a strong correlation. (There are some statistical issues in plotting data this way which might lead to exaggeration of the correlation when measurement error is substantial but I’ve gone to some lengths in the recent paper to show that this is unlikely.)


When you think about it though the relationship is actually quite unremarkable. What both studies are showing is that kids with the most severe problems to start with are the most likely to show improvements. To a certain extent this is common sense – if two kids both improve by 30% then the child with the biggest problem to start with will show the biggest change in absolute units.

What interests me though is that if we only look at the average changes in each group we will reach the conclusion that the group as a whole have improved. If we are not careful we might conclude that all the group has improved. Thissimply isn’t the case. The full truth is that the kids who have the biggest problems have improved a lot those with milder problems haven’t improved very much (in absolute terms).

The Botulinum toxin study became the basis for an industry sponsored randomised controlled trial (Baker et al. , 2002). In that trial although we included baseline readings as a covariate in the statistical analysis but we only ever reported group results. That is still probably the most rigorous trials of lower limb injection of Botulinum Toxin in the literature. The message that almost everyone has taken out of that study  from the data we presented is that kids with spastic diplegia will benefit form Botulinum toxin. Had we presented the data more carefully the conclusion should have been that the more severely affected kids will benefit from Botulinum Toxin big time, but that  the milder kids may not benefit at all.

As it stands the paper is really convenient for the company because it suggests that a wider group of kids will benefit from an expensive drug than is actually the case. Given that bigger responses to treatment in more severely affected people is likely in almost all conditions that affect people across a range of severity I suspect that a similar phenomena spread across almost all of . I wonder how much profit the drug companies are making as a consequence?

Leave a comment or double click “n comments” link at top of post to view discussion.

Baker, R., Jasinski, M., Maciag-Tymecka, I., Michalowska-Mrozek, J., Bonikowski, M., Carr, L., . . . Cosgrove, A. (2002). Botulinum toxin treatment of spasticity in diplegic cerebral palsy: a randomized, double-blind, placebo-controlled, dose-ranging study. Dev Med Child Neurol, 44(10), 666-675.

Baker, R., McGinley, J. L., Schwartz, M. H., Beynon, S., Rozumalski, A., Graham, H. K., & Tirosh, O. (2009). The gait profile score and movement analysis profile. Gait Posture, 30(3), 265-269.

Eames, N. W. A., Baker, R., Hill, N., Graham, K., Taylor, T., & Cosgrove, A. (1999). The effect of botulinum toxin A on gastrocnemius length: magnitude and duration of response. Dev Med Child Neurol, 41(4), 226-232.

Rutz, E., Donath, S., Tirosh, O., Graham, H.K., Baker, R. (2003). Explaining the variability improvements in gait quality as a result of single event multi-level surgery in cerebral palsy. Gait Posture, published on-line http://dx.doi.org/10.1016/j.gaitpost.2013.01.014

Little boxes

GMFCS  is a categorical scale (Palisano et al., 1997, 2008). Children and adolescents are allocated to one group or another. There’s absolutely no evidence, of course, that there is anything in the condition (or group of conditions) we call cerebral palsy to suggest that children’s gross motor abilities are distributed in such neat little packages. The spectrum of cerebral palsy is almost certainly a continuum and the gross motor abilities are almost certainly distributed along a continuum as well. The categories of the GMFCS do not represent actual discrete groups of children with gross motor abilities that are qualitatively different from those in the other groups. Rather, they are an administrative convenience. Medicine, and life in general, is littered with examples of continuously distributed parameters divided into essentially arbitrary categories simply because this is the easiest thing to do.

Bill Reid's depiction of GMFCS II (ROyal Children's Hospital, Melbourne)

Bill Reid’s depiction of GMFCS II (Royal Children’s Hospital, Melbourne)

Remembering this is important when we engage in discussion about how clearly children can be allocated to the different categories and also how stable that categorisation is over time. If the classification is actually a convenient division of a continuous spectrum then there will be a number of children who fall very close to the border line between these  groups. Some of them will lie sufficiently close to the boundary that they can’t be reliably categorised. One day they will illustrate the characteristics of one group and another day the characteristics of another. Alternatively one assessor will make a subjective decision to put the marginal patient in one group whereas another assessor will put them in the other group. Neither is wrong – it is just a consequence of taking people on a continuum and trying to put them in boxes. Just how many children inhabit this marginal space is unclear but in assessing the reliability of the classification system we should be anticipating at least some borderline children for whom it is not possible to allocate a definitive GMFCS level. I may not have been reading carefully enough but I’ve never seen any discussion of this in the relevant literature.

This also impacts on studies of stability of the GMFCS over time. We should expect that a fairly modest improvement in gross motor function should take a child who has been graded at the top end of one category at one time to lead them to be graded at the lower end of the next category up on a later occasion. Equally we should expect some children at the lower range of ability for any given range to drop a level if they deteriorate quite mildy. Some transition between neighbouring groups is thus an inevitable consequence of how the groups are defined and should be expected.


Palisano, R., Rosenbaum, P., Walter, S., Russell, D., Wood, E., & Galuppi, B. (1997). Development and reliability of a system to classify gross motor function in children with cerebral palsy. Dev Med Child Neurol, 39(4), 214-223.

Palisano, R. J., Rosenbaum, P., Bartlett, D., & Livingston, M. H. (2008). Content validity of the expanded and revised Gross Motor Function Classification System. Dev Med Child Neurol, 50(10), 744-750.

GMFCS based research -are we asking the right questions?

It’s fifteen years since the publication of the first paper on the GMFCS (Palisano et al., 1997). Since then it has become ubiquitous in the field of cerebral palsy. More and more measures are being found that correlate with it. It just seems like magic. But the more things we discover that show this correlation the more I wonder whether this really is magic. Have we missed something? Are we asking the right questions?

Cerebral palsy is an extremely heterogeneous condition affecting some kids extremely severely and others very mildly. In terms of gross motor function the range is from a child with  hemiplegia and a mild foot drop right through to those with severe total body involvement who are essentially immobile. GMFCS allows us to group children (and now adolescents, Palisano et al., 2008) in terms of that function. In other words, GMFCS is essentially a classification of the severity of CP as indicated by gross motor function.

When we think about it most of the other indices, scores and scales we look at can be considered to be measures of the severity of CP as indicated by other aspects of the condition. When we get a correlation between GMFCS and another measure we are thus really saying there is a correlation of the severity of CP indicated on the basis of Gross Motor Function and severity of CP as indicated by hip dysplasia (Robin et al., 2008, see Figure below) , or gait quality (Baker et al., 2009) or physical activity (Bjornson et al., 2007). We shouldn’t really be surprised that there is a correlation – in fact the thing that should really surprise us is if there isn’t.


Correlation of hip dysplasia (migration percentage) with GMFCS (Robin et al., 2008)

A more nuanced approach to research in CP might be to anticipate the underlying correlation between indicators of severity of CP and accept it as unremarkable. Measures that don’t correlate are actually more remarkable and further investigation of these, when they are identified, might be more productive than investigation of those that do. Detailed consideration of individual children that buck the trends may also give important clinical insights.


Baker, R., McGinley, J. L., Schwartz, M. H., Beynon, S., Rozumalski, A., Graham, H. K., & Tirosh, O. (2009). The gait profile score and movement analysis profile. Gait Posture, 30(3), 265-269.
Bjornson, K. F., Belza, B., Kartin, D., Logsdon, R., & McLaughlin, J. F. (2007). Ambulatory physical activity performance in youth with cerebral palsy and youth who are developing typically. Phys Ther, 87(3), 248-257.
Palisano, R., Rosenbaum, P., Walter, S., Russell, D., Wood, E., & Galuppi, B. (1997). Development and reliability of a system to classify gross motor function in children with cerebral palsy. Dev Med Child Neurol, 39(4), 214-223.
Palisano, R. J., Rosenbaum, P., Bartlett, D., & Livingston, M. H. (2008). Content validity of the expanded and revised Gross Motor Function Classification System. Dev Med Child Neurol, 50(10), 744-750.