Journals such as DMCN publish increasing numbers of papers reporting gait analysis data as a primary outcome measure in intervention studies for children with cerebral palsy (CP). Despite increasing standardization of the formatting of gait data and improved reliability, a wide variety of statistical approaches to data analysis and presentation have been undertaken. A recurrent question raised by clinical and statistical reviewers is the inclusion of data from both lower limbs. The variability in the approach taken by authors and the response by reviewers can lead to frustrating delays or even the rejection of some studies by this and other journals.
Ambulant children with CP have either one affected lower limb (hemiplegia) or both lower limbs are affected (diplegia). In spastic diplegia, gait pathology may be symmetric or asymmetric. What is usually unclear is the problem of the ‘non-independence of the two lower limbs,’ i.e. how much does one limb affect the other in gait?
Starting at the top of the skeletal chain for gait, there is one pelvis; it can be viewed morphologically as a semi-rigid bony box or ring, and it is modelled as rigid in instrumented gait analysis. Therefore, whatever the right side of the pelvis does during gait, the left side must do the same (for pelvic tilt) or the exact opposite (for pelvic rotation or obliquity) at exactly the same time. In hemiplegia, if the right side is retracted, the left side will be protracted by the same amount.
In spastic hemiplegia, even when there is clearly neurological involvement on only one side, there are well known compensatory changes in gait pattern on the unaffected side. Most intervention studies on patients with hemiplegia concern a unilateral procedure on the involved side. Data are therefore presented from this side and pre- to post-intervention changes can reasonably be attributed to the intervention. Now that the use of summary statistics of gait, such as the Gait Profile Score and the Gait Deviation Index are more widely used, it should be remembered that these are derived from multiple parameters, from multiple levels, and from both lower limbs. Therefore, an improvement in the Gait Profile Score after an intervention on the hemiplegic limb, receives contributions directly from the effects on the affected limb and also from resolution of compensatory gait deviations on the unaffected limb. For example, some children with equinus gait secondary to hemiplegia choose to adopt an equinus pattern on the unaffected side (vaulting) to improve clearance. Correction of the equinus deformity on the affected side usually leads to resolution of the compensatory equinus on the unaffected side.
The complexity of the interrelationship of the two lower limbs in spastic diplegia is more complex. In children with symmetric diplegia, the gait pathology may be symmetrical, the fixed musculoskeletal deformities may be symmetrical, and symmetric interventions may be advised during multilevel surgery. However, many children with diplegia have asymmetric neurological involvement, asymmetric gait pathology and asymmetric musculoskeletal pathology. Such children have complex interactions between levels of the same lower limb and between the two lower limbs. Surgical prescriptions are, therefore, often asymmetric. For example, there may be a decision to perform a procedure on one side only or to use a different ‘surgical dose’. Bilateral femoral derotations are commonly performed and just as commonly, asymmetric degrees of rotation are required and performed. In children with asymmetric equinus deformities, asymmetric surgery is often required, e.g. a gastrocnemius recession on one side and a gastrocsoleus recession on the other side.
In summary, interventions may be unilateral, bilateral but asymmetric, or bilateral and symmetric and a study cohort may include participants belonging to more than one category. There are currently no widely agreed methods of dealing with these complexities in study design, data analysis, or data presentation. A distinction should be made as to whether the outcome measure is a summary statistic of gait (involving both sides, Gait Profile Score or Gait Deviation Index) or whether the outcome measure is the direct effect (involving one level for one side, e.g. Gait Variable Score), such as an ankle dorsiflexion Gait Variable Score for gastrocsoleus lengthening.
If the primary outcome measure is a summary statistic of gait, then this will be calculated from the multiple parameters in both lower limbs according to the description by the creators of these indices.[1, 2] In this case, cohorts composed of only unilateral or only bilateral and symmetric procedures do not present major difficulties. In principle, unilateral and bilateral procedures and individuals with hemiplegia and diplegia should not be assessed together. If they are combined, the statistical analysis should still only include the one gait summary statistic per participant but distinguish between individuals who have had a unilateral procedure and individuals who have had bilateral procedures and estimate the effects separately.
If the outcome of a procedure is the direct effect at a single level then researchers have several options. Two simple approaches would be to either report the effect on both sides together (A1), or report only one side, even when bilateral procedures have been performed, (A2). Given that, for example, 10 children with spastic diplegia have 20 lower limbs, it is obvious that approach A2 halves the sample size and discards information with the consequence of less precise estimates of effect. Approach A2 may also introduce bias to the estimated effect if the chosen side has not been selected either randomly or according to a clear rule determined prior to data collection. Many authors reporting small data sets are keen to report data from both sides, i.e. take approach A1. However, analysing data from both sides without taking care of the correlation between the two lower limbs of an individual study participant falsely narrows confidence intervals and may lead to incorrect conclusions.
Another approach is to recognize that the two lower limbs are not independent, and to make a statistical adjustment to take into account correlations between the lower limbs of the same patient. Such adjustment may use one of the two following options: a marginal approach, with generalized estimating equations and robust standard error calculations to allow for excess correlation between outcomes in the two sides of the same individual; or a conditional approach with a mixed effect model containing a random effect for each study participant.
There remains the problem that the degree of interdependence or dependence between the two sides has usually not been studied nor is it clearly known. Studies in which the non-independence of the two lower limbs is recognized, explicitly stated, and in which a statistical adjustment has been made, are likely to be more favourably reviewed than those studies which simply ignore the problem. However, this is clearly an avenue for further research. Given the number of new reporting tools, including the Movement Analysis Profile and Gait Variable Score, it should be possible to gain insight to the relative dependence/interdependence of the two sides and the multiple levels in children with asymmetric spastic diplegia.
Certainly, better estimates of effect will be obtained from larger studies and use of all available data, in the light of a more informed understanding of inter-limb dependence in gait.