Page 1 of 24: Number of Words (Abstract Until References) : 3523 Number of Figures: 2 Number of Tables: 3

Page 1 of 24
1
This article has been peer-reviewed and accepted for publication, but has yet to undergo copyediting and proof correction. The final published version may differ from this proof.
Title: Using post-traumatic amnesia to predict outcome following traumatic brain injury.
Running Head: Post traumatic amnesia and outcome prediction

Journal: Journal of Neurotrauma
Running Title: Predicting outcome using PTA
Using post-traumatic amnesia to predict outcome following traumatic brain injury. (doi: 10.1089/neu.2015.4025)
Authors:
Jennie L Ponsford, BA (Hons), MA (Clin Neuropsych), PhD
School of Psychological Sciences, Monash University, Melbourne, Australia, and Monash-
Epworth Rehabilitation Research Centre, Epworth Hospital, Melbourne, Australia
Jennie.Ponsford@monash.edu
Gershon Spitz, BA (Hons), PhD

Gershon.Spitz@monash.edu
Journal of Neurotrauma
Dean McKenzie BA (Hons), PhD

Research Development & Governance, Epworth Healthcare, Melbourne, Australia
and School of Public Health and Preventive Medicine, Monash University, Melbourne,
Australia
Dean.McKenzie@epworth.org.au
Correspondence to: Jennie L Ponsford, BA (Hons), MA (Clin Neuropsych), PhD

Jennie.Ponsford@monash.edu, Tel: +61 3 9905 1552
Number of words (Abstract Until References): 3523

Number of Figures: 2
Number of Tables: 3
Page 2 of 24
2
Abstract
Duration of post-traumatic amnesia (PTA) has emerged as a strong measure of injury severity
following traumatic brain injury (TBI). Despite the growing international adoption of this
measure, there remains a lack of consistency in the way in which PTA duration is used to
classify severity of injury. This study aimed to establish the classification of PTA that would
best predict functional or productivity outcomes. We conducted a cohort study of 1041
individuals recruited from inpatient admissions to a TBI rehabilitation centre between 1985
and 2013. Participants had a primary diagnosis of TBI, emerged from PTA prior to discharge
from inpatient hospital, and engaged in productive activities prior to injury. Eight models that
classify duration of PTA were evaluated, six that were based on the literature and two that
were statistically-driven. Models were assessed using area under the receiver operating
characteristic curve (AUC) as well as model-based Akaike Information Criterion (AIC) and
Bayesian Information Criterion (BIC) statistics. All categorisation models showed longer
PTA to be associated with a greater likelihood of being non-productive at one year following
TBI. Classification systems with a greater number of categories performed better than two-
category systems. The dimensional (continuous) form of PTA resulted in the greatest AUC,
and lowest AIC as well as BIC, of the classification systems examined. This finding indicates
that the greatest accuracy in prognosis is likely to be achieved using PTA as a continuous
variable. This enables the probability of productive outcomes to be estimated with far greater
precision than that possible using a classification system. Categorising PTA to classify
severity of injury may be reducing the precision with which clinicians can plan the
management of patients following TBI.
Keywords: traumatic brain injury, post traumatic amnesia, outcome prediction

Page 3 of 24
3
Traumatic brain injury (TBI) is a leading cause of mortality and morbidity, which
results in a range of physical, cognitive, behavioural, and emotional changes. These changes
reduce the likelihood of return to functional independence, employment, study, leisure
pursuits, as well as personal and social relationships. 1 Improving the prediction of long-term
outcomes following TBI is, therefore, a crucial goal in order to guide clinical management
and long-term planning.
Measurement of injury severity is vital following TBI as it is considered to be one of
the strongest predictors of outcome. Several indices are used to assess severity of injury,
including length of coma (LOC), the Glasgow Coma Scale (GCS), and duration of post-
traumatic amnesia (PTA).2, 3

Length of coma has proven to be a less reliable index due to
differences in its measurement. The GCS has traditionally been most widely used as the gold-
standard measure, associated with in-hospital mortality and functional outcome.4
Duration of PTA has emerged as another strong measure of injury severity,
representing the length of time from injury until return of orientation and continuous memory
for events.5, 6
PTA duration has been associated with presence or extent of skull fracture,
intracranial haemorrhage, raised intracranial pressure, residual neurological deficits, extent of
neuropathology, as well as with longer-term functional outcomes and return to employment. 6-

9
Recent studies investigating individuals surviving to discharge from hospital provide
support for PTA as a stronger predictor of longer-term functional outcome, return to
employment, and cognitive impairment, compared to GCS or LOC. 10-12 It also accounts for
more variance in outcome than socio-demographic factors.13, 14
The measurement of PTA has evolved substantially, from retrospective estimates
relying on patient responses regarding their first memory following emergence from PTA to
the development of scales that objectively, and prospectively, monitor depth and recovery of
Page 4 of 24
4
PTA.15 PTA is generally measured in terms of minutes, hours, or days, dating from the
moment of injury until the individual is orientated in person and place, while also being able
to lay down new memories.16
Several different classification systems have been proposed in an attempt to
characterise injury severity. Russell and Smith6 and Jennett and Teasdale17 proposed that PTA
below 5 minutes should be considered very mild; 5-60 minutes mild; 1-24 hours moderate; 1-
7 days severe; 1-4 weeks very severe, and more than 4 weeks considered extremely severe
TBI. The above researchers assessed PTA by interviewing patients retrospectively. This
classification system, however, was based on clinical observations of patient cohorts, rather
than statistical analysis of outcome data. Several recent studies have utilised cohorts from the
National Institute on Disability and Rehabilitation Research funded TBI Model Systems.
These studies documented PTA duration exceeding 24 hours, assessed prospectively at least
three times weekly using either the Galveston Orientation and Amnesia Test or Orientation-
Log (O-Log). End of PTA was defined as the first of the two consecutive days of GOAT > 75
or O-Log > 24 scores or one day following hospital discharge. Walker et al.12 reported that
the probability of Good Recovery (the definition including return to previous employment
capacity) on the Glasgow Outcome Scale at one year following injury is less than 10% for
individuals with PTA duration of 56 days or longer. Brown et al. 10 identified a PTA duration
of 48 days to be indicative of poor employment outcomes at one year following injury. The
same group showed that at one year following injury competitive employment was predicted
by PTA extending into week four after injury.18 In addition, Nakase-Richardson et al.19 found
that TBI patients differed significantly in productivity at one year when classified according
to the Mississippi PTA intervals. The latter define moderate severity as PTA duration
between 0 and 14 days, moderate severe 15 to 28 days, severe 29 to 70 days, and very severe
as PTA duration longer than 70 days. In these studies return to productivity/employment was
Page 5 of 24
5
defined as being a full-time or part-time student, competitively employed or taking care of
house or family.
Categorisation of continuous, or dimensional, variables such as duration of PTA is
commonly performed in medical research (e.g. diabetes, depression, BMI, waist-hip-ratio,

systolic and diastolic blood pressure) in an effort to simplify both clinical interpretation and
statistical analysis. In practice, however, such categorisation often leads to a loss of statistical
power. Alternatively, in circumstances involving possible confounding variables such as age,
categorisation may result in an increase in type one errors or false alarms in rejecting the null
hypothesis.20 Categorisation may also affect interpretation; for example a recent Australian
study of invitation to an outpatient cardiac rehabilitation program by age, showed very
different effects depending upon whether age was categorised into five- or ten-year
intervals.21 Although the authors of recent applied statistical texts continue to discourage the
use of categorisation, it is still often employed in clinical research and practice.22, 23
In order to directly ascertain the effects of categorising PTA, the present study
compared different methods of categorising PTA with methods that retain the duration of
PTA as a dimensional variable. The aims of the study were two-fold: firstly, to establish the
best classification of PTA that would predict functional or productivity outcomes, and
secondly to compare the various classification systems with a simple dimensional variable
approach based upon standard logistic regression.
METHODS
Study population and Design
The study was approved by the Human Ethics Committees of Epworth Healthcare and
Monash University. Participants were recruited as part of a longitudinal head injury outcome
study from consecutive inpatient admissions to a TBI rehabilitation centre in the context of a
no-fault accident compensation system administered by the Transport Accident Commission

Page 6 of 24
6
(TAC) or Worksafe, which are state agencies help avoid and manage compensation claims
following injury. Patients were admitted between 1985 and July 2013. Inclusion criteria
included a primary diagnosis of non-penetrating TBI with emergence from PTA, age at injury
greater than 16 years, and pre-injury engagement in productive employment or study.
Exclusion criteria included no recorded productivity status at either premorbid or one-year

post-accident time-points, no recorded duration of PTA, permanent confusional or amnestic
state, or not being engaged in productive activities prior to injury.
Procedure
Patient demographic and injury-related information (gender, age at injury, number of
years of education, marital status, living location, premorbid productivity, premorbid medical
history, injury cause, inpatient CT scan results, and physical injuries) was retrieved from
medical files at hospital admission. One-year productivity outcomes were collected at follow-
up for the longitudinal study.
Assessment of Productivity Status
Participants pre-morbidly unemployed or not currently in the labour force were
classified as non-productive. Participants who were classified as apprentices, full-time or
part-time employees, or as secondary or tertiary students were classified as productive. One-
year productivity status was captured on a structured outcome questionnaire. Individuals
classified as medically/surgically delayed, not in the labour force, engaged in work trial, work
conditioning or retraining, unemployed, or non-vocational were categorised as non-
productive. Participants were considered as productive if they were employed in their
previous position or an alternative position full time or part-time.

Page 7 of 24
7
Assessment of PTA Duration
The WPTAS24 is a 12-item measure that assesses orientation for person, place and
time, as well as anterograde memory. The WPTAS was administered daily by a
neuropsychologist until patients had emerged from PTA, after having answered all items
correctly on three consecutive days, or otherwise cleared by a neuropsychologist.

PTA classification systems
Eight PTA classification systems were compared, six that were previously proposed
in the literature and two that were derived statistically using the WPTAS data collected in the
current study. The classification models recommended in the literature were those of:
Russell/Jennett,17 Walker,12 Brown,10 Brown,18 Mississippi,19 and Collapsed Mississippi.19
Russell and Jennett17 categorised PTA into <1 day, 1-7 days, >7-28 days, > 28 days. In
predicting Good recovery on GOS,Walker et al.12 categorised PTA into 0-56 days and >56
days. In predicting return to productivity defined as full- or part-time work or study or home
duties, Brown et al.10 categorised PTA into 0-48 days and >48 days; Brown et al.18
categorised PTA into 0-28 days and >28 days; the Mississippi19 system categorised PTA into
0-14 days, 15-28 days, 29-70 days, >70 days; and the Collapsed Mississippi19 system
categorises PTA into 0-14 days, 15-28 days, >28 days. In these studies, PTA was assessed
either using the GOAT, O-Log or retrospective file audit, with PTA estimated as length of
stay plus one day in cases still in PTA on discharge. Two other classifications were examined
in the current study that were derived through statistical analysis using the WPTAS data
25,26
collected in the current study. These were Classification and Regression Trees (CART)
and a simple logistic regression approach that left duration of PTA in its
continuous/dimensional form. These methods are further explained below.

Page 8 of 24
8
Statistical Analysis
Statistical analyses were conducted using Stata Version 1227 and CART Version 7.25
The original dataset was randomly divided into a learning sample comprising 75% of
observations used to develop the statistically-derived PTA classification models, and a hold-
out sample of 25% of observations, used to test the models. PTA classification schema were
initially developed using the learning dataset and cross-validated on the hold-out dataset.
Logistic regression was used to predict productivity group membership at one year
using PTA in its continuous or dimensional form. CART is a classification tree approach that
divides dimensional variables into binary splits at each stage of the tree-growing process, and
then “prunes” the tree using cross-validation.26 CART firstly grows a decision tree until the
resulting subgroups are too small for further analysis (generally five observations or less).
This tree is then pruned, based upon how well the performance of the tree generalises to other
data. Performance is assessed using 10-fold cross-validation.26 The dataset is firstly divided
into ten equally sized sub-sets, trees are then constructed using nine sub-sets and then tested
for generality on the excluded sub-set, with each sub-set being excluded in turn. CART has
been used in medical research for many years28 and has previously been employed in the
analysis of PTA data by Brown et al.10
The appropriateness of each PTA classification and regression model was assessed
using the area under the Receiver Operating Characteristic (ROC) curve,29 calculated using
Stata version 12. Areas under the ROC for different models were calculated using the
nonparametric method of De Long et al.30 The Akaike Information Criterion (AIC) and
Bayesian Information Criterion (BIC) information criteria, which attempt to balance model
complexity (e.g. number of categories) with model performance were used to compare each
of the models with one another.31 PTA classification systems were evaluated based on their
AUC, with greater values representing overall better classifications, as well as their
Page 9 of 24
9
associated AIC and BIC model statistics, with lower values representing better model
performance balanced against the number of parameters of the model (e.g. the number of
categories). The BIC also takes the number of observations into account.31
RESULTS
Patient Characteristics
Of the 1461 patients admitted between 1985 and July 2013, 279 were engaged in
non-productive activities prior to injury, 23 were aged under 16 years, 89 had no recorded
premorbid or one-year productivity status, 25 had no recorded duration of PTA, and 3
participants had unresolved PTA. The study sample comprised the remaining 1041
participants (Table 1). Of these, 455 (43.7%) were classified as non-productive at 12 months
post-injury and 586 (56.3%) as productive.
----------------------------------------------------------------------------------------------------------------
INSERT TABLE 1
----------------------------------------------------------------------------------------------------------------
PTA Classifications
A minority of participants in the current study had a PTA duration less than a day
(Table 2). Conversely, only 5.7% of the sample had a PTA exceeding 70 days. All
classification systems showed a greater likelihood of being non-productive at 12 months post-
injury with greater length of PTA.
----------------------------------------------------------------------------------------------------------------
INSERT TABLE 2
----------------------------------------------------------------------------------------------------------------
The dimensional (continuous) form of PTA, analysed using logistic regression,
obtained the greatest AUC of the selection of classification systems (Figure 1). This was the
case in both the learning and hold-out datasets. In the learning dataset, dimensional PTA
performed significantly better (as measured using AUC) than all other classification systems.
Classification systems with a greater number of categories, including the Russell/Jennett,
Mississippi, and Collapsed Mississippi, performed better than those with only two categories.
Interestingly, the CART statistically-derived approach found a two category solution, below
Page 10 of 24
10
or above 30.5 days of PTA. Although a simple binary classification, this system did not
perform as well as systems with a greater number of categories. Similar AUCs were observed
in the hold-out dataset with the reduced sample size. The only change was in the
Russell/Jennett classification, which displayed a poorer AUC. Dimensional PTA achieved a
statistcally superior AUC compared to the Russell/Jennett, CART, Brown 2010,18 and Walker
classifications. All other classifications remained in the same ordering, based on AUC,
between the learning and hold-out samples.
----------------------------------------------------------------------------------------------------------------
INSERT FIGURE 1
----------------------------------------------------------------------------------------------------------------
The relationship between the probability of being non-productive at one-year
following injury and duration of PTA appears to be monotonic (increasing), if not linear
(Figure 2, Continuous PTA panel). On the other hand, it is evident that artificial constraints
may be imposed by categorisation of PTA. For example, leaving PTA as a dimensional
variable suggests that individuals with PTA duration between 29 and 70 days have a
probability of being non-productive of 43% to 70%, respectively. Conversely, a PTA
duration between 29 and 70 on the Mississippi scale would indicate that all individuals with a
duration between 29 and 70 days have a 57% chance of being non-productive at one-year
following injury, when assessed by logistic regression using the Mississippi scale as a
categorical predictor. A greater number of cut-offs were found to increase the AUC, as has
been observed in other studies.33 Conversely, the Walker and Brown 2010 classifications,
having only a single cut-off, were found to perform the worst, with lower performance than
the single cut-off CART decision rule.
Sensitivity for all models was generally low, correctly classifying only between 16 to
39% of non-productive individuals at one year for the learning dataset, and between 18 to
45% in the hold-out dataset (Table 3). Specificity for the models was quite high, correctly
Page 11 of 24
11
classifying 84 to 98% of productive individuals in the learning dataset, and between 77 to
95% of productive individuals in the hold-out dataset. Both AIC and BIC model-based
statistics suggest that PTA in its continuous form is the superior model of the set of
competing classification systems, both in the learning and hold-out datasets.

----------------------------------------------------------------------------------------------------------------
INSERT FIGURE 2
----------------------------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------------------------
INSERT TABLE 3
----------------------------------------------------------------------------------------------------------------
DISCUSSION
PTA duration is increasingly used as a measure of injury severity to predict outcome
following TBI. Despite the growing international adoption of this measure, there remains a
lack of consistency in the way in which PTA duration is being used to classify severity of
injury. Methods most commonly employed, such as those proposed by Russell and Smith and
Teasdale and Jennett are not based on systematic analysis of clinical outcome data.6, 17
More
recently, classifications based on clinical data analysis, such as the Mississippi PTA Intervals,
have been proposed, but these have not been widely adopted. The present study compared
different methods of categorising PTA with methods that retain the duration of PTA as a
dimensional variable in terms of their prediction of productivity outcomes in a consecutive
sample of patients with TBI admitted to rehabilitation.
Eight classification models were examined, six that were based on the literature and
two that were statistically-driven, and applied to our sample. All systems showed longer PTA
to be associated with a greater likelihood of being non-productive at one year following TBI.
Of the categorical classification systems, those with more categories, including the
Russell/Jennett, Mississippi, and Collapsed Mississippi intervals, performed better than two-
category systems, including the statistically derived CART binary cut-off of 30.5 days.
Page 12 of 24
12
Although the Russell/Jennett classification did not perform as well when tested in the hold-
out dataset, the Mississippi and Collapsed Mississippi classifications continued to perform
relatively well. However, the dimensional (continuous) form of PTA, resulted in the greatest
AUC of the classification systems examined. This was the case in both the learning and hold-
out datasets. In addition, the dimensional model of PTA displayed the lowest AIC and BIC
model statistics of the models assessed, providing further support for this model.
This finding indicates that the greatest accuracy in prognosis is likely to be achieved
using PTA as a continuous variable. This enables the probability of productive outcomes to
be estimated with far greater precision than that possible using a classification system. For
example using the Mississippi PTA intervals, all individuals with PTA ranging from 29 to 70
days are treated the same with regard to predicted outcome. Using the dimensional approach
it is possible to ascertain that, based on the current dataset, an individual with 29 days of PTA
has a 46% probability of being non-productive at one year post-injury, whereas a person at
the mid-point with PTA of 49 days has a 60% probability of being non-productive and a
person with a PTA of 70 days has a 74% probability of being non-productive. Rather than
categorising individuals into a discrete set of PTA duration bands, clinicians may be able to
make more accurate predictions of productivity outcomes using continuous measures of PTA.
An important question concerns the utility of such probability calculations. It is not difficult
to compute individual probabilities for each unique value of PTA duration. This would lead
to increased precision of prognosis than could be achieved by 2, 3, or 4 categories used in the
Mississippi or other classification systems. The computations for prediction using logistic
regression equations may appear complex at first glance, as they involve exponentiation. It is
a relatively straightforward procedure to program the calculations in a spreadsheet on a tablet,
however, or as an application for the simplest of ‘smart’ phones.

Page 13 of 24
13
It needs to be acknowledged that outcome was only measured in terms of return to
productivity for those individuals engaged in productive activities prior to injury. Return to
productivity was measured in a similar way to previous studies of Model Systems samples,
except that it did not include engagement in home duties pre- or post-injury as productive
activity. The findings may not generalise to groups who were not productive prior to injury or
to prediction of other forms of outcome, such as functional status. In addition, the current
study did not investigate the extent that duration of PTA predicts recovery of specific
domains within the ‘productive’ umbrella, such as cognition, motor function, and behaviour.
PTA was assessed using different measures in the Model Systems studies from that used in
the present study. This may have resulted in some systematic differences in PTA duration
across studies which may in turn influence categorisation.

Although PTA is an essential factor to consider when examining prognosis, it must
also be acknowledged that other factors may play a role in outcome prediction, including
demographic variables, particularly age, premorbid neurological conditions, personality, and
social support. However, in previous studies PTA has been shown to be a stronger predictor
of productivity and functional outcome than these other factors. 13-14 Interestingly, Brown et
al.10 used CART, logistic, and discriminant analysis with a range of predictor variables. They
found that CART performed best at developing useful multivariate models that predicted
outcome. This study shows that PTA is an important variable that may assist in making
clinical judgements, but should also be considered in the context of other demographic and
social predictors
The study focused on a consecutive sample admitted for rehabilitation in the context
of a government-funded non-fault accident compensations system that provided rehabilitation
to individuals injured in motor vehicle or work-related accidents regardless of fault. The
sample, therefore, included a predominance of transport or work-related injuries, with

Page 14 of 24
14
relatively few falls, assaults or sporting injuries and very few mild injuries. The other studies
used to create PTA cut-offs included a higher proportion of injuries due to falls, assaults and
other injuries (approximately 33% vs 15% in current study), and participants had a slightly
higher mean age of around 37 years as opposed to 31 years in the present study. These factors
might have differentially influenced return to productivity outcomes in these studies. The
purpose of this study, however, was not to compare outcomes per-se, but to evaluate the use
of PTA in predicting outcome. None of these studies included many participants with mild
TBI. There is already some evidence to suggest that PTA duration is less predictive of
outcome following mild TBI. One other issue raised by the study findings concerns the need
for examination of the nomenclature used to describe injury severity. There is considerable
variability across studies in the methods and measures used to classify the severity of injury.
The findings of the present study suggest there is a need to review the way in which PTA
duration is initially recorded and. in turn, is being used to describe the severity of TBI.
The present study has shown that using PTA as a continuous variable allows for
prediction of productivity outcomes in individuals with moderate to severe TBI with greater
precision than appears possible using categorical systems. This underscores the importance of
obtaining continuous, prospective measures of PTA using validated scales,32 notwithstanding
the need to consider the costs of adding a daily assessment procedure lasting 5-10 minutes
into patient care routines. Given that the use of the GCS has now become standard practice in
the management of TBI, it should be possible to work towards further integrating the
assessment of PTA into routine clinical practice.

This article has been peer-reviewed and accepted for publication, but has yet to undergo copyediting and proof correction. The final published version may differ from this proof. Page 15 of 24
Acknowledgement
Author disclosure statement:
No competing financial interests exist.

for Safety, Compensation and Recovery Research (ISCRR).
This project is funded by the Transport Accident Commission (TAC), through the Institute
15
Page 16 of 24
16
References
1. Ponsford JL, Sloan S, Snow P. Traumatic brain injury: Rehabilitation for everyday
adaptive living. 2nd Edition. London: Psychology Press; 2012.
2. Teasdale G, Jennett B. (1974). Assessment of coma and impaired consciousness: A
practical scale. Lancet. 304(7872): 81-4.
3. Symonds C, Ritchie Russell W. (1943). Accidental head injuries: Prognosis in service
patients. Lancet. 241(6227): 7-10.
4. Teasdale G, Maas A, Lecky F, Manley G, Stocchetti N, Murray G. (2014). The
Glasgow Coma Scale at 40 years: standing the test of time. Lancet Neurol. 13(8): 844-
54.
5. Tate RL, Pfaff A, Baguley IJ, et al. (2006). A multicentre, randomised trial examining
the effect of test procedures measuring emergence from post-traumatic amnesia. J
Neurol Neurosurg Psychiatry. 77(7): 841-9.
6. Russell WR, Smith A. (1961). Post-traumatic amnesia in closed head injury. Arch
Neurol. 5(1): 4-7.
7. Wilde EA, Bigler ED, Pedroza C, Ryser DK. (2006). Post-traumatic amnesia predicts
long-term cerebral atrophy in traumatic brain injury. Brain Inj 20(7): 695-9.
8. Sigurdardottir S, Andelic N, Roe C, Schanke AK. (2009). Cognitive recovery and
predictors of functional outcome 1 year after traumatic brain injury. J Int Neuropsychol
Soc. 15(5): 740-50.
9. van der Naalt J, van Zomeren AH, Sluiter WJ, Minderhoud JM. (1999). One year
outcome in mild to moderate head injury: the predictive value of acute injury
characteristics related to complaints and return to work. J Neurol Neurosurg

Psychiatry. 66(2): 207-13.
10. Brown AW, Malec JF, McClelland RL, Diehl NN, Englander J, Cifu DX. (2005).
Clinical elements that predict outcome after traumatic brain injury: A prospective
multicenter recursive partitioning (decision-tree) Analysis. J Neurotraum. 22(10):
1040-51.
11. Sherer M, Struchen M, Yablon S, Wang Y, Nick T. (2008). Comparison of indices of
traumatic brain injury severity: Glasgow Coma Scale, length of coma and post-
traumatic amnesia. J Neurol Neurosurg Psychiatry. 79(6): 678-85.
12. Walker W, Ketchum J, Marwitz J, et al. (2010). A multicentre study on the clinical
utility of post-traumatic amnesia duration in predicting global outcome after moderate-
severe traumatic brain injury. J Neurol Neurosurg Psychiatry. 81(1): 87-9.
13. Ponsford J, Draper K, Schönberger M. (2008). Functional outcome 10 years after
traumatic brain injury: Its relationship with demographic, injury severity, cognitive and
emotional status. J Int Neuropsychol Soc. 14(02), 233-242.
14. Schönberger M, Ponsford J, Olver J, Ponsford M, Wirtz M. (2011). Prediction of
functional and employment outcome one year after Traumatic Brain Injury: A
Structural Equation Modelling approach. J Neurol Neurosurg Psychiatry. 82(8):936-
941.
15. Forrester G, Encel J, Geffen G. (1994). Measuring post-traumatic amnesia (PTA): an
historical review. Brain Inj. 8(2): 175-84.
16. Russell WR, Nathan P. (1946). Traumatic amnesia. Brain. 69(4): 280-300.
17. Jennett B. (1984). Management of head injury in adults. Injury. 16(3): 212.
18. Brown AW, Malec JF, Mandrekar J, et al. (2010). Predictive utility of weekly post-
traumatic amnesia assessments after brain injury: A multicentre analysis. Brain Inj.
24(3): 472-8.
Page 17 of 24
17
19. Nakase-Richardson R, Sherer M, Seel RT, et al. (2011). Utility of post-traumatic

amnesia in predicting 1-year productivity following traumatic brain injury: comparison
of the Russell and Mississippi PTA classification intervals. J Neurol Neurosurg
Psychiatry. 82(5): 494-9.
20. MacCallum RC, Zhang S, Preacher KJ, Rucker DD. (2002). On the practice of
dichotomization of quantitative variables. Psychol Methods. 7(1): 19.
21. Williams JS. (2011). Assessing the suitability of fractional polynomial methods in
health services research: a perspective on the categorization epidemic. J Health Serv
Res Policy. 16(3): 147-52.
22. Field, A. (2013). Discovering statistics using IBM SPSS. 4 th ed. Sage, London.
23. Osborne J.W. (2015) Best practices in logistic regression. Sage: Thousand Oaks,
California.
24. Shores E, Marosszeky J, Sandanam J, Batchelor J. (1986). Preliminary validation of a
clinical scale for measuring the duration of post-traumatic amnesia. Med J Australia.
144(11): 569-72.
25. Salford Systems. (2013). CART: Version 7. San Diego, California.
26. Breiman L, Friedman J, Olshen RA, Stone CJ. (1984). Classification and regression
trees. Belmont, California: Wadsworth.
27. StataCorp. (2011). Stata Statistical Software: Release 12. College Station , Texas:
StataCorp LP.
28. Goldman L, Weinberg M, Weisberg M, et al. (1982). A computer-derived protocol to
aid in the diagnosis of emergency room patients with acute chest pain. New Engl J Med.
307(10): 588-96.
29. Hanley JA, McNeil BJ. (1982). The meaning and use of the area under a Receiver
Operating Characteristic (ROC) curve. Radiology. 143(1): 29-36.
30. De Long ER, De Long DM, Clarke-Pearson DL. (1988). Comparing the area under two
or more correlated receiver operating characteristic curves: a nonparametric approach.
Biometrics. 44, 837-845.
31. Chatterjee S, Hadi A.S. (2012). Regression analysis by example. 5th ed. Wiley,
Hoboken, New Jersey.
32. Bayley MT, Tate R, Douglas JM, et al. (2014). INCOG guidelines for cognitive
rehabilitation following traumatic brain injury: methods and overview. J Head Trauma
Rehabil. 29(4): 290-306.
Nonparametric comparisons of AUC comparing each PTA classification with one-another in

the training and hold-out datasets. Superscript next to each classification systems represent
those systems that performed significantly more poorly, based on AUC.
Graph displays the predicted probabilities obtained through logistic regressions in the hold-
out dataset. AUC for each classification system is also included.
Table 1. Demographic and Injury Characteristics of Participants

Variable Nonproductive a Productive Total sample
M (SD), range M (SD), range M (SD), range
Page 18 of 24
18
Age, years 33.9 (14.71), 16-76 28.3 (11.14), 16-78 30.7 (13.11), 16 - 78
23.8 (26.08), .001 -
Duration of PTA, days 32.9 (32.87), .02-183 16.8 (16.05), .001-120
183
% (n) % (n) % (n)
Sex
Male 73.6 (335) 77.1 (452) 75.6 (787)
Female 26.4 (120) 22.9 (134) 24.4 (254)
Education
< 12 years of education 61.2 (259) 46.1 (255) 52.7 (514)
> 12 years of education 38.8 (164) 53.9 (298) 47.3 (462)
Marital status
Married/defacto 42.3 (190) 30.8 (178) 35.9 (368)

Widowed/divorced/Never
57.7 (259) 69.2 (399) 64.1 (658)
married/single/separated
Living location
Metropolitan 57.2 (258) 62.4 (364) 60.2 (622)
Country/interstate 42.7 (193) 37.6 (219) 39.8 (412)
Cause of injury
Car occupant 62.9 (286) 57.0 (333) 59.6 (619)
Pedestrian 12.3 (56) 13.7 (80) 13.1 (136)
Motorcycle 11.0 (50) 14.2 (83) 12.8 (133)
Bicycle 2.0 (9) 5.5 (32) 3.9 (41)
Fall 4.0 (18) 3.6 (21) 3.8 (39)
Work related 5.7 (26) 4.5 (26) 5.0 (52)
Other 2.2 (10) 1.6 (9) 1.9 (19)
CT Scan
Abnormal 85.6 (387) 83.6 (485) 84.5 (872)
Normal 13.3 (60) 15.2 (88) 14.3 (148)
Not performed 1.1 (5) 1.2 (7) 1.2 (12)

Injuriesb
Back 29.6 (134) 24.6 (143) 26.8 (277)
Chest 42.1 (190) 34.1 (198) 37.6 (388)
Abdomen 23.7 (107) 20.0 (116) 21.6 (223)
Limb 57.2 (259) 51.3 (298) 53.9 (557)
Facial 34.4 (155) 35.2 (205) 34.8 (360)
a
Note. Nonproductive and productive status refers to participant productivity status at one
year following TBI.
b
Refers to individuals who have moderate or severe physical injuries defined as injuries
resulting in a fracture or requiring surgery.
Table 2. PTA classification frequencies for total sample and as a percentage of those non-
productive at 12 months following injury from the current sample
Duration of PTA Sample a Non-productive at 12 months b, c
n % N %
Derived from literature
Russell/Jennett
<1 41 3.9 13 31.2
1-7 264 25.4 79 29.9
>7 – 28 447 42.9 177 39.6
> 28 289 27.8 186 64.4
Walker
0-56 949 91.2 379 39.9
>56 days 92 8.8 76 82.6
Brown (2005)
0-48 912 87.6 354 38.8
> 48 days 129 12.4 101 78.3
Page 19 of 24
19
Brown (2010)
0-28 752 72.2 269 35.8
>28 days 289 27.8 186 64.4
Mississippi
0-14 516 49.6 172 33.3
15-28 236 22.7 97 41.1
29-70 230 22.1 131 57.0
>70 59 5.7 55 93.2
Collapsed Mississippi
0 – 14 516 49.6 172 33.3

15 – 28 236 22.7 97 41.1
>28 289 27.8 186 64.4
Statistically-derived
CART
< 30.5 768 73.8 275 35.8
>30.5 273 26.2 180 65.9
a
Presents percentage of individuals that are categorised into PTA bands depending on
classification system.
b
Presents percentage of individuals non-productive at 12-months following injury within each
categorisation band.
c
Definition of non-productivity and productivity used in the current study differed in relation to
previous studies.
PTA classification systems derived from literature or through statistical methods.

Russell/Jennett classification derived from Russell and Jennett16 ; Walker classification obtained
from Walker et al.12 ; Brown 2005 classification obtained from Brown et al. 10 ; Brown (2010)
classification obtained from Brown et al.18 ; Mississippi and Collapsed Mississippi classifications
obtained from Nakase-Richardson et al.19
Table 3. Classification model statistics for training and hold-out datasets

Learning Hold-out
Sensitivity Specificity AIC BIC Sensitivity Specificity AIC BIC
(% ) (% ) (% ) (% )
Continuous 39.4 84.2 994.1 1003.40 37.1 88.4 328.3 335.38
PTA
Russell/Jennett 39.4 84.2 1025.3 1034.63 45.7 77.4 341.4 348.48
Walker 16.3 97.9 1024.4 1033.67 18.1 95.5 342.1 349.18
Brown (2005) 20.6 95.8 1025.5 1034.78 27.6 93.6 332.9 339.99
Brown (2010) 39.4 84.2 1022.31 1031.63 45.7 77.4 339.5 346.59
Mississippi 39.4 84.2 1012.0 1021.37 45.7 77.4 331.5 338.64
Collapsed 39.4 84.2 1024.0 1033.31 45.7 77.4 337.9 345.01

Mississippi
CART 38.0 86.1 1017.6 1026.90 44.8 78.7 338.7 345.83
Note.AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion

20
Page 20 of 24
21
22
Page 22 of 24
23
24
Page 24 of 24

Page 1 of 24: Number of Words (Abstract Until References) : 3523 Number of Figures: 2 Number of Tables: 3

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Page 1 of 24: Number of Words (Abstract Until References) : 3523 Number of Figures: 2 Number of Tables: 3

Uploaded by

Copyright:

Available Formats

Page 1 of 24

Running Head: Post traumatic amnesia and outcome prediction

Gershon Spitz, BA (Hons), PhD

Dean McKenzie BA (Hons), PhD

Correspondence to: Jennie L Ponsford, BA (Hons), MA (Clin Neuropsych), PhD

Number of words (Abstract Until References): 3523

best predict functional or productivity outcomes. We conducted a cohort study of 1041

management of patients following TBI.

Keywords: traumatic brain injury, post traumatic amnesia, outcome prediction

reduce the likelihood of return to functional independence, employment, study, leisure

and long-term planning.

Measurement of injury severity is vital following TBI as it is considered to be one of

traumatic amnesia (PTA).2, 3

standard measure, associated with in-hospital mortality and functional outcome.4

Duration of PTA has emerged as another strong measure of injury severity,

intracranial haemorrhage, raised intracranial pressure, residual neurological deficits, extent of

neuropathology, as well as with longer-term functional outcomes and return to employment. 6-

support for PTA as a stronger predictor of longer-term functional outcome, return to

more variance in outcome than socio-demographic factors.13, 14

The measurement of PTA has evolved substantially, from retrospective estimates

to lay down new memories.16

Several different classification systems have been proposed in an attempt to

defined as being a full-time or part-time student, competitively employed or taking care of

Categorisation of continuous, or dimensional, variables such as duration of PTA is

commonly performed in medical research (e.g. diabetes, depression, BMI, waist-hip-ratio,

power. Alternatively, in circumstances involving possible confounding variables such as age,

study of invitation to an outpatient cardiac rehabilitation program by age, showed very

use of categorisation, it is still often employed in clinical research and practice.22, 23

approach based upon standard logistic regression.

no-fault accident compensation system administered by the Transport Accident Commission

greater than 16 years, and pre-injury engagement in productive employment or study.

Exclusion criteria included no recorded productivity status at either premorbid or one-year

post-accident time-points, no recorded duration of PTA, permanent confusional or amnestic

state, or not being engaged in productive activities prior to injury.

Patient demographic and injury-related information (gender, age at injury, number of

up for the longitudinal study.

Assessment of Productivity Status

Participants pre-morbidly unemployed or not currently in the labour force were

classified as non-productive. Participants who were classified as apprentices, full-time or

part-time employees, or as secondary or tertiary students were classified as productive. One-

year productivity status was captured on a structured outcome questionnaire. Individuals

conditioning or retraining, unemployed, or non-vocational were categorised as non-

productive. Participants were considered as productive if they were employed in their

previous position or an alternative position full time or part-time.

Assessment of PTA Duration

time, as well as anterograde memory. The WPTAS was administered daily by a

correctly on three consecutive days, or otherwise cleared by a neuropsychologist.

PTA classification systems

Russell/Jennett,17 Walker,12 Brown,10 Brown,18 Mississippi,19 and Collapsed Mississippi.19

continuous/dimensional form. These methods are further explained below.

analysis of PTA data by Brown et al.10

premorbid or one-year productivity status, 25 had no recorded duration of PTA, and 3

post-injury and 586 (56.3%) as productive.

classification systems showed a greater likelihood of being non-productive at 12 months post-

injury with greater length of PTA.

Classification systems with a greater number of categories, including the Russell/Jennett,

Russell/Jennett classification, which displayed a poorer AUC. Dimensional PTA achieved a

between the learning and hold-out samples.

The relationship between the probability of being non-productive at one-year

may be imposed by categorisation of PTA. For example, leaving PTA as a dimensional

probability of being non-productive of 43% to 70%, respectively. Conversely, a PTA

the single cut-off CART decision rule.