Rasch analysis of UK FIM+FAM in patients with TBI
DISCUSSION
This study represents the first Rasch analysis of the UK
FIM+FAM scale in patients with TBI. The aim was to
determine whether the instrument fits the Rasch model
in a similar manner to patients with stroke, and if so, to
derive a conversion table of Rasch-transformed scores
for potential clinical use.
The best fit model was obtained when items were
combined into the 3 super-items, and the scale fitted
the Rasch model reasonably without any item dele-
tion, amendment or rescoring. However, as with the
stroke patients, at both ends of the scale the relations-
hip between the Rasch measure and the raw scores is
extremely steep, while in the middle of the scale the
distribution of the scores is notably ”flat”. Table V also
confirms that the difference in the summed raw scores
between 97 and 197 (100 units on the horizontal axis
in Fig. 3) corresponds to a difference in the summed
Rasch interval admission scores between 120 and 140
(i.e. just 20 units on the vertical axis). That many raw
score points are to be found within a narrow logit range
of functioning reflects the fact that the thresholds from
different items fall closely together across this functio-
nal space and presents a good example of the tension
between a Clinimetric and Psychometric perspective
(31). The clinical information is important at indivi-
dual level, as patients are improving on tasks within a
clinical framework that is recognisable by clinicians,
but the psychometric evidence is that such improve-
ment (i.e. each additional raw score point) is marginal
in the overall scale of functioning, and could give a
misleading impression of the speed of recovery. The
opposite is true for those at the margins of the scale.
Linearisation thus confers the potential advantage of
greater discrimination in the upper and lower ends,
together with more robust statistical properties – for
example enabling calculation of valid change scores,
which may be important for research, but transformed
scores are less recognisable by clinicians, which may
explain their limited uptake to date in clinical settings.
We suggest that, while the Rasch-transformed scores
may be useful for research purposes, their widespread
adoption into routine clinical practice will require
further work at the translational level.
Because people were expected to have a preponde-
rance of lower scores on admission and higher scores
on discharge, our main Rasch analysis was conducted
on a mixed admission and discharge sample, to ensure
that the full range of the response categories were
represented. As illustrated in Table III, this resulted in
a preponderance of scores in the lowest and highest
response categories in the sample, which effectively
creates a U-shaped distribution, as opposed to the
more normal distribution that would be expected in a
573
cross-section of a disabled population. This bi-modal
distribution reflects the chosen time-points at each end
of the programme, rather than being a reflection of the
scaling properties of the UK FIM+FAM (i.e. people
not being able to endorse answers in the middle range
of the scale). However, while the scarcity of mid-range
scores affected the PSI by inflating measurement error,
it should not have any influence on the Rasch logistic
model shape because the Rasch model has no distri-
butional assumptions. Given the extreme (bi-modal)
distribution of the TBI scores, the PSI value of 0.81
found in this study may be considered satisfactory
for group analyses, but use for individual assessment
warrants further clinical testing.
This relationship between study sample distribu-
tion and the Rasch model fit is an interesting finding
that deserves further exploration with psychometric
techniques, including generalizability theory (32). The
floor and ceiling effects of the UK FIM+FAM are well
recognized for the population as a whole (14, 15). Ho-
wever, as with many scales used for outcome evaluation
in rehabilitation, the level descriptors were designed
to capture the types of change that may be expected to
occur within an inpatient rehabilitation setting, as the
patient progresses to the level of independence required
for transition to the community. In this context the range
is quite suitable for the population of inpatients with
complex disabilities requiring specialist rehabilitation
following acquired brain injury (which is the setting
in which it is used in the UK(20)). But, because the
FIM+FAM is rated only on admission and discharge
in UK clinical settings, there is limited opportunity to
capture progression through the range of scores that
may be expected to occur during the rehabilitation
journey. Were the instrument to be rated at frequent
intervals throughout the programme (as occurs with
the FIM in some US rehabilitation programmes) one
might expect a more normal distribution of scores. In
view of the time commitment, this is not feasible in a
busy clinical setting. A more normal-shaped distribution
could be also achieved by purposive selection, but this
would not reflect normal clinical experience.
Limitations
In addition to the limitations above, the authors re-
cognize that the UKROC population is a selected
group of patients with severe TBI and highly complex
rehabilitation needs requiring specialist rehabilitation.
The findings may not therefore not be generalizable
to the overall TBI population. However, they would
be relevant to other groups of TBI patients requiring
treatment in specialist or tertiary centres, which are
probably the main services that are likely to be using
the UK FIM+FAM as an outcome measure.
J Rehabil Med 51, 2019