Reference Curves for the Gross Motor Function Measure: Percentiles for Clinical Description and Tracking Over Time Among Children With Cerebral Palsy

Cerebral palsy (CP) is a group of disorders affecting the developmentof movement and posture and causing activity limitations thatare attributed to nonprogressive disturbances that occurredin the developing fetal or infant brain.¹ Cerebral palsy occursin about 2 to 2.5 per 1,000 live births² and is the most prevalentchildhood neuromuscular condition seen by pediatric rehabilitationpractitioners, including physical therapists. The prognosisfor gross motor function among children with CP is extremelyvariable. This variability has been a key challenge for scientificdescriptions of motor function as children with CP develop.It is also a fundamental consideration for clinicians who regularlydeal with practical issues of examination, prediction, interventionplanning, and outcome evaluation³ when working with childrenwith CP.

In 2002, Rosenbaum and colleagues⁴ published a landmark longitudinalcohort study that dramatically improved knowledge of the developmentof gross motor function among children with CP. Rosenbaum etal⁴ used population-based sampling methods to conduct longitudinalassessments of the gross motor function of 657 children overapproximately 4 years. Children were stratified by age and bythe severity of the condition on the basis of the Gross MotorFunction Classification System (GMFCS) for CP.⁵ The motor examinationswere conducted with the 66-item version of the Gross Motor FunctionMeasure (GMFM-66), an evaluative instrument for use with childrenwith CP.^6–9 The result was a set of 5 motor developmentcurves, corresponding to each of 5 GMFCS levels of severity.The curves describe changes in GMFM-66 motor function scoreswithin strata of severity, in terms of the rate of developmentand a presumed limit of functional ability (Fig. 1).

View larger version (23K):
[in this window]
[in a new window]

Figure 1. Gross motor development curves representing average development predicted by the Gross Motor Classification System. The diamonds on the vertical axis identify 4 items of the 66-item Gross Motor Function Measure (GMFM-66) that predict when children are expected to have a 50% chance of completing that item successfully. The GMFM-66 item 21 (diamond A) assesses whether a child can lift and maintain his or her head in a vertical position with trunk support by a therapist while sitting, item 24 (diamond B) assesses whether a child can maintain a sitting position on a mat without support from his or her arms for 3 seconds, item 69 (diamond C) measures a child's ability to walk forward 10 steps without support, and item 87 (diamond D) assesses the task of walking down 4 steps by alternating feet with arms free. Reprinted with permission from Rosenbaum PL, Walter SD, Hanna SE, et al. Prognosis for gross motor function in cerebral palsy: creation of motor development curves. JAMA. 2002;288:1357–1363. Copyright 2002, American Medical Association. All rights reserved.

These gross motor measures and curves are now widely cited inthe clinical and scientific literature^10–12 as usefuland valid descriptions of gross motor changes among childrenwith CP. Scientific or clinical questions of prediction canbe approached from an initial GMFCS assignment that establishesthe appropriate range of functional expectations for an individualchild. Evaluating a child's GMFCS severity level is straightforward.¹³Once a child is assigned to a GMFCS level, the correspondingmotor change curve shows the average pattern of change in GMFM-66motor scores up to 12 years of age for children at that level.Rosenbaum et al⁴ also provided data about the degree of variabilityin limits of gross motor function expected among children withineach level.

With the curves provided by Rosenbaum et al,⁴ it is possibleto crudely evaluate children's gross motor capability relativeto the average for their age and GMFCS level. The user's manualfor the GMFM-66⁸ also provides item maps that aid in criterion-referencedinterpretations of scores by relating total GMFM-66 abilityscores to the probability of attaining motor tasks such as lyingand rolling, sitting, crawling and kneeling, standing, walking,running, and jumping. However, it has been difficult for cliniciansto make normative evaluations of children's motor capabilitywithin GMFCS levels because reference percentiles have not beenavailable. In this article, we develop and present referencepercentiles for GMFM-66 motor scores by age and GMFCS level.First, we review the clinical utility of the GMFM-66, the GMFCS,and the motor development curves as they pertain to examination,prediction, and intervention planning by using 3 case examples.After presenting the development of the reference percentiles,we revisit these examples to demonstrate how the percentilescan enhance assessment. We discuss the use of the percentilesboth for understanding a single GMFM-66 assessment and for trackinga child's motor ability over time.

Case Examples

Top

Abstract

Introduction

Case Examples

Motor Measures

We selected real gross motor function data for 3 children fromthe data set of Rosenbaum et al⁴ to use as case examples. Descriptionsof the children, with their names changed, are provided in Table 1.Each of these children had a diagnosis of spastic diplegia,although the severity of involvement varied widely (GMFCS levelsI, III, and IV), highlighting the variability for children withinthe classic descriptions of type of motor disorder and distributionof involvement.

View this table:
[in this window]
[in a new window]

Table 1. Description of Case Examples^{^a}

Motor Measures

Top

Abstract

Introduction

Case Examples

Motor Measures

As shown in Table 1, David, Jennifer, and Hardeep were in GMFCSlevels I, III, and IV, respectively. The GMFCS has become theinternational standard for classifying the severity of CP.¹²Table 2 shows basic descriptions of the gross motor functionabilities expected for children between the ages of 6 and 12years, according to the 5 GMFCS levels.⁵ The GMFCS is straightforwardand reliable when used by therapists having minimal contactwith children or gaining information from parental reports.¹⁴Children in level I have the fewest limitations in gross motorfunction and mobility, and children in level V have the greatestlimitations, such that little voluntary movement is possible.The GMFCS levels are assigned on the basis of separate criteriaaccording to age bands (<2 years, 2–4 years, 4–6years, and 6–12 years). The GMFCS levels can be used toestablish broad expectations for motor development and achievement.For instance, we expect that David (level I) will continue todevelop in gross motor abilities such that he will be able toperform all of the activities of a 5-year-old child who is developingtypically, with minor limitations in speed, balance, and coordination.Jennifer (level III) will be able to walk using a walker orcanes on a level surface and will likely use a manual wheelchairfor functional mobility. Hardeep (level IV) will need assistancefrom an adult for transfers and will use either a manual ora powered wheelchair for mobility. This information is usefulfor realistic goal setting in physical therapy as well as forlong-term planning by families with respect to arranging foraccessible home, school, and community environments.

View this table:
[in this window]
[in a new window]

Table 2. Expected Functional Abilities for Children Aged 6 to 12 Years, According to the Gross Motor Function Classification System (GMFCS)

The GMFCS also provides a context for interpreting more detailedexaminations with the GMFM.⁶ The full version of the GMFM contains88 items (GMFM-88) measuring children's abilities related tolying and rolling, sitting, crawling and kneeling, standing,walking, running, and jumping, with the most difficult itemson the scale representing abilities attained by children developingtypically by 5 years of age. Each item is scored by observationon a 4-point ordinal scale. The GMFM-66 was developed and validatedby use of Rasch analysis to create a unidimensional motor abilityscore with interval-level measurement properties.⁹ The Raschmodel that underlies the GMFM-66 also estimates the difficultiesof the items,⁷ so that a child's total ability score can beeasily related to the probability of attaining common motormilestones. Total possible scores range from 0 to 100.

Each of the 3 children selected as case examples had spasticdiplegia. On the initial occasion of measurement, David, a boywho was 3 years 2 months old, was in GMFCS level I and had aGMFM-66 score of 57.6. Both Jennifer and Hardeep were olderthan David, had more severe limitations, and had lower GMFM-66scores, 51.6 and 43.3, respectively. These scores, which areclearly not solely related to age, are useful for describingthe current motor abilities of individual children, particularlywhen used with the item maps described earlier.⁸ For example,David was able to walk independently but probably could notachieve standing through either left or right half-kneeling,lower himself to sitting on a bench without hand support, pickup an object from the floor without hand support, or walk upor down stairs by holding onto a railing and alternating feet.Jennifer could pull herself to a standing position by usinga surface and could cruise to left or right when holding on,but she could not stand without arm support.

Normative and predictive interpretations are facilitated byreferencing a child's GMFM-66 score against the average developmentalpattern for children in his or her GMFCS level. Such referencingcan be done with the motor development curves⁴ shown in Figure 1.For instance, the average child in GMFCS level I is expectedto achieve a GMFM-66 score of approximately 88 as he or sheapproaches adolescence. Rasch item maps from the user's manualfor the GMFM-66⁸ suggest that such a child has a high probabilityof being able to perform tasks such as independent walking ona level surface and walking down steps by alternating feet witharms free. However, there is considerable variability withinGMFCS levels; therefore, it is clear that the average predictedcapability is not the only factor. For instance, Rosenbaum etal⁴ reported variability in the ultimate limit of motor capabilitysuch that there is only a 50% chance of children in GMFCS levelI achieving scores of between 80.1 and 92.3. In addition tothe real variability in achievement, there is also measurementerror, meaning that no one assessment falls exactly on the truedevelopmental trajectory for a child.

Normative Percentiles

Top

Normative Percentiles

For a child of 38 months of age, David's GMFM-66 score (57.6)was well below the average score (68.8) predicted from the motordevelopment curve for children of this age in GMFCS level I.Two questions of interpretation immediately arise. First, giventhe variability in GMFM-66 at this age, is this difference fromthe average "large"? Second, given that this child's developmentmay be below average for his GMFCS level at age 3, will hisdevelopment remain below average as he ages?

Clinicians are generally familiar with normative reference percentilesfor answering the first question. Reference percentiles arecommonly used for interpreting measurements in medicine, suchas physical growth,¹⁵ and for defining the normal ranges fora wide variety of laboratory measurements.¹⁶ Many clinical measuresin pediatric rehabilitation also provide percentiles to facilitatenormative interpretation.^17–19 Reference percentiles areconstructed by selecting a clinically appropriate comparisongroup and developing a statistical summary of the distributionof scores for this group. This summary is transformed to a percentilescale, such that a child's percentile represents the percentageof children in the normative sample that he or she outperforms.Reference percentiles answer questions about function in termsof relative standing.

The question of whether David's motor ability will remain belowaverage relates to issues of both prediction and longitudinalvariability or tracking of measurements.²⁰ It is common fortherapists and families to assume that percentile rankings arestable, but in fact they can be highly variable over time.¹⁷Thus, whenever therapists want to monitor function over time,it will be important to assess the likely degree of variationin percentiles over time. If clinicians do not consider thisvariation, it can be mistaken for evidence of clinically significantchange.

In the present study, we used data from the sample of Rosenbaumet al⁴ to construct reference percentiles for GMFM-66 scoreswithin GMFCS levels for children with CP. We also evaluatedthe degree to which percentile rankings are stable over time,and we discuss how this factor affects the appropriate use ofpercentiles for clinical monitoring of gross motor function.Our goal was to increase the clinical applicability of the GMFM-66and GMFCS by providing reference curves in a form familiar tophysical therapists and other service providers who work withchildren with CP.

Method

Top

Method

Subjects

The sample used in the present study was described in detailby Rosenbaum et al.⁴ Children with CP and their families wererecruited to the study through 19 publicly funded regional ambulatorychildren's rehabilitation centers in the province of Ontario,Canada, beginning in 1996. These programs provide a range ofdevelopmental therapies and services, including physical therapy,occupational therapy, speech-language therapy, and recreationaltherapy. Because these centers are publicly funded and eachprogram serves the majority of eligible children in its treatmentarea, Rosenbaum et al⁴ argued that the sample was closely representativeof the population of children with CP in Ontario. Children wereincluded in the study if they had received a diagnosis of CPor if they had neuromotor findings consistent with CP, suchas spasticity or reflex abnormalities. Children were excludedif they had other neuromotor conditions, such as spina bifidaor muscle diseases. They also were excluded if they had everreceived botulinum toxin in the lower limbs or dorsal rhizotomyor were currently receiving intrathecal baclofen. The samplewas stratified by GMFCS severity level and age and, therefore,was not necessarily representative with respect to those factors.

From 936 randomly selected children and families who were eligibleand after refusals to participate (n=217; 23%), other lossesof contact, and later exclusions because of reevaluation ofeligibility, 657 children were included in the original motordevelopment analyses. Children under 6 years of age were assessedevery 6 months, and older children were assessed every 9 to12 months. This protocol yielded a total of 2,632 GMFM-66 assessments,with an average of approximately 4 observations per child.

The method used to create percentiles (see section on data analysisbelow) required that the observations be treated as cross-sectionalrather than longitudinal. To minimize the effects of the longitudinaldesign, while still including as many data as possible, we selecteda subsample from the original 2,632 observations to avoid multipleobservations from a child occurring close in time. To accomplishthis, we established ten 1-year age bands from 2 years up toand including 12 years of age. Observations were selected sothat there was no more than one observation per child per 1-yearage band. As a result, statistical smoothing operations usedto create the percentiles gave substantial weight to only oneobservation for each child at a given age. It is important tonote that these age bands were used only for creating the subsampleand were not used for fitting the percentiles. Observationswere included only for ages from 2 to 12 years, restrictinginferences to ages for which large amounts of data were available.This sampling also is consistent with the appropriate use ofthe GMFCS, which is currently applicable to children up to 12years of age.

As a result of this subsampling, a total of 1,940 observationsfrom 650 children were available for creating cross-sectionalreference curves. The distribution and characteristics of thissample within 5 levels of GMFCS severity are shown in Table 3.

View this table:
[in this window]
[in a new window]

Table 3. Description of the Sample Used to Create Reference Curves^{^a}

Outcome Measures

The severity of CP was classified with the GMFCS. The GMFCSis a reliable and valid system that classifies children withCP on the basis of the major age-appropriate gross motor activitiesthat they can typically accomplish, with particular emphasison functional mobility.^5,13 The use of the GMFCS requires familiaritywith the child but does not require formal training. Interraterreliability is good for children aged 2 to 12 years (kappa=.75)and not so strong for children under 2 years of age (kappa=.55).Although the GMFCS was recorded at each observation, we followedthe protocol of Rosenbaum et al⁴ in relying on the first availableGMFCS assignment for the children in the present study. Doingso ensured correspondence with previous research and reflectstypical clinical goals, in which an initial classification withthe GMFCS frames subsequent intervention planning and outcomeevaluation.

Gross motor function was examined with the GMFM-66.⁸ The validityof the GMFM-66 was established in several ways. Face validitywas established by examining the hierarchy of items by use ofRasch analysis.⁷ Sensitivity to change was determined by demonstratingthat children who were younger changed more than children whowere older and that more change occurred in children in GMFCSlevels I and II than in those in the other levels.⁷ Reliabilitywas initially established with trained raters, who obtainedintraclass correlation coefficients of .99 for both interraterreliability and test-retest reliability.⁷ In the study of Rosenbaumet al,⁴ all examining therapists were trained on the administrationand scoring of the GMFM.

Data Analysis

The LMS method of Cole and Green²¹ was used to construct referencepercentiles. The method summarizes the changing distributionof GMFM-66 scores as a function of age in terms of 3 curvesrepresenting the skewness (L), the median (M), and the coefficientof variation (S). Smooth, nonlinear fitting of the 3 LMS curvesis accomplished with cubic splines by use of a penalized likelihoodcriterion. Assuming that the original data are skew-normal,the resulting curves approximate a standard normal distributionat any given age and thus can be combined to produce percentiles.The degree of smoothness for each of the 3 LMS functions iscontrolled by selecting smoothness parameters, often expressedas expected degrees of freedom (edf), for each curve. In turn,the smoothness of the resulting percentile curves depends onthe smoothness of the LMS curves. Higher values for edf allowfor more complexity and less smoothing with respect to eachof the components of the curve—skewness (L), median (M),and variability (S)—with the goal being to select thesimplest model (ie, with lower values for edf) that preservesclinically plausible details in the percentile curves. The LMSmethod is a standard method for constructing cross-sectionalreference percentiles. Unlike many ad hoc methods, it producessmooth curves that incorporate the changing variability andskewness in the sample and does not require the arbitrary binningof subjects into crude age bands.

Determining the appropriate degree of smoothness for the LMScurves is necessarily a balance of smoothing out irregularitiesarising from sampling error without eliminating features ofreal clinical interest. For assistance in identifying the appropriatedegree of smoothness, changes in edf were evaluated as likelihoodratio tests based on changes in the penalized likelihood.²¹In a well-fitting model for the percentiles, the observationsshould conform to a standard normal distribution at any givenage after LMS transformation. Thus, the goodness of fit forcandidate models was evaluated by use of Kolmogorov-Smirnovtests of normality for the transformed data and by inspectionof Q-Q plots of the estimated percentiles.²¹

The LMS percentiles were constructed separately for each GMFCSlevel by use of the statistical routines provided by Carey,²²implemented in the R statistical programming language.²³ Separatecurves were not considered for boys and girls because of samplesize and because Rosenbaum et al⁴ found that gender was nota significant predictor of gross motor function trajectoriesin the sample studied.

To examine the degree of stability of GMFM-66 percentiles overtime, we exploited the longitudinal aspect of the original data.A subsample of the observations used in constructing the referencepercentiles was selected such that 2 GMFM-66 observations fromeach child were contributed to the sample. Children for whomonly one observation was available were excluded. For childrenwith more than 2 available measurements, the earliest 2 measurementswere selected. The resulting sample contained pairs of measurementsfor 570 children. For each pair of observations, estimated percentileswere extracted, and the means and standard deviations of thedifferences were calculated for each GMFCS level.

Results

Top

Results

Reference Curves for Gross Motor Function

Figures 2, 3, 4, 5, and 6 show the estimated reference curvesfor each GMFCS level, plotted at the 3rd, 5th, 10th, 25th, 50th,75th, 90th, 95th, and 97th percentiles. Note that the referencecurves begin at the earliest observed age in each data set,which varies slightly by GMFCS level.

View larger version (52K):
[in this window]
[in a new window]

Figure 2. Gross Motor Function Classification System level I percentiles. GMFM-66=66-item Gross Motor Function Measure. Figure 2 may not be used or reproduced without written permission from the authors.

View larger version (50K):
[in this window]
[in a new window]

Figure 3. Gross Motor Function Classification System level II percentiles. GMFM-66=66-item Gross Motor Function Measure. Figure 3 may not be used or reproduced without written permission from the authors.

View larger version (49K):
[in this window]
[in a new window]

Figure 4. Gross Motor Function Classification System level III percentiles. GMFM-66=66-item Gross Motor Function Measure. Figure 4 may not be used or reproduced without written permission from the authors.

View larger version (49K):
[in this window]
[in a new window]

Figure 5. Gross Motor Function Classification System level IV percentiles. GMFM-66=66-item Gross Motor Function Measure. Figure 5 may not be used or reproduced without written permission from the authors.

View larger version (50K):
[in this window]
[in a new window]

Figure 6. Gross Motor Function Classification System level V percentiles. GMFM-66=66-item Gross Motor Function Measure. Figure 6 may not be used or reproduced without written permission from the authors.

For GMFCS levels I to IV, well-fitting percentiles were obtainedat edf of 3, 5, and 3 for the L, M, and S curves, respectively.For level V, there were numerical problems in estimating anL parameter for adjusting skewness; therefore, the L parameterwas set to 1.0 to yield no adjustment for skewness, with edfof 4 and 4 for the M and S curves, respectively.

Kolmogorov-Smirnov tests of normality within each level werenot significant, indicating that the percentiles conform tothe expected normal distribution and that the model used inconstructing the percentiles fits the data well. To assess fitat particular age ranges, the data for each level were splitinto 5 age bands of equal width, and Kolmogorov-Smirnov testswere applied within each age band. None of these tests weresignificant at any age band, within any GMFCS level, indicatinga good fit at all ages. In addition, inspection of Q-Q plotsyielded no evidence that the percentiles fit poorly at particularages.

An approximate percentile can be obtained by consulting thefigure for a child's GMFCS level and finding the percentilecurve nearest the intersection of the child's age and the GMFM-66score. For most clinical purposes, it will be sufficiently accurateto visually interpolate between adjacent percentile curves (eg,halfway between the 10th and the 5th percentiles). If greateraccuracy is desirable, tabulated percentiles are available online²⁴or by contacting the corresponding author.

Longitudinal Stability of Percentiles

The means and standard deviations of the changes in percentilesby GMFCS level are shown in Table 4. The median time betweenobservations for each child was 1.0 year for each level, andneither time difference nor baseline age was correlated significantlywith the amount of change in percentiles.

View this table:
[in this window]
[in a new window]

Table 4. Mean Changes in Percentiles Over 2 Assessments, With Probability Intervals^{^a}

Table 4 shows intervals of expected change between 2 percentilemeasurements, corresponding to 20%, 50%, and 80% coverage probabilities.Thus, for a child in GMFCS level III, a second percentile measurementhas about an 80% chance of being within ±15.9 of thefirst measurement but only a 50% chance of being within ±8.4.As shown in Table 4, the expected within-child variability inpercentiles was substantial for all GMFCS levels, with the largestvariability being observed for levels I and II.

Discussion and Conclusion

Top

Discussion and Conclusion

By using the GMFCS, GMFM-66, and motor development curves (Fig. 1),physical therapists have had clinical instruments to assistwith examination, prediction, intervention planning, and outcomeevaluation for children with CP. However, percentile norms withinGMFCS levels have not been available to assist with the interpretationof relative standing or change over time. In this article, weprovide GMFM-66 percentile reference curves that are usefuladditional tools when therapists want to understand a child'scapability in the context of the usual development of otherchildren with CP who are classified at the same functional level.Next, using the case examples, we illustrate the applicationof these percentiles (Tab. 1), and then we discuss some generalissues of use and interpretation.

The reference curves are straightforward for use in evaluatinga child's relative capability at a single point in time. Thepercentiles for the case examples are shown in Table 5. Theywere obtained at this level of precision from the LMS output,but in a real application, tabulated percentiles or visual interpolationfrom Figures 2, 3, 4, 5, and 6 would normally be sufficient.On the first occasion of measurement (time 1), David, Jennifer,and Hardeep had percentile rankings of 14.9, 54.6, and 60.7relative to other children in GMFCS levels I, III, and IV, respectively.David had the highest GMFM-66 score of the 3 children but, nonetheless,had a much lower percentile ranking. This inconsistency betweencapability and percentile rankings occurred because the 3 childrenwere of different ages and in different GMFCS levels. In transformingscores to percentiles, the clinical meaning of original scoresin relation to functional criteria is set aside in favor ofrelative ranking. Clinicians and families must adjust theirinterpretations accordingly. This is especially obvious forthe GMFM-66 reference curves, because children's percentilesdepend on both age and GMFCS level. As another example, a 6-yearold child who scores 52 on the GMFM-66 is near the 95th percentileif she is in level IV but is near the 50th percentile if sheis in level III. Her motor capability is the same in eithercase. Nonetheless, if expectations and intervention planningare being based on the GMFCS, it may be of interest to knowthat she is a very highly functioning child in level IV. Incontrast to measures that derive their interpretation entirelyfrom norm referencing, such as many intelligence tests, usersof the GMFM-66 now have access to both functional and percentileinterpretations; it is likely that many examinations will relyon both.

View this table:
[in this window]
[in a new window]

Table 5. Changes in 66-Item Gross Motor Function Measure (GMFM-66) and Percentiles for Case Examples

The appropriate use of reference percentiles relies on a wisechoice of reference group. Percentiles derived from the scoresof children developing typically may be useful for screeningbut are of extremely limited value for examining function amongchildren with CP and other chronic disabling conditions.²⁵ Suchtests will generally fail to detect meaningful variations incapability among children in a clinical population who all fallbelow the extreme lowest percentiles of the typical developmentreference distribution. Furthermore, such a measure is incongruentwith the usual goals of service provision, which have littlerelevance to achieving "normal function." The GMFM-66 percentileshave important advantages in this regard. The 5 GMFCS levelsdefine clinically meaningful subpopulations that are alreadywidely used by therapists to organize examination and interventionplanning.¹²

These percentiles are based on a large data set that, withinGMFCS levels, is likely to be representative of the populationof children with CP that most therapists in Europe and NorthAmerica will serve. This is a key requirement for the effectiveinterpretation of normative comparisons and is an importantstrength of the present study. Rosenbaum et al⁴ did not controltherapeutic interventions during their study; therefore, thesample was representative of children who received a range ofaccepted medical, orthopedic, and developmental therapy services.Children receiving dorsal rhizotomy, botulinum toxin, and intrathecalbaclofen prior to their study were excluded. At the time, thesewere newer therapies with largely unknown effects, and theywere not readily available in Ontario. In any event, as Rosenbaumet al⁴ pointed out, these interventions are used only in highlyselective subgroups of children and would have little effecton the estimated distribution of GMFM-66 scores as a whole.

Cross-sectional percentiles are most easily applied when therapistswish to evaluate relative standing at a single point in time.Therapists routinely use them to evaluate longitudinal changein standing as well, although this use is not strictly validwithout consideration of the typical stability of percentiles.For example, a therapist who may be alarmed upon finding thata child in level I has dropped from the 50th percentile to the40th percentile upon reexamination should consult Table 4 andconsider that changes at least this large are quite common.For instance, for level I, 80% of reexaminations are expectedto change by up to 20 percentile points in either direction,meaning that 20% of changes are larger than this. The referencecurves require additional information to assist with the interpretationof change over time, as illustrated next.

Table 5 shows outcome data on David, Jennifer, and Hardeep aftera period of approximately 1 year. For David, who was in GMFCSlevel I, the GMFM-66 score changed from 57.6 to 66.0, a changeof 8.4 points. This change translates to a percentile rankingof 14.9 at the beginning of the year and a ranking of 19.6 atthe end of the year, an increase of 4.7. An examination of Table 4shows that this amount of change means that David is developingas might be expected; his change is well within the intervalwithin which 80% of children change.

In contrast, for Jennifer, classified in level III, the GMFM-66score changed from 51.6 to 49.9 (a decrease of 1.7 points).This change represents little or no loss of function but translatesto a change in percentile ranking from 54.6 to 37.5 over 1 year,a decrease of 17.1. Table 4 shows that this amount of changeis relatively uncommon in level III; there is an 80% chancethat the percentile will remain within 15.9 points of the initialranking on retesting 1 year later. Jennifer is not far outsidethis range, but the results may suggest that she is fallingbehind the developmental trajectory expected for children inlevel III.

The example for Jennifer shows how the percentiles can be helpfulin revising goals and interventions over the subsequent timeperiod to ensure that further unexpected changes do not occur.However, this example also illustrates that interpretationsof changes in percentile rankings should be tempered by knowledgeof the child's clinical situation, and we emphasize the hazardsof overinterpreting longitudinal comparisons. For Jennifer,an uncommonly large decrease in percentile ranking resultedfrom a decrease of only 1.7 points in GMFM-66 scores, a findingthat may have little importance in functional terms. Indeed,such a small raw score change may not be statistically significantgiven the standard error of the measure; if not, one cannotbe confident that any functional change has occurred. A largedownward percentile change arises nonetheless, because the distributionof scores is changing; Jennifer may not be changing, but otherchildren are, and she is falling behind. However, "falling behind"is not the same as losing function. When large changes in percentilesoccur, we encourage therapists to consider the functional meaningof the raw scores by using interpretative aids in the user'smanual for the GMFM-66.⁸

Finally, for Hardeep, classified in level IV, the GMFM-66 scorechanged from 43.3 to 47.1, a modest increase of 3.8 points.This change translates to a change in percentile ranking from60.7 to 77.8, an increase of 17.1. Table 4 shows that thereis an 80% chance that percentiles for children in level IV willnot change more than 15.1 points on retesting after 1 year.For Hardeep, one can conclude that his development was betterthan expected over the preceding year.

These examples illustrate how to use the reference curves toapproximately interpret changes in percentiles. The resultsshown in Table 4 suggest that large changes in percentile rankingsover 1 year are quite common at all levels of the GMFCS. Thissuggestion is consistent with the findings for gross motor developmentamong children developing typically.^17,26 Our approach to approximatelyquantifying the longitudinal stability of GMFM-66 percentileswill be useful for interpreting changes. We are presently developinga longitudinal approach that will further improve the predictionof changes in GMFM-66 percentiles. In the meantime, the cross-sectionalpercentiles presented here are important new tools for the clinicalassessment of motor function among children with CP.

Footnotes

Dr Hanna, Dr Bartlett, Ms Rivard, and Dr Russell provided concept/idea/researchdesign and consultation (including review of manuscript beforesubmission). Dr Hanna and Dr Bartlett provided writing and projectmanagement. Dr Hanna and Ms Rivard provided data analysis. Wethank Dr Peter Rosenbaum and the other members of the researchteam for their support and access to the data. In addition,we acknowledge the support of the faculty and staff at the CanChildCentre for Childhood Disability Research. CanChild is supportedby a grant from the Ontario Ministry of Health and Long-TermCare.

The study was approved by the Faculty of Health Sciences ResearchEthics Board at McMaster University.

Collection of the original source data was supported by grantsfrom the Canadian Institutes of Health (MT-13476) and the NationalCenter for Medical Rehabilitation Research of the US NationalInstitute of Child Health and Human Development (R01-HD-34947).

Dr Russell, as one of the authors of the Gross Motor FunctionMeasure (GMFM-66 and GMFM-88) User's Manual, receives royalties,which are deposited into a research account and not taken forpersonal use.

References

Top