Discourse recovery after severe traumatic brain injury: exploring the first year

Abstract

Objectives:

Although much is known about discourse impairment, little is known about discourse recovery after severe traumatic brain injury (TBI). This paper explores discourse recovery across the critical first year, controlling for pre-injury, injury and post-injury variables.

Design and methods:

An inception cohort comprising 57 participants with severe TBI was examined at 3, 6, 9 and 12 months post-injury and compared to a cross-section of matched healthy control participants. A narrative discourse task was analyzed with main concept analysis (MCA). A mixed linear model approach was used to track recovery controlling for pre-injury, injury and post-injury variables.

Results:

An upward trajectory of recovery was observed, with peak periods of improvement between 3–6 and 9–12 months and all time points were significantly below controls. Years of education and PTA duration were significant covariates in the recovery model. Presence of aphasia also influenced the recovery model.

Conclusions:

Individuals with TBI typically improve over the first year, however many will continue to have discourse deficits at 12 months. Years of education, PTA duration and aphasia should be considered when planning services. The 3–6- and 9–12-month periods may offer optimal periods for discourse recovery and increased supports may be beneficial between 6-9 months.

There is currently very little research detailing the nature of discourse recovery across the first year following a severe traumatic brain injury (TBI) (1). There are several reasons why we need to strengthen the existing body of research into discourse recovery and why we need to focus on the first year in particular. The first year following a TBI is a crucial stage for recovery. It is during this period that individuals typically receive the bulk of their rehabilitation (2-4). Individuals with TBI and their families often face significant personal and emotional challenges during the first year after their injury. Interventions and supports for discourse difficulties can ease these challenges (5-7), yet knowing when is the right time for this therapeutic support remains unclear. By examining the trajectory of discourse recovery, we can obtain a better understanding of the optimal timing for these interventions, which will subsequently improve the cost-effectiveness of therapy services, an important consideration for health services (8-10). The first 12 months is additionally a time when many individuals are reintegrating back into the community (11). The majority of individuals who have a TBI typically experience challenges with reintegration into at least one or more of their premorbid everyday, social or work roles (12). The first year is additionally the foundation for long-term recovery. Maximizing recovery during this first year may promote long-term improvements (13) and minimize chances of decline (14,15). Furthermore, exploring recovery across the first year may offer insights into prognosis. Examining critical factors such as injury variables and the relationship of these variables to discourse recovery may aid predictions of real-world functioning. Exploring prognosis may therefore be of benefit for treatment planning such as providing clues around the best timing of rehabilitation and predicting future support needs. It is important that there is a strong body of evidence around the first year of recovery to ensure that individuals with TBI and their families are getting the support that they require at the right times, to maximize the cost-effectiveness of rehabilitation programs, to aid prognosis and to support effective timing of rehabilitation.

Only four studies have explored discourse recovery in detail. The first of these studies conducted by Snow, Douglas and Ponsford (16) examined a group of 26 individuals who had sustained a severe TBI. The participants were initially evaluated between 3 and 6 months post-injury and were reviewed at a minimum of 2 years post-injury using a conversational assessment. Results indicated no significant improvements in the group over time; however, there was a sub-group that did improve and this was characterized by greater severity of injury and longer period of Speech Pathology intervention. This study also reported that approximately a quarter of the variance in discourse skills at 2 years could be predicted by conversational performance at 3–6 months (16). A further study by Snow, Douglas and Ponsford (17) utilized the same 26 participant cohort as described above and also examined recovery across the same period. However, the focus of this study was not on conversation but on narrative discourse ability, which was evaluated with the use of a monologic picture description task. The key finding was that discourse skills recovered to within normal limits; when compared to controls. This study also explored associations between narrative skills and injury severity, however this was not significant. It is possible the picture description discourse measure selected in this study may not have been sufficiently challenging to identify discourse deficits (17). Another key discourse recovery study utilized a longitudinal design to describe and track changes in the spoken discourse of two participants who had severe TBI using a narrative retell task and a control comparison group (18). The first participant was reviewed on six occasions between five and nine months post-injury. This participant showed improved ability to communicate the main ideas of the narrative by nine months but had ongoing difficulty with cohesion. The second participant was evaluated between 3.5-6.5 months post-injury. This participant showed persisting discourse difficulties with content and cohesion. Although some improvements were noted in cohesion between the first and final evaluations, the scores were still well below controls. This study also noted a potential prognostic relationship between initial appropriateness of verbal output and later ability to express complex relationships through language (18).

The final notable study, compared a group of 46 individuals with severe TBI between three and six months post-injury with 37 healthy controls using a procedural discourse task (19). This study found improvements in the production of relevant content or macro-structural elements of discourse, but not on productivity measures between the two time points. Dysarthria was identified as a potential factor of the significantly different speech rates found between individuals with TBI and controls (19).

This small body of literature shows a pattern of discourse improvement for some, but not all, individuals in the first couple of years following injury and limited exploration of the factors influencing discourse recovery (16-18). With regards to study design, only one of these studies used multiple data points across the first year to adequately describe recovery and this was limited by a very small sample size (18). Furthermore, variables influencing discourse recovery such as injury severity were mostly reported qualitatively and were not factored into the recovery process. These seminal studies have revealed some of the key challenges with measuring discourse recovery over time such as needing to consider the impact of communication partner on dialogic discourse tasks (16), using tasks that are sufficiently challenging to measure subtle changes over time (17) and the need to consider a range of pre-injury, injury and post-injury factors when examining recovery. The use of narrative tasks may address several of these issues, as they are inherently monologic in nature and elicit complex language structures (20-22) while the use of modern statistical modeling (23,24) may enable improved clarification of the influence of pre-injury, injury and post-injury factors on discourse recovery. It is clear that there is a lack of information regarding discourse recovery, particularly across the first year. In order to strengthen this body of research we need to consider study design as well as optimal discourse measures for the study of recovery.

MCA is a tool that shows promise for identifying subtle difficulties in narratives of people with TBI over time (25,26). MCA evaluates the person’s ability to convey a general gist or idea (25,27), considering both accuracy and completeness. Unlike many of the existing discourse analyses that may focus on one or two discourse features (28), MCA can offer insights into the micro-linguistic, micro-structural, macro-structural and super-structural elements of discourse (29). For example, micro-structural difficulties may include incorrect use of pronoun referents, macro-linguistic features may include reduced coherence while super-structural difficulties may include poor chronological sequencing. It is of benefit to have an analysis that offers multiple insights due to the heterogenous nature of the discourse patterns reported in TBI populations (30).

Although occurring less frequently than social, cognitive and discourse impairments, aphasia is among the cognitive-communication disorders that result from a TBI and it can impact on discourse performance. Aphasia typically occurs in a relatively small proportion of individuals with TBI, with incidence ranging from 2% to 32% (31,32). The impact of aphasia on discourse competence in TBI is largely unknown and this is complicated by the unique cognitive-linguistic profile observed following TBI (33,34). The Western Aphasia Battery-Revised (WAB-R) (35) is the most widely used and recommended standardized measure for assessment of aphasia in TBI (36-39). Judicious use of the WAB-R, alongside additional cognitive and discourse measures may assist our understanding of discourse competence and recovery following TBI.

Two primary areas of discourse recovery in the first year that require exploration are the patterns of recovery and the influence of pre-injury, injury and post-injury variables. These need to be examined within the context of a strong longitudinal research design, using sufficient sample sizes and using appropriate discourse measures. Research into discourse recovery provides an opportunity to identify optimal phases for treatment as well as factors that influence discourse performance. Such information is beneficial for rehabilitation services and clinicians to guide and support the individual and their close others along this important stage of recovery.

Aims

In a longitudinal cohort of individuals with severe TBI:

1. Describe the changes in discourse performance using three monthly intervals over the first year post-injury in relation to a matched control group.

2. Explore the influence of pre-injury, injury and post-injury variables on the recovery model.

Method

Study design

This study utilized a longitudinal design with an inception cohort and a cross-sectional control group comparison. Participants with TBI were initially reviewed at 3 months post-injury and were then followed up at 6, 9 and 12 months post-injury. Participants were reviewed as close as possible to each data point; however, a 1-month margin was used to allow for unforseen circumstances. The mean days post-injury for each time point were as follows: 3 months (M = 96.71, SD = 20.36), 6 months (M = 196.43, SD = 18.15), 9 months (M = 291.93, SD = 11.64) and 12 months (M = 383.48, SD = 14.44).

Participants

Recruitment

Participants with TBI were recruited through three specialist metropolitan brain injury rehabilitation services in New South Wales, Australia between June 2011 and March 2013. A representative from each brain injury service screened all admissions to the brain injury rehabilitation services during the recruitment period against the inclusion/exclusion criteria.

Inclusion/exclusion criteria

The inclusion criterion comprised of: (a) severe TBI determined by a Glasgow Coma Scale score less than eight (40,41) and/or post-traumatic amnesia (PTA) duration of greater than 24 h (40,42); (b) participant aged between 16 and 65 years at the time of the injury; (c) cleared of PTA by time of assessment; (d) medically stable; (e) proficient English speaker (f) residing within Sydney metropolitan area or within 3h traveling distance. Individuals were excluded in the following circumstances: (a) where consent was unable to be obtained from the person with TBI or a significant other; (b) more than 7 months post-injury at time of initial assessment; (c) history of previous neurological illness or injury or significant medical history such as developmental delay; (d) persisting PTA; (e) unable to be followed up for at least one appointment.

Participant characteristics

The total number of initial referrals was 79. Of these 79 initial referrals seven were unable to be contacted; six had time limitations; three indicated no interest; three were excluded at the request of their rehabilitation teams and three had medical complications. This resulted in a final cohort of 57 participants. Ten of the participants were recruited at 6 months post-injury.

The average age for the participants was 35.25 years (SD = 13.11) with a range between 16 and 66 years. The majority of the sample were male (80%). Average years of education was 13.58 years (SD = 2.99) with a range between 8-20 years. The average PTA duration was almost 2 months (M = 52.89; SD = 40.02) with a range of 6–215 days. Initial GCS scores averaged 6.82 (SD = 3.47) with scores ranging from 3–15. TBI cause was mostly motor vehicle accidents (n = 38), followed by falls (n = 12) then assaults (n = 5) and other causes (n = 2). A checklist was used to screen for hearing or visual impairments. While all participants had adequate hearing for the purposes of the assessment, nine participants reported a mild difficulty in this area. One participant had a significant visual impairment but was still able to complete the task without pictorial support. The scores for this participant were of a similar pattern to the cohort and were therefore included in the analysis. A further nine participants reported mild visual difficulties, however these did not impact on the assessment. Five participants identified English as their second language, with primary languages including; Singhalese, French, Urdu, Arabic and Mandarin. One participant was multi-lingual and identified Indi as their primary language followed by Mandarin and then English. Reduced intelligibility at conversation level (Frenchay Dysarthria Assessment-2; Intelligibility Subtest; Item c; score < 7) (43) was identified in 6% of cases at 3 months, 12% at 6 months, 8% at 9 months and 2% at 12 months. The discourse samples of these individuals were all included in the study; however, these were reviewed multiple times and by a second speech pathologist to ensure accuracy of transcription.

Pre-injury employment data are provided in Table 1 while a summary of neuropsychological cognitive testing across 12 months is provided in Table 2. The Living Skills subtest from the Sydney Psychosocial Reintegration Scale-2 is presented in Table 3 to highlight relative independence and support needs for everyday living skills such as personal care and travel. The proportion of participants receiving active speech pathology rehabilitation was 85% at 3 months, 65% at 6 months, 51% at 9 months and 43% at 12 months.

Table 1.

Pre-injury employment data.

Occupational status		Workload			Occupational category
Employed	44 (77%)	Full-time	34 (76%)	Technician/Trade	12 (27%)	Sales	4 (9%)
Unemployed	9 (16%)	Part-time	2 (4%)	Professional	7 (16%)	Machinery	3 (7%)
On leave	1 (2%)	Casual 9	(20%)	Clerical/Admin	6 (13%)	Personal srvices	2 (4%)
Homemaker	3 (5%)			Labourer	5 (11%)	Student	1 (2%)
Student	1 (2%)			Managers	5 (11%)

Table 2.

Summary of neuropsychological cognitive testing (z scores).

Time point	Cognitive domain	Attention and processing speed				Memory		Executive function
	Test	DSF	DSB	Writ	Oral	HVLT	BVMT	Inhib	Switch	Maze	Jdge	Cat	WG
3 months n = 49	Mean	−0.40	−0.35	−2.18	−1.98	−2.12	−1.49	0.18	0.14	−1.21	−0.80	−0.92	−0.82
	(%a)	27	22	78	78	65	59	12	14	45	37	35	35
	Sub-total			−1.22 (57%)		−1.81 (67%)						−0.47 (16%)
	Totalb											−1.08 (45%)
	Mean	−0.24	−0.31	−1.97	−1.72	−2.14	−1.22	0.36	0.30	−0.95	−0.27	−0.71	−0.50
6 months n = 54	(%)	26	20	63	61	66	48	4	8	39	22	33	28
	Sub-total			−1.06 (43%)		−1.69 (59%)						−0.29 (13%)
	Total											−1.01 (41%)
	Mean	−0.01	−0.21	−1.38	−1.16	−1.48	−1.42	0.57	0.41	−0.76	−0.28	−0.42	−0.15
9 months n = 45	(%)	9	7	47	38	53	47	4	4	33	20	20	11
	Sub-total			−0.70 (27%)		−1.48 (47%)						−0.12 (4%)
	Total											−0.84 (22%)
	Mean	−0.25	−0.55	−1.59	−1.40	−1.58	−1.18	0.41	0.13	−0.66	−0.20	−0.14	−0.48
12 months n = 46	(%)	22	30	50	50	46	43	4	13	35	20	15	28
	Sub-total			−0.70 (37%)		−1.33 (43%)						−0.13 (11%)
	Total											−0.73 (24%)

^a% of participants identified as impaired (<1 SD)

^bCognitive Index; DSF = Digit span forwards; DSB = Digit span backwards; Writ = Written symbol digit modalities tests; Oral = Oral symbol digit modalities tests; HVLT = Hopkins verbal learning test-revised; BVMT = Brief visuospatial memory test-revised; Inhib = Inhibition score from Delis-K Executive Function System (D-KEFS) Colour-word Interference Test; Switch = Inhibition/Switching score from (D-KEFS) Colour-word Interference Test; Maze = Mazes subtest from Neuropsychological Assessment Battery (NAB); Jdge = Judgement subtest from NAB; Cat = Categories subtest from NAB; WG = Word generation subtest from NAB.

Table 3.

Sydney psychosocial reintegration scale-2 (SPRS) (living skills subtest^a).

Degree of supportb	3 months	6 months	9 months	12 months
High	15 (36%)	8 (19%)	4 (14%)	5 (14%)
Moderate	16 (38%)	9 (21%)	5 (17%)	2 (6%)
Low	10 (24%)	19 (45%)	14 (48%)	25 (69%)
Nil	1 (2%)	6 (14%)	6 (21%)	4 (11%)

^aRelative version is reported to minimize potential impact of reduced insight in self-reported outcomes

^bDegree of reported support required for everyday living skills: High = High support needs (sub-test score 0–7); Moderate = Moderate support needs (sub-test score 8–11); Low = Low support needs (sub-test score 12–15); Nil = Nil support needs/fully independent (sub-test score 16).

The distribution of the Western Aphasia Battery-Revised (WAB-R) Aphasia Quotient (AQ) scores at 3 months are observed in Figure 1, highlighting a continuum of linguistic skill and a sub-group of individuals with AQ scores below 90 (excluding the severe outlier). Table 4 provides detailed aphasia data on the group overall as well as identified subgroups. This data highlight a continuum of cognitive-linguistic impairment and shows that those with aphasia clearly have more severe cognitive-linguistic impairments compared to those without aphasia. Furthermore, a subgroup with an AQ below 90 demonstrated even more pronounced cognitive-linguistic impairment as well as frank aphasia characteristics such as perseveration. Furthermore, (4/7) 57% of the participants in this subgroup had persistent linguistic impairment at 12 months and the remaining 3/7 remained in the NABW range. There is clearly a difference in the discourse scores of each subgroup and between the groups with and without aphasia.

Figure 1.

Normal Q-Q plot for Aphasia Quatient (WAB-R). Note the sub-group with AQ

Table 4.

Summary of aphasia and related data (3 months).

Sub-group	n	Western Aphasia Battery-Revised					Discourse	BNT	Cognition			Aphasia % 12m b
AQ		SS	AVC	REP	NWF	AQ			APS	MEM	EF
<90a	7 (16%)	16.43	9.36	9.50	7.67	85.91	16.71	*27.33	−1.88	−2.64	−0.80	57%
90–93.79	16 (37%)	17.50	9.77	9.69	8.98	91.86	20.67	42.81	−1.41	−2.27	−0.70	0%
≥93.8	20 (47%)	18.85	9.92	9.90	9.67	96.68	35.05	50.32	−0.97	−1.21	−0.23	0%
All	43	17.95	9.77	9.76	9.09	93.13	26.86	44.02	−1.29	−1.85	−0.50	12%

^aExcluding one outlier with severe aphasia

^*Qualitative features: Perseveration (real and non-real words), non-word production and/or non-word phonemic paraphasia (e.g. icorn for unicorn), multiple real word phonemic paraphasia (e.g. draft for dart) or unrelated semantic paraphasia (e.g. scissors for clothes hanger), high frequency of related semantic paraphasia (e.g. boat for canoe).

^bBased on AQ cut-off < 93.8; however, all participants with aphasia at 3 months continued with a Not Aphasic By WAB presentation (Fromm, 2016) and none of these participants scored 100 at final review.

Control sample

The control sample comprised a cross-section of 57 healthy controls with no history of neurological disease or injury. Control transcripts were obtained through aphasiabank.org; which contains an online password-protected repository of discourse samples including control samples (38). Permission was granted to use the control data for this study. The average age for the control group was 32.14 years (SD = 13.98) with a range from 19 to 66 years. Males comprised 60% of this group. The mean number of years of education was 14.43 (SD = 1.54). The control group was matched to the TBI group on age (U = 1607.50, p = 0.923, r = −0.09), gender (X²(1, n = 114) = 1.21, p = 0.271) and education level (t(108) = −1.86, p = 0.065). The control participants were all born in the USA and all control participants spoke English as their primary language.

Procedure

Primary measure

The primary measure for this study was the Main Concept Analysis (MCA) (25,27) score obtained from a narrative retell of the story of Cinderella (38). The full elicitation instructions and stimuli for the Cinderella retell can be found at https://tbi.talkbank.org/ (44). Troubleshooting questions can be found at: https://aphasia.talkbank.org/protocol/ (45). The group with TBI completed the narrative retell at approximately 3, 6, 9 and 12 months post-injury. The control group comprised of a cross-sectional sample with a single data point only.

Elicitation and stimuli.

The elicitation for the Cinderella retell followed the protocol established by MacWhinney, Fromm, Forbes and Holland (38) and is outlined below:

1. Present Cinderella picture book (46), with the text covered.

2. ‘I’m going to ask you to tell a story. Have you ever heard the story of Cinderella?’ If yes continue to 3. If answer is no, ask participant to tell a fairy tale s/he knows.

3. ‘Do you remember much about it? These pictures might remind you of how it goes. Take a look at the pictures and then I’ll put the book away, and ask you to tell me the story in your own words.’

4. Allow participant to look through book; assisting with page turning if needed; and then, if necessary, prompt: ‘Now tell me as much of the story of Cinderella as you can. You can use what you know about the story, as well as the pictures you just looked at.’

5. If participant gives a response of fewer than three utterances, or seems to falter, allow 10 seconds, then prompt: ‘What happened next?’ or ‘Go on’.

6. Continue until participant concludes story or it is clear s/he has finished. If the participant is unable to generate a sufficient response, use troubleshooting question ‘Did Cinderella go to the ball and meet the prince?’.

Transcription.

Cinderella narratives were transcribed verbatim and separated into T-units. The T-unit separation helped with identification of the main concepts but was not critical to the analysis.

Main concept analysis.

MCA involves firstly identifying if an utterance corresponds with a main concept. For example, the utterance; ‘But she has to return by a certain time otherwise it’ll all turn into pumpkin, into a pumpkin’ corresponds with the main concept ‘She knew she had to be home by midnight because everything will turn back at midnight’. The Cinderella narrative has 34 main concepts that each contains one subject, one main verb and, in most cases, an object (25,27). Each main concept of the Cinderella narrative has essential content and acceptable variations as detailed in Richardson and Dalton (27). For example, ‘She knew she had to be home by midnight because everything will turn back at midnight’ (27) contains three essential components; (i) she, (ii) had to be, (iii) home by mid-night. Acceptable variations for each includes, respectively; (i) Cinderella; (ii) must be, needs to be, must return; (iii) leave by midnight. Alternatively, the participant could also state ‘the fairy godmother told her that if she wasn’t home by midnight [result] would happen’ (27).

The second stage of MCA involves rating the accuracy and completeness of each utterance and is focused on evaluating the person’s ability to convey the general ‘gist’ or idea (25,27). Five codes were used to rate each utterance as described in Table 5. Appendix 1 offers worked examples of the analysis under both control and TBI conditions. The control example offers a complete and fully correct analysis, highlighting all of the main concepts of the Cinderella narrative. Appendix 2 shows a case example of the analysis over time (27).

Table 5.

Main concept analysis coding and scoring descriptions.

Coding		Description	Score
Accurate-complete	AC	No essential elements of the main concepts are incorrect. The speaker must have produced every essential element of the main concept.	3
Accurate-incomplete	AI	No incorrect content but missing at least one essential element.	2
Inaccurate-complete	IC	Contained at least one incorrect piece of information but mentioned all essential elements.	2
Inaccurate-incomplete	II	Statement corresponds with a main concept but fails to include at least one essential element and contains at least one incorrect essential element.	1
Absent	AB	None of the essential information is given.	0

Reliability.

Inter-rater and intra-rater reliability was completed for the MCA. Sample size for reliability was set at 20% based on 95% confidence level and 15% margin of error. Samples were randomly selected using a computerized program. Inter-rater reliability was 80.1% with 100% agreement achieved through consensus. Intra-rater reliability was completed approximately 6 months post the initial ratings with 81.78% agreement.

Data analysis

Determining discourse impairment and severity

One standard deviation around the control mean was used to distinguish impaired from non-impaired performance. Mild impairment was one to two standard deviations below the control mean, moderate impairment was two to three standard deviations below the control mean and severe impairment was greater than three standard deviations below the control mean. One standard deviation was selected as the cut-off for impaired performance to ensure that subtle cognitive-communication difficulties were identified (47). Indeed, other studies have found one standard deviation to be a sensitive and accurate reflection of cognitive communication (47) and discourse impairment (48), with a two standard deviation cut-offs possibly resulting in Type II errors (48). Furthermore, it is uncommon for individuals with severe TBI to have an absence of cognitive-communication difficulties (32,49), minimizing the risk of Type I errors. A continuum of discourse impairment has also been reported in other discourse studies (48) yet there have been no previous attempts to categorize severity of discourse impairment. The mild, moderate and severe descriptions in the present study are clearly defined; using well-known standard deviation measures; to aid description along this continuum.

Statistical analysis

Statistics were completed using IBM SPSS version 24. Descriptive statistics were firstly used to describe the nature of the discourse data. Next, to address Aim 1, a mixed linear models full information maximum likelihood (FIML) analysis was completed (23). The data were normally distributed, linear and homoscedastic, meeting the assumptions for the mixed model (23). This baseline model examined the dependent variable, MCA score, using the four time points (3, 6, 9, 12 months) as the repeated measures factor while covarying for Age, Education, PTA duration in a mixed linear models full information maximum likelihood (FIML) analysis. To address the control comparison within Aim 1, independent t-tests were used to compare the difference between the single control group data point and the TBI group at each data point. Descriptive statistics were also utilized to identify potential recovery sub-groups relating to Aim 1.

Aim 2 was firstly addressed by examining the pre-injury and injury covariates in the baseline model as described above. Only covariates that showed significant correlations with the dependent variable were included in the analysis. Thus, age, education years and PTA duration were included, while GCS score and gender were excluded. The post-injury components of Aim 2 were analyzed with two further models. Two therapy conditions were explored in the first model. ‘Therapy minimal’ was defined as therapy across 1–2 time periods while ‘therapy maximal’ referred to therapy across 3–4 time periods. A two-condition (therapy minimal and therapy maximal) between group × four-condition (time: 3, 6, 9, 12 months) within group model covarying for Age, Education, and PTA duration, was tested using a mixed linear models full information maximum likelihood (FIML) analysis to compare differences between therapy conditions on the MCA score. The second model incorporated the condition of presence or absence of aphasia. An aphasia quotient score equal to or above 93.8 on the WAB-R was considered non-aphasic (35). A two-condition (aphasia and no aphasia) between group × four-condition (time: 3, 6, 9, 12 months) within group model covarying for Age, Education, PTA duration, was tested using a mixed linear models full information maximum likelihood (FIML) analysis to compare differences in recovery between individuals with aphasia or without aphasia at 3 months. Likelihood-ratio chi-squared tests were used to compare the baseline and nested models. Alpha was set at p = 0.050 with a least significant difference adjustment applied for multiple comparisons

Reliable change

Clinically significant change was calculated with the Morley and Dowzer (50) Leeds Reliable Change Indicator based on the original work of Jacobson, Roberts, Berns and McGlinchey (51). Clinically significant change was also considered if a participant moved into or out of the normal range (52).

Ethical approval

This study was part of a larger longitudinal project (NHMRC #632681) examining recovery beyond 2 years post-injury. The project was approved by the Australian National Human Research Ethics Committee.

Results

Description of discourse data

Overall discourse scores

The mean MCA score for the control group was 64.56 (SD = 19.84). The mean MCA scores for the group with TBI were 26.16 (SD = 17.53) at 3 months, 29.96 (SD = 18.02) at 6 months, 30.78 (SD = 18.45) at 9 months and 36.35 (SD = 19.03) at 1 year.

Frequency and severity of discourse impairments

Discourse impairments were identified in 41 out of 48 (85%) participants assessed at 3 months post-injury, with eight of these participants falling within a severe range. Discourse deficits remained prevalent at 6 and 9 months, evident in 43 out of 54 (80%) participants at 6 months and 37 out of 46 (80%) participants at 9 months. By 1 year, 31 out of 47 (66%) participants continued to have discourse deficits with two participants remaining in the severe range. Further details are provided in Table 6.

Table 6.

Frequency and severity of discourse impairments.

Severity			Data point
Description	Score range	SD below control mean	3 months	6 months	9 months	12 months
WNL	45–80	Within 1SD	7	11	9	16
Mild	25–44	1SD-2SD	16	20	22	14
Moderate	6–24	2SD-3SD	17	19	13	15
Severe	0–5	3SD-4SD	8	4	2	2

Discourse patterns

Participants who scored within the severe range (0–5) typically produced one accurate-complete main concept and omitted over 30 main elements. In the moderate range (6-24) the typical discourse pattern included between one to five accurate-complete main concepts, the omission of 20–30 main elements and up to seven inaccurate and/or incomplete errors. Participants in the mild range (25-44) tended to produce between 5 and 10 accurate-complete main concepts with 15–20 absent elements and up to 12 inaccurate and/or incomplete elements. Participants who scored within the normal range (45+) demonstrated a pattern of 10–20 accurate-complete elements and 10–20 absent elements with up to 10 inaccurate and/or incomplete concepts. The control participants typically produced 15–30 accurate-complete concepts, with up to 15 elements omitted and up to four inaccurate and/or incomplete errors. Further ratings for each severity level are outlined in Table 7.

Table 7.

Discourse patterns.

Severity			Score range
Description	Score range	SD below control mean	AC	IC/ AI	II	AB
Controls	45+		15–30	0–3	0–3	0–15
WNL	45–80	Within 1SD	10–20	0–7	1–3	10–20
Mild	25–44	1SD-2SD	5–10	3–6	0–6	15–20
Moderate	6–24	2SD-3SD	1–5	1–3	1–4	20–30
Severe	0–5	3SD-4SD	1	3	0	30 +

Discourse recovery (aim 1)

Baseline recovery model

The baseline linear mixed model analysis revealed a statistically significant main effect of time [F(3, 135.97) = 18.40, p < 0.001, η_p² = 0.09]. Post hoc pairwise comparisons, with least significant difference adjustment, indicated lower scores at 3 months compared to 6 months (p < 0.001, d = 0.45) and lower scores at 9 months compared to 1 year (p = 0.031, d = 0.24). The change between 6 and 9 months was not significant (p = 0.284, d = 0.11). Covariates analyzed in this model are discussed below in relation to Aim 2. The recovery trajectory is illustrated in Model 4.

Control comparison

The group with TBI had statistically significant lower scores on the discourse measure when compared to the control group at 3 months [t(100) = −10.34, p < 0.001, d = 2.35], 6 months [t(109) = −9.69, p < .001, d = 1.99, 9 months [t(100) = −8.90, p < 0.001, d = 1.87] and 12 months [t(101) = −7.36, p < 0.001, d = 1.67]. The control data are also included in Figure 2.

Figure 2.

Trajectory of recovery across 1 year. Note the control data were taken at a single time point only but are charted across the graph to aid comparison. The shaded region shows the standard error of the control mean.

Discourse recovery sub-groups

Further analysis of the recovery data revealed two overall differing patterns of recovery. The first sub-group, comprising almost one-third of the sample (n = 18) recovered at a relatively equal rate to the other group between 3 and 9 months but then showed a faster rate of recovery between 9 and 12 months. Most of these participants were initially within a mild range and improved to within normal limits by 12 months. This sub-group included 11 participants who showed a clinically significant change. The remaining seven participants in this sub-group showed a similar recovery pattern as described above and while the scores did not reach a clinically significant level, many were approaching this level. In this group, 6/18 (33%) participants were identified with aphasia by WAB-R, however only 1/18 (6%) participants had an aphasia quotient score below 90. The second group (n = 37) recovered at a stable rate with a relatively slower rate of recovery compared to sub-group 1 between 9 and 12 months. This group typically started at moderate and improved to a mild range. In this second group, 18/39 (46%) participants were initially diagnosed with aphasia by WAB-R, with 8/39 (21%) scoring below 90. The two patterns are illustrated in Model 3. One participant showed a clinically significant decline, moving from normal to mild impairment.

Factors influencing recovery (aim 2)

Pre-injury and injury-related variables

The baseline linear mixed model analysis revealed a statistically significant main effect of time [F(3, 135.97) = 18.4, p < 0.001, η_p² = 0.09], with years of education [F(1, 56.16) = 4.71, p = 0.034, η_p² = 0.09] and post-traumatic amnesia duration [F(1, 57) = 26.89, p ≤ 0.001, η_p² = 0.28] emerging as significant covariates in the model. Age was not a significant covariate in this model [F(1, 55.32) = 2.35, p = 0.131, η_p² = 0.02].

Post-injury variables

Including treatment amount as a main effect, did not alter the recovery model [χ²(1) = 0.06, p = 0.806, V = 0.02]. Likewise, the main effect of treatment amount was not significant [F(1, 55.43) = 0.01, p = 0.918, η_p² = 0.01], while the main effect of time remained significant [F(3, 135.95) = 18.34, p ≤ 0.001, η_p² = 0.09]. Years of education [F(1, 55.95) = 4.19, p = 0.045, η_p² = 0.09] and post-traumatic amnesia duration [F(1, 56.4) = 22.19, p ≤ 0.001, η_p² = 0.24] also continued to be significant covariates in the model.

Analysing aphasia as a main effect significantly altered the recovery model [χ²(1) = 7.24, p = 0.007, V = 0.21]. The main effect of aphasia however, was not significant [F(1, 153.90) = 0.08, p = 0.773, η_p² = 0.00] and, although the main effect of time remained significant, the F value was reduced [F(3, 128.25) = 11.16, p ≤ 0.001, η_p² = 0.05]. In this model, post-traumatic amnesia duration remained a significant covariate [F(1, 62) = 24.29, p ≤ 0.001, η_p² = 0.21] however, years of education was no longer a significant covariate [F(1, 55.23) = 3.42, p = 0.07, η_p² = 0.07]. The models are visually shown in Appendix 3.

Discussion

The aim of the present study was to explore discourse recovery across the first year following severe Traumatic Brain Injury to address the lack of empirical evidence on this topic. Four data points were used to examine changes in discourse performance and this was compared to a matched control sample. The influences of pre-injury, injury and post-injury variables on the recovery model were also investigated.

Discourse recovery

Results of the linear mixed model revealed an overall significant improvement in discourse skills between three and 12 months following severe traumatic brain injury however comparisons with the control sample showed that the group with TBI remained significantly below the controls across the 12-month period. The overall change reflected an improvement from conveying 1–5 accurate-complete elements (moderate) to conveying between 5 and 10 accurate-complete concepts (mild). This is in contrast to the 15–30 accurate-complete concepts produced by the control group. Improvements in discourse have been reported in other single case designs (18) and longer-term studies (17) but this is the first study to report these findings in a cohort with TBI using multiple intervals across the first year. Such findings reinforce the need for Speech-Language Pathologists to have an active role in supporting discourse for at least the first year following injury, in line with the INCOG Guidelines (39).

The model also showed peak periods for change between the 3–6- and the 9–12-month data points with the 6–9-month period showing a slower rate of change. A possible explanation for this 6–9-month dip was a potential reduction in therapy during this time. Inclusion of treatment as a blocking variable however, did not impact on the recovery model and therefore did not adequately explain this pattern. Further research exploring treatment in detail may reveal further insights into this pattern. Varied rates of change have been reported in other single-case discourse studies during the first year (18,53); however, this is the first study to report this finding in a group cohort with relatively equal spacing of intervals. This finding has potential implications for the timing of rehabilitation. For example, clinicians may choose to capitalize on the peak periods of recovery by offering increased therapy during those periods. Alternatively, speech pathologists might offer extra supports during the 6–9-month period to help maintain a pattern of improvement.

Discourse recovery sub-groups

Two overall sub-groups were identified with regards to discourse recovery. One third of the participants in the cohort were identified as belonging to an ‘Improved to Within Normal Limits (WNL)’ sub-group. On average, participants in this group improved from a mild discourse impairment to within a normal range by 12 months. This sub-group was characterized by higher initial scores and faster recovery rate from 9 to 12 months. This sub-group was additionally characterized by higher education levels, shorter post-traumatic amnesia duration and relative absence of aphasia at 3 months. In contrast, approximately two-thirds of the cohort were described as ‘Slow to recover’. Participants in this sub-group typically moved from a moderate to a mild profile by 12 months and while this group showed improvement, the rate of improvement was relatively slower between 9 and 12 months compared to the ‘Improved to WNL’ sub-group. This sub-group had lower initial scores in addition to lower education levels, longer post-traumatic amnesia duration and higher incidence of aphasia at 3 months. The aforementioned pre-injury, injury related and post-injury variables were not significantly different in this small sub-group comparison and larger sample sizes would be required to strengthen these findings.

This is the first study to explore sub-groups using multiple data points across the first year and these relatively smaller time intervals have offered new insights into the nature of recovery. One other discourse study has identified a sub-group of participants who improved however this was within a cohort that, overall, did not show change (16). Results from the present study suggest that the 9–12-month period may be a critical stage for recovery, as this was the point at which the sub-groups recovery trajectories departed. The faster rate of recovery in the ‘Improved to WNL’ sub-group during the 9–12-month stage could possibly be related to improvements in insight and awareness in this group that facilitated improved recovery. For example, participants in this group may have had increased capacity to engage in metacognitive discourse activities. Indeed, another study of the present cohort showed the 12-month point as critical to changes in awareness and insight (54) lending credibility to this explanation. These findings strengthen the need for Speech Pathologists to be actively involved in managing discourse during the first year. Also, supporting awareness and insight around discourse deficits might be particularly helpful for mild discourse impairments between 9 and 12 months, although further research is required in this area.

Other findings

There were several other noteworthy findings from this study. Firstly, discourse impairments were highly prevalent across the first year in this cohort of individuals with severe TBI. Discourse impairments were evident in a very high proportion of the cohort at 3 months (85%) and were still identified in a large proportion of the cohort (66%) at 12 months. These findings add to a growing body of prevalence studies that indicate high prevalence rates of discourse deficits after TBI (48,55,56).

Secondly, one participant had a clinically significant decline, moving from within a normal range to showing mild discourse deficits. Decline has been reported in another discourse study (17). This finding is of interest as this participant could have potentially been discharged in the early stages of recovery based on the findings of a relatively normal performance between 3 and 6 months. Another, related finding was that the small group of TBI participants who scored within normal limits, nonetheless showed a slightly different pattern to the control group performance, despite achieving scores within the same range. This group appeared to produce less accurate-complete utterances than controls but typically produced more inaccurate or incomplete sentences than the controls, which nevertheless resulted in a similar overall score. These findings suggest that there are a group of individuals who perform similarly to controls but may still not be communicating with the same efficiency and clarity as controls. This group of individuals appear to have subtle difficulties with efficiency and clarity that may not necessarily impact on everyday conversations with family but are likely to significantly impact on successful work reintegration, where efficiency and clarity of communication are paramount. These findings suggest the need for Speech Pathologists to continue to monitor all patients, even those who perform within a normal range at least up to a year and possibly beyond. Assessment for this group may be strengthened by evaluation of discourse patterns in addition to assessment tasks that require more sustained communication effort and complexity such as the Functional Assessment of Verbal Reasoning and Executive Strategies (FAVRES) (47).

Finally, the results from this study suggest that MCA of the Cinderella narrative is a useful tool for evaluating discourse. This is also the first study to offer normative data on narrative discourse performance using an equally sized matched control group. This analysis, combined with the normative data, was able to: (i) differentiate impaired versus not impaired discourse, (ii) identify different levels of discourse impairment severity, (iii) identify subtle qualitative differences in those who were within normal limits, (iv) sensitively show change over three monthly intervals and (v) it was suitable for administration across the cohort. Future research exploring the reliability and feasibility of MCA in clinical settings would be useful as would further exploration of its suitability to shorter time intervals and at earlier or later stages of recovery.

Factors influencing recovery

Pre-injury and injury variables

Years of education and post-traumatic amnesia duration emerged as significant covariates in the recovery model. Greater years of education and shorter post-traumatic amnesia duration were related to optimum recovery outcomes. These patterns were also observed in the discourse recovery sub-groups however potential significance was limited by the sample sizes of the sub-groups. Injury severity, as measured by post-traumatic amnesia duration, has been qualitatively reported in one other discourse recovery study (17) and alongside years of education, has been reported as strongly associated with broader psychosocial outcomes following TBI (57). As injury severity has been identified as a key variable for both discourse recovery and psychosocial outcomes, further research exploring the relationship between discourse and psychosocial outcomes would be valuable (58).

Post-injury variables

Speech pathology treatment did not significantly influence the discourse recovery model. This is potentially related to the nature of the therapy data, which very broadly captured the amount of therapy but not the nature or quality of the provided therapy. Indeed, other discourse recovery studies have reported a potential relationship between greater duration of speech therapy and improved conversational discourse outcomes (16) and there is a growing body of evidence supporting the effectiveness of speech pathology interventions following TBI (6,59-62). Further research is evidently needed to determine the extent to which different treatment programs influence the spontaneous recovery process.

On the other hand, the presence of acute aphasia did significantly alter the discourse recovery model. Aphasia is not the predominant communication deficit after TBI (2–32% incidence) (31,32), and the prevalence of discourse or cognitive-communication impairment is higher. In the current study, 40% of individuals had acute mild aphasia compared to 85% of individuals with early discourse impairments. Individuals with aphasia appear to have a different recovery pattern to those without aphasia. A potential explanation is that the language impairments associated with aphasia present a barrier to recovery. Also, those with aphasia may also have more severe injuries overall and thus, more severe cognitive profiles.

Indeed, closer examination of language and cognitive functioning at 3 months revealed a continuum of cognitive-linguistic performance across the cohort as well as a frankly aphasic sub-group 16% (7/43) with relatively prominent language deficits and more severe cognitive deficits, particularly with memory, compared to the remainder of the cohort (Table 4). The predominant feature of frank aphasia was the presence of naming and word-retrieval deficits however all other primary language functions such as auditory comprehension were also affected (Table 4). One outlier presented with severe aphasia (AQ 50). Furthermore, all participants with aphasia at 3 months had a persisting Not Aphasic by WAB presentation at final review (26). These findings reinforce that individuals who present with aphasia following TBI may form a vulnerable sub-group for recovery. The frankly aphasic sub-group was of insufficient size to be incorporated into the recovery model. However, further research exploring and understanding the cognitive-linguistic interaction of the aphasia sub-group, particularly those with frank aphasia characteristics, may facilitate improved recovery and outcomes.

Implications

Findings from this study have a number of implications for rehabilitation services assisting those with severe traumatic brain injuries. Firstly, Speech Language Pathologists (SLPs) should maintain an active role in supporting discourse for at least the first year following injury including qualitatively examining discourse and monitoring those who are within normal limits. Secondly, SLPs may wish to capitalize on peak periods of recovery by offering increased therapy during 3–6 and 9–12 periods. Between 9 and 12 months, awareness and insight training for discourse may be helpful, particularly for individuals who initially present with a mild discourse deficit. Extra supports across the 6–9-month stage such as maintaining frequency of therapy could facilitate ongoing improvement. SLPs may also wish to use MCA to assist with identifying and monitoring discourse deficits. Five key questions have been developed for rehabilitation services to promote the translation of these findings to clinical practice (Table 8). Examples of how this information could be applied are also offered. Clinicians are encouraged to consider each question in relation to their service and to develop a list of appropriate strategies for their service.

Table 8.

Service considerations and applications.

Areas for consideration	Example application
1. Does this service have systems in place for monitoring discourse recovery?	Implement a policy of evaluating all patients at 3, 6, 9 and 12 months post-injury with MCA. Ensure that all clinicians have completed training in MCA.
2. What processes does this service use to maximize discourse recovery outcomes?	Patients are offered at least two intensive blocks of therapy across the first year. Intensive blocks refer to therapy being offered at least two times per week over a 12-week period.
3. How does this service prevent adverse discourse recovery outcomes?	Patients who are not engaged in intensive rehabilitation are reviewed at least fortnightly over the first year.
4. How does this service engage patients and their close others in the recovery process?	Patients and their families are given information around the prognosis of discourse recovery by 6 months post-injury. Patients are their families are invited to participate in a discussion around discourse recovery by 6 months post-injury.
5. What processes are used to assist in planning of future services?	Planning of future services incorporates some consideration of PTA duration, years of education and aphasia.

Limitations and future directions

One of the limitations of this study was that the control group was cross-sectional with a single data point and was not followed longitudinally with the group who had TBI. This posed a barrier to evaluating practice effects, which was instead addressed with alternative methods. Reliable change measures (50-52) were utilized, revealing improvements that reflect recovery of function that cannot be explained by practice effects. Evidence also suggests that clinical populations and tasks of high complexity have low susceptibility to practice effects (63,64).

Furthermore, the control group was comprised of individuals from the US whereas the group with TBI was comprised of individuals from Australia. The narrative task is part of an internationally standardized protocol for western cultures, developed in consultation with representatives from Australia, Canada, the UK and the US. The Cinderella narrative also relies on an overlearned genre, common to both Australian and US cultures and all participants indicated familiarity prior to completing the task. Furthermore, English was the primary language for both of these groups. However, there may have been some subtle linguistic variations that impacted on the analysis. For example, Australians may have a tendency to speak in a more informal or colloquial manner than those who speak US English. Also, there may have been cultural variations between the two groups with exposure to the Cinderella story.

Another limitation of the existing study was that participants were not assessed before 3 months post-injury. Participants could have potentially scored lower at initial assessment and thus shown more change over the first year however, PTA would then have been an important factor to control (53). The overall sample size was a further limitation of the current study, impacting on the effect sizes.

Additionally, a number of limitations relating to the discourse measure were identified. Firstly, while the discourse task was carefully selected to meet the aims of the study, the use of a single cognitive-communication task and single discourse genre may have limited the results (1,48,65,66). Secondly, even though reliability ratings were within an acceptable range (28), there may be discrepancies using this tool in other settings, particularly with untrained users. Notably, the reliability of this analysis appears to be improved in non-impaired populations (27).

A number of future research directions have been identified from this study. Firstly, further exploration of pre-injury, injury and post-injury variables with larger sample sizes than the present study may enable improved understanding of recovery sub-groups. Secondly, exploring the impact of different treatment programs on the recovery process is warranted. Aphasia was another post-injury variable that was implicated in the recovery process, with an identified need for future research to differentiate aphasia in TBI through examination of language and cognition. Additionally, ongoing research into awareness and insight across the first year may provide further clues around facilitating discourse recovery. Further exploration of the reliability and suitability of MCA to clinical settings and mild-moderate TBI would also be of benefit, as would development of an Australian control sample. Finally, exploring longer-term discourse recovery as well as the link between discourse and psychosocial outcomes would strengthen the overall body of research.

Conclusions

The first year following a TBI is clearly an important time period for discourse recovery. This study found that; although discourse improved in a cohort with severe TBI over the first year, individuals with TBI remain significantly below matched controls at the anniversary of their injury. New insights offered by this project include the finding that there are peak periods for recovery between 3–6 and 9–12 months and that there may be slower periods of recovery between 6 and 9 months. A sub-group of individuals improved to within normal limits and this sub-group was characterized by a mild impairment in the earlier stages post-injury and increased recovery rate between 9 and 12 months, possibly related to changes in insight. Years of education, PTA duration and aphasia were identified as important factors in the recovery process. This study contributes to the growing body of evidence reporting high prevalence of discourse disorders following TBI. The present study also offers normative data on narrative discourse performance. MCA was identified as a useful tool for measuring cognitive-communication recovery across a spectrum of discourse deficits and highlighted the need to monitor all patients, even those whose discourse is within a normal range. Services are encouraged to revisit their practices around Speech Pathology service provision across the first year in light of this research. Future research directions could include further exploration of the nature of aphasia in TBI, further investigation of the influence of treatment on discourse recovery as well as replication with larger sample sizes. The services and supports suggested from this research provide a step forward for Speech Pathologists and rehabilitation services in supporting an individual to maximize their cognitive-communication recovery journey.

Figure 3.

Chart of the sub-group trajectories over one year (n = 57). Note ‘WNL’ is the abbreviation for within normal limits.

Appendix

Appendix 1

Worked example of Main Concept Analysis with Cinderella narrative (Control^a)

Content (Essential content bolded)	Element	Code
Once there was a little girl called Cinderella whose dad got remarried to a woman with two daughters.	1	AC
Cinderella’s dad died and she was left with the stepmother and stepsisters.	2	AC
But the stepmother and stepsisters were really mean to Cinderella.	3	AC
Cinderella then had to be the maid to the stepmother and stepsisters.	4	AC
She had to do all the chores around the house.	5	AC
Meanwhile, over at the palace the king thinks the prince need to get married.	6	AC
So the king announces that there is going to be a ball to find a wife for his son.	7	AC
Cinderella and the stepsisters received an invitation for the ball.	8	AC
And they are very excited to go.	9	AC
But Cinderella is not allowed to go unless she gets all the chores done first.	10	AC
She ends up with so many chores that she has no time to make her dress but the little mice and animals that she has befriended make the beautiful dress for her.	-	-
The stepsisters see her in the beautiful dress and are very jealous so they tear the dress off her.	11	AC
So the stepsisters go off to the ball.	12	AC
Cinderella started crying in the garden.	13	AC
As she was crying, a fairy godmother appeared out of nowhere.	14	AC
The fairy godmother then turned a pumpkin into a carriage and the mice are turned into horses to pull the carriage.	15	AC
Then the fairy godmother turns Cinderella’s ragged clothes into a beautiful dress with glass slippers.	16	AC
And so off she went to the ball.	17	AC
But Cinderella had to be home by midnight because that is when the magic would wear off.	18	AC
Cinderella ended up dancing with the prince.	19	AC
And the prince starts to fall in love with her.	20	AC
Cinderella was having so much fun that she lost track of time and only when she heard the bells tolling did she realize that it was midnight.	21	AC
She had to quickly run away.	22	AC
And as she was going down the stairs she lost one of her glass slippers.	23	AC
But she keeps running as everything is starting to go back to the way it was.	25	AC
And she makes it home just in time.	26	AC
Back at the palace, the prince finds the lost shoe	24	AC
And the prince orders his men to search the kingdom for Cinderella.	27	AC
They eventually come to the house where Cinderella and the stepsisters live.	28	AC
The stepsisters try to make the glass slipper fit them.	29	AC
But of course it doesn’t fit either of them.	30	AC
They try the slipper on Cinderella.	31	AC
And it fits perfectly.	32	AC
So Cinderella and the prince then got married.	33	AC
And they lived happily ever after.	34	AC

^aThe control sample was edited for this Appendix as none of the control participants achieved an AC score for all items.

Worked examples of Main Concept Analysis with Cinderella narrative (TBI)

Example 1	Element	Code
Okay um Cinderella’s a little girl that um goes to live with her father and a stepmother with two stepsisters ugly stepsisters.	2	AC
Ah they’re treated really nice.	-	-
And she’s da made to be the maid of the house, clean up and look after everyone, clean up after the sisters get what they want.	4	AC
	5	AI
When she gets a bit older there’s a ball at the local palace.	7	AC
And ah ah ah someone from the palace comes around and invites everyone to the town for the ball.	8	AC
Do um Cinderella Cinderella um finds herself a nice dress and jewellery in a box that she can wear.	-	-
The two stepsisters find her wearing wearing the things and tear the clothes off of her.	11	AC
Told her she’s not allowed to go to the ball which she’s not allowed to.	10	AC
They get dressed up and everything to go to the ball.	12	AC
She’s comes she’s not real happy bit disappointed.	13	AC
Comes across a fairy godmother that finds her.	14	AC
She um dresses her in nice nice ah gown with ah glass slippers.	16	AC
Example 2	Element	Code
and so then one day she saw a magical lady.	14	IC
and she turned her into a um Cinderella like a nice princess for a person.	16	AI
and then all the, she went to a dance.	17	AC
and then something happened.	-	-
someone probably upset her.	-	-
or the bloke… danced on the lawn or something.	19	AI
and so she left.	22	AI
and when she was going down the stairs, she lost her, like on the way out, she lost her shoe.	23	AC
Example 3	Element	Code
and the horses were made from mice by memory.	15	AI
and then she goes to the ball.	17	AC
but the problem was that at twelve o’clock her costume turned back to what is originally was.	18	AI
so she had to um. at twelve o’clock or when it got close to twelve o’clock.	21	AC
she did the runner on the prince to go home.	22	AC
and lost her shoes, or shoe.	23	AC
anyway the prince had met the girl Cinderella and liked her a lot.	20	AC
and wanted to find out who she was.	-	-
so he sent out all his workers to find out who the shoes fit the best.	27	AC
and they went round from house to house.	-	-
and tried the shoe on all the single ladies.	-	-
and in the end the shoe fits her.	32	AC

Appendix 2

Case example of discourse recovery

Targeted Content	3 Months	6 Months	9 Months	12 Months
Stepmother/ stepsisters were mean to Cinderella	and they were mean	(Absent)	the stepsisters didn't really like them didn't really like Cinderella and always teased her and ah made fun of her	the two the two sisters from the mother don't like the the one from the dad and they always give her hell
Coding	Inaccurate-Completea	Absent	Accurate- Complete	Accurate-Complete
She knew she had to be home by midnight because everything will turn back at midnight.	She can go there until twelve o'clock and and then the her um everything turns into a pumpkin	but she has to return by a certain time otherwise it'll all turn into pumpkin into a pumpkin	and if she if Cinderella doesn't get back by twelve o'clock and she her carriage her carriage gets turns into a pumpkin	but she has to be out of there by midnight, otherwise she turns in a pumpkin. Her carriage turns back into a pumpkin
Coding	Inaccurate-Completeb	Inaccurate-Completec	Accurate-Complete	Accurate-Complete
The prince searched (was trying to find/looked for) for Cinderella (for the person who would fit into the glass slipper/ for the girl from the ball)	(Absent)	so the prince um the prince she was dancing with tries gets everyone to try the shoe to find out who who was there	and he needs to find out who the slipper belongs to so he gets everyone to try on the slipper	and then the prince wants to find her again and the only way to find her is through the slipper so he gets everyone to try on the slipper to find out who he was dancing with
Coding	Absent	Inaccurate-Complete d	Inaccurate-Completee	Accurate-Complete
Total MCA score	11	32	33	53

Note: This shows 3 examples from the total 34 potential elements.

^a“they” was an unspecified pronoun and considered incomplete

^b“can go there” was considered inaccurate

^c“by a certain time” was not specific

^dno indication of searching/looking

^epronoun was not specified.

Appendix 3

Visual displays of linear mixed models

Covariantes appearing in the model are evaluated at the following values: Age = 37.32, Education Years = 14.35, PTA = 51.03

Covariantes appearing in the model are evaluated at the following values: Age = 37.32, Education Years = 14.35, PTA = 51.03

Covariantes appearing in the model are evaluated at the following values: Age = 37.00, Education Years = 14.59, PTA = 52.10