Young Adults With Acquired Brain Injury Show Longitudinal Improvements in Cognition After Intensive Cognitive Rehabilitation

Abstract

Purpose:

The aim of this study was to assess the effect of an intensive cognitive and communication rehabilitation (ICCR) program on language and other cognitive performance in young adults with acquired brain injury (ABI).

Method:

Thirty young adults with chronic ABI participated in this study. Treatment participants (n = 22) attended ICCR 6 hours/day, 4 days/week for at least one 12-week semester. Deferred treatment/usual care control participants (n = 14) were evaluated before and after at least one 12-week semester. Pre- and postsemester standardized cognitive assessment items were assigned to subdomains. Between-groups and within-group generalized linear mixed-effects models assessed the effect of time point on overall item accuracy and differences by item subdomain. Subdomain analyses were adjusted for multiple comparisons.

Results:

Between-groups analyses revealed that treatment participants improved significantly faster over time than deferred treatment/usual care participants in overall item accuracy and specifically on items in the verbal expression subdomain. Investigating the three-way interaction between time point, group, and etiology revealed that the overall effects of the treatment were similar for individuals with nontraumatic and traumatic brain injuries. The treatment group showed an overall effect of treatment and significant gains over time in the verbal expression, written expression, memory, and problem solving subdomains. The control group did not significantly improve over time on overall item accuracy and showed significant subdomain-level gains in auditory comprehension, which did not survive correction.

Conclusions:

Sustaining an ABI in young adulthood can significantly disrupt key developmental milestones, such as attending college and launching a career. This study provides strong evidence that integrating impairment-based retraining of language and other cognitive skills with “real-world” application in academically focused activities promotes gains in underlying cognitive processes that are important for academic success as measured by standardized assessment items. These findings may prompt a revision to the current continuum of rehabilitative care for young adults with ABI.

Supplemental Material:

https://doi.org/10.23641/asha.19320068

Acquired brain injury (ABI) encompasses a variety of etiologies, including traumatic brain injury (TBI), stroke, tumor, anoxic/hypoxic injury, and encephalitis among others. While sustaining an ABI at any age can have significant consequences, sustaining an ABI in young adulthood can significantly derail the trajectory of an individual's academic, career, and social development (Committee on Improving the Health and Well-Being of Young Adults et al., 2015). Unfortunately, young adults are one of the most frequently affected groups to sustain TBI (i.e., ages 15–24 years; Taylor, 2017) and nonfatal opioid overdose, which can lead to anoxic/hypoxic injuries (i.e., age ranges between 15 and 34 years; Office of the Assistant Secretary for Planning and Evaluation, 2019; Vivolo-Kantor, 2020). Furthermore, the rate of stroke in college- and working-age individuals (i.e., 18–50 years) has been on the rise over the past several decades (Benjamin et al., 2019) due to an increase in vascular risk factors in this group (e.g., hypertension and diabetes; Singhal et al., 2013). Concern over the growing number of young adults with ABI in need of rehabilitation services to get their lives back on track is underscored by the fact that the majority of cognitive rehabilitation programs (i.e., approaches that target attention, visuospatial functioning, language and communication skills, memory, and executive function; Cicerone et al., 2019) do not provide services at the frequency and intensity necessary to prepare them for college (e.g., Babbitt et al., 2016; Kennedy & Krause, 2011; Klonoff et al., 2006; Todis & Glang, 2008).

ABI Impacts Cognitive Processes

ABI often leads to long-term deficits in a range of cognitive domains, such as language, attention, memory, executive function, and visuospatial/constructional processes. As cognitive processes are supported by large-scale brain networks (Kljajevic, 2014; Petersen & Sporns, 2015), there is considerable overlap in impaired cognitive processes across ABI etiologies, despite differences in the nature of the injury (e.g., focal vs. diffuse). Aphasia is common after focal injury, such as left hemisphere stroke, but it has also been demonstrated after TBI (Kiran, 2012; McAllister, 2011; Norman et al., 2013), especially in more moderate to severe cases. Attention, memory, and executive function are consistently impaired after diffuse injury (i.e., TBI [McAllister, 2011; Rabinowitz & Levin, 2014] and anoxic/hypoxic injury [Cullen & Weisz, 2011; Shah et al., 2004]) but can also occur after focal injury, for example, in the context of stroke-induced aphasia (Gilmore, Meier, et al., 2019; Kertesz & McCabe, 1975; Lang & Quitz, 2012; Purdy, 2002; Seniów et al., 2009; Simic et al., 2017; Villard & Kiran, 2015). Visuospatial/constructional deficits occur across ABI etiologies (Arciniegas & Anderson, 2004; Cullen & Weisz, 2011; Gehring et al., 2010; Hokkanen et al., 1996; McKay et al., 2008; McKenna et al., 2006; Shah et al., 2004; Tonning Olsson et al., 2014), with some variation in the frequency or severity based on location of injury (Wilde, 2006, 2010). In summary, individuals with ABI have overlapping patterns of cognitive deficits in language, attention, memory, executive function, and visuospatial/constructional processes, which are important for academic success.

Cognitive Processes Important for College

The same cognitive processes (i.e., language, attention, memory, executive function, and visuospatial/constructional) that are frequently impaired in ABI are often relied upon by young adults in college. There is a great deal of research emphasizing the importance of various cognitive domains on college performance with a general pattern of higher performance in the cognitive domain of interest accompanying higher academic achievement. For example, neurotypical college students attained higher grades (Weyandt et al., 2013) and were at lower risk for academic challenges (Weyandt et al., 2017) than college students with attentional impairments. College freshmen with higher working memory performance had higher grade point averages (GPAs; Hannon, 2014), a standard metric of academic achievement, than students with lower working memory performance. In terms of executive function, studies have shown that students with greater conceptual reasoning ability (Rohde & Thompson, 2007), study skills (Hartwig & Dunlosky, 2012; Hassanbeigi et al., 2011), strategy usage (Taraban et al., 2000), self-regulation (Cohen, 2012), and self-efficacy (Krumrei-Mancuso et al., 2013) earned higher GPAs than their counterparts. As expected, positive relationships have also been found between visuospatial processing ability and performance in science (Castro-Alonso & Uttal, 2019) and math (Rohde & Thompson, 2007). Finally, the reliance on language skills in college is unarguable (Hargie, 2006; Mahmud, 2014; Morreale & Pearson, 2008; Rubin & Graham, 1988). For example, students with better listening performance (Feyten, 1991) and reading comprehension (Royer et al., 1990) had greater success in college than their counterparts with worse performance in those domains.

Based on the pathophysiology of ABI, it is not surprising that young individuals with ABI struggle with academics after their injury. Students with TBI report that deficits in attention, executive function, and memory function impact their academic performance (Kennedy, Krause, & Turkstra, 2008). This group also endorses having to review material to a greater extent than preinjury and having difficulty understanding course material (Cahill et al., 2014). Some students with TBI modified their academic status by taking fewer courses per semester than before their injury and even changed their career goals (Kennedy, Krause, & Turkstra, 2008; Todis & Glang, 2008). Predictably, young adults with disability (including TBI) graduated from post–secondary education less often than peers without disability (Sanford et al., 2011). Over 40% of young adults with stroke demonstrate long-term language and other cognitive impairments, which can impede return to work and school (Yahya et al., 2020). Unfortunately, the impact of stroke on academic advancement for young adults with ABI has been understudied relative to TBI, and not surprisingly, services for this unique group are often inadequate and disjointed (Radford & Walker, 2008). One study investigating the academic experiences of young adults with stroke-induced aphasia revealed self-endorsed difficulty taking notes, recalling what the professor said, and remembering what they had read (Mattuzzi & Pfenninger, 2018). Study participants also ranked class activities involving speaking (e.g., oral presentations) as the most stressful and reported feeling anxious about their language difficulty in class. Although individuals with TBI and stroke-induced aphasia may experience difficulty with the same academic activities (e.g., recalling information from the lecture), in many cases, this difficulty is driven by different underlying deficits that should be considered when targeting these activities in therapy (e.g., individuals with aphasia may not be able “to recall information” because of auditory comprehension impairments that affected encoding or lexicosemantic impairments that affected access and retrieval; individuals with TBI may not be able “to recall information” because of attention impairments that affected encoding or memory impairments that affected retrieval).

Individuals with tumor and encephalitis also experience academic challenges after injury. Parsons et al. (2012) report that over half of young adult cancer survivors (i.e., first cancer diagnosis between 15 and 29 years of age) endorsed challenges with return to work or school that were cognitive in nature (e.g., “trouble keeping up with work or studies,” “forgetting things,” and “hard to pay attention at work or school”). Young adults with encephalitis also experience academic challenges postinjury and may need specific strategies to succeed (Obrecht & Patrick, 2002). In fact, Fraas and Bellerose (2010) investigated the effects of a mentoring program for a young adult with encephalitis who experienced difficulty adjusting to school postinjury due to persistent memory impairment, emotional deficits, and fatigue. In summary, young adults sustain ABI when they are on the precipice of launching their educational and career goals. Associated language and other cognitive impairments can substantially disrupt their academic and vocational trajectories. Thus, it is paramount that this unique population receives cognitive rehabilitation that is specifically tailored to their personal goals, such as getting back on track toward postsecondary education and a future career, and clinical deficit profiles (e.g., aphasia and executive dysfunction).

Current Cognitive Rehabilitation Approaches

Many young adults with ABI receive cognitive rehabilitation to address deficits in the domains discussed in the preceding sections. Cognitive rehabilitation can take several forms, including restorative, compensatory, comprehensive, and/or contextualized approaches (Cicerone et al., 2019; Hart, 2010; Institute of Medicine, 2011; Wilson, 1997, 2002; Ylvisaker et al., 2002). It can also be modular, targeting a cognitive domain (i.e., attention, visuospatial functioning, language and communication skills, memory, and executive function; Cicerone et al., 2019) in isolation (e.g., Sohlberg et al., 2000), or be multimodal, targeting multiple cognitive domains simultaneously (e.g., Cicerone et al., 2008).

Despite the availability and application of cognitive rehabilitation approaches, there is no clear evidence that existing programs substantially contribute to the advancement of young adults with ABI to college. Comprehensive rehabilitation programs report positive functional outcomes (i.e., productivity and independence; Cicerone et al., 2000, 2005, 2011). However, the frequency of return to school (Klonoff et al., 2006; Sarajuuri et al., 2005) is difficult to discern as it is often combined with return to work (Cicerone et al., 2004, 2008; Goranson et al., 2003; Vanderploeg et al., 2008) or not reported (Cooper et al., 2017; Mills et al., 2006; Schönberger et al., 2006; Svendsen & Teasdale, 2006). Additionally, one rehabilitation program, designed to support young adults with ABI by providing coaching support for studying and learning, time management, and interpersonal interaction, reported modest postprogram benefits for the two individuals included in the study (Kennedy & Krause, 2011). However, the Kennedy and Krause (2011) program was designed for young adults who have already been admitted into college and thus does not serve those with more moderate to severe impairments that may require intensive, academically focused rehabilitation to advance to college.

Our own prior work in this area has demonstrated the feasibility of implementing an intensive, academically focused cognitive rehabilitation program specifically for young adults with ABI who wish to pursue college but currently cannot due to the severity of their language and/or other cognitive deficits. The Intensive Cognitive and Communication Rehabilitation (ICCR) program includes classroom-style lectures, individual therapy, and technology training for 6 hours/day, 4 days/week, and 12-week iterations. A central tenet of the program is that the integration of impairment-based retraining of language and other cognitive skills with “real-world” application in academically focused activities (e.g., listening to a lecture and taking notes, studying for quizzes, and answering discussion questions) should drive change in underlying cognitive processes as measured by standardized assessment items (Meier et al., 2017)—an alternative approach to interventions that target impairment and measure change in function (e.g., Cantor et al., 2014; Doesborgh, 2003). Full details of the initial efficacy study are reported elsewhere (Gilmore, Ross, & Kiran, 2019), and thus, the results will only be summarized here. Six young adults with chronic ABI were enrolled in the study (n = 4 treatment participants, n = 2 control participants). Before and after each treatment/no-treatment period, all participants underwent a battery of standardized assessments examining global cognitive function. Treatment participants showed statistically significant gains in at least one standardized assessment of cognitive function, whereas control participants did not, suggesting that the improvements achieved by the treatment participants were likely attributable to the intervention.

Summary of the Problem

Young adults rely on executive function, attention, memory, visuospatial processing, and language domains to succeed in college. These domains are often impaired in young adults with chronic ABI, and cognitive deficit profiles overlap across ABI etiologies. Treatment approaches are commonly segregated by ABI etiology, despite obvious benefits in including individuals with different ABI etiologies in the same intervention (e.g., provision of a peer rehabilitation group, balance of impaired and spared processes in a group context that may facilitate collaboration and empowerment). Cognitive rehabilitation programs for young adults with ABI struggling to advance to college should focus on impaired cognitive domains that have been shown to support academic success in healthy young adults. Nevertheless, academic outcomes for existing cognitive rehabilitation programs are limited in the literature. Furthermore, cognitive function is not consistently or thoroughly assessed as an outcome measure for such programs (e.g., Cooper et al., 2017; Svendsen & Teasdale, 2006). Some studies have reported an aggregate score (e.g., Cicerone et al., 2004, 2008), but these types of composite or summary scores derived from commonly used standardized outcome measures (e.g., Western Aphasia Battery–Aphasia Quotient [WAB-AQ]; Kertesz, 2006; and Repeatable Battery for the Assessment of Neuropsychological Status total index score [RBANS-Total]; Randolph, 2012) are coarse and may obscure treatment-related gains in specific cognitive domains targeted by an intervention. Although subtest scores can be inspected as an alternative, this approach can also be flawed. The analysis in this study leveraged rich item level data from four commonly administered standardized assessments of cognitive function to overcome some of these challenges and capture subtle improvements in specific cognitive domains.

This study investigated the effect of the ICCR program, which combined targeted retraining of language and other cognitive skills with repeated opportunities for application in a functional context (i.e., classroom-based activities) on a range of underlying cognitive domains as measured via standardized assessment battery items in a group of young adults with ABI pursuing postinjury college enrollment. This overall study objective was addressed via the following specific aims:

1. comparing overall cognitive function and performance on specific language and other cognitive domains—known to be impaired in individuals with ABI, important for academic success, and the focus of this multifaceted integrated intervention—over time between a group of young adults with ABI who participated in ICCR (i.e., treatment) and a group of young adults with ABI who did not (i.e., deferred treatment/usual care control),

2. examining longitudinal performance in overall cognitive function and specific language and other cognitive domains for the treatment and control groups individually, and

3. assessing whether changes in overall cognitive function for the treatment versus control group over time differed for young adults with traumatic versus nontraumatic ABI etiologies.

Method

Study Design

The study employed a longitudinal nonrandomized intervention design (Moerbeek, 2008; Sedgwick, 2017). Participants who met the eligibility criteria were given the choice to enroll in the treatment or defer for a semester. If they chose to defer treatment, they were given the standardized assessment battery (see the Assessment section). Before the start of the next semester, the study team contacted them to complete the assessment battery again, and they were again given the option to enroll in the intervention (as participation in multiple semesters was permitted) or continue to defer. The deferred treatment control phase always preceded the treatment phase. While in the deferred treatment/usual care group, participants were asked to refrain from taking college courses but otherwise were able to participate in their daily lives (e.g., volunteer, work, and attend outpatient therapy). Participants who did not attend outpatient speech therapy in the community during the control phase were considered “deferred treatment” controls, and those who sought outpatient speech therapy in the community of their own accord during the control phase were considered “usual care” controls. See Supplemental Material S1 for details about the deferred treatment/usual care control participants' activities during the study.

Recruitment

Participants were recruited from the greater Boston area and nationally for this longitudinal study via the following methods: (a) word of mouth; (b) referrals from speech-language pathologists, neuropsychologists, and physicians; (c) posting on professional message boards; (d) social media; and (e) conference presentations. Primary eligibility criteria for this study's enrollment included the following: (a) young adult between the ages of 18 and 40 years (Erikson, 1997; McLeod, 2018); (b) sustained an ABI; (c) presence of language and/or other cognitive deficits as determined by performance below normal limits on the WAB-R AQ (Kertesz, 2006; < 93.8) and/or the RBANS-Total (Randolph, 2012; < 85); (d) goal of enrolling in and/or returning to post–secondary education; and (e) adequate hearing for conversation and adequate vision for functional reading based on medical records review, self/caregiver report, and/or clinical judgment. Potential participants with concomitant neurological disease (e.g., epilepsy and attention-deficit disorder) were considered for inclusion on an individual basis. Individuals with neurodegenerative disease were excluded.

Participants

Between fall 2016 and fall 2020, 37 individuals were screened for the study. Seven individuals were excluded (i.e., five individuals did not meet inclusion criteria and two individuals declined to pursue the program after screening). The remaining 30 participants were enrolled in this study.

Sixteen unique young adults enrolled into the treatment group immediately. Fourteen unique young adults (eight men, age M = 25.99 years, SD = 5.64; months postonset M = 57.77, SD = 46.27; TBI = 7, non-TBI = 7; WAB-AQ M = 84.15, SD = 15.73, range: 43.7–99.5; RBANS-Total M = 57.93, SD = 10.37, range: 45–79) enrolled as deferred treatment/usual care control participants. Six of these 14 deferred treatment/usual care participants (P13/C7, P14/C10, P17/C12, P18/C2, P19/C13, and P22/C11) transitioned to the treatment group after completing their control study phase(s), increasing the treatment group to 22 young adults (15 men, age M = 24.24 years, SD = 4.43; months postonset M = 52.00, SD = 39.10; TBI = 10, non-TBI = 12; WAB-AQ M = 78.78, SD = 20.93, range: 18.8–99.6; RBANS M = 55.09, SD = 10.84, range: 44–78). See Figure 1 for flowchart of recruitment, enrollment, self-allocation to groups, and analysis. See the Model Building and Structure section for how the six participants who contributed data to both groups were managed in the analyses.

Figure 1.

Flow diagram for recruitment, enrollment, self-grouping, and analysis. *Data from both of their study phases were included in the analyses. See the Model Building Structure section in Method for how these data were managed.

All participants provided written consent to participate in the study in line with human subjects policies and procedures put forth by the Boston University Institutional Review Board. They each had attained at least a high school education by study enrollment, although a high school degree was not required for inclusion. The treatment and deferred treatment/usual care control groups did not significantly differ on age, months postonset, WAB-R AQ, RBANS-Total, or education level based on Welch's two-sample t tests (p > .05 level). See Table 1 for additional demographic details, including any premorbid history of mental health conditions or learning disabilities/differences endorsed during screening.

Table 1.

Demographic details.

ID	Age	Sex	MPO at enrollment	ABI etiology broad	ABI etiology specific	Education level	WAB-R-AQ	Severity of language impairment	RBANS-Total	Severity of cognitive impairment	No. of time points contributed	Premorbid MH or LD Dx
Treatment participants
P1	24.09	F	99.02	NTBI	Tumor	13	81.8	Mild	44	Severe	4	No
P2	29.16	M	70.2	NTBI	Stroke	15	78.8	Mild	64	Mod	4	No
P3	24.64	F	44.25	NTBI	Stroke	16	58.7	Mod	44	Severe	2	No
P4	21.01	M	49.28	TBI	TBI	12	61.9	Mod	45	Severe	4	No
P5	25	M	96.06	TBI	TBI	12	62.5	Mod	46	Severe	4	No
P6	35.21	M	97.24	TBI	TBI	16	18.8	Very severe	47	Severe	2	No
P7	22.12	F	14.95	TBI	TBI	14	93.8	Mild	78	Mild	4	No
P8	27.53	M	13.14	TBI	TBI	14	96.8	WNL	59	Mod	2	Yes, MH
P9	25.35	M	68.86	TBI	TBI	12	67.5	Mod	49	Severe	4	No
P10	29.67	M	97.77	NTBI	Stroke	15	87.5	Mild	52	Severe	4	Yes, MH
P11	25.73	M	35.68	TBI	TBI	15	90.6	Mild	53	Severe	4	No
P12	21.26	M	13.11	TBI	TBI	13	92.2	Mild	49	Severe	2	No
P13 a	21.89	M	4.57	NTBI	Tumor	15	94.2	WNL	52	Severe	4	Yes, LD
P14a	20.45	M	53.88	NTBI	Stroke	13	92.8	Mild	74	Mild	4	Yes, LD
P15	18.02	F	37.02	NTBI	Stroke	12	99.6	WNL	74	Mild	4	No
P16	21.55	M	19.97	NTBI	Tumor	14	97.2	WNL	68	Mod	4	Yes, LD
P17a	25.37	M	144.94	NTBI	Encephalitis	12	96.5	WNL	60	Mod	4	No
P18a	20.5	F	22.24	NTBI	Tumor	13	65	Mod	46	Severe	3	No
P19a	21.04	M	18.53	TBI	TBI	12	43.7	Severe	45	Severe	3	No
P20	22.17	F	31.34	NTBI	Stroke	12	65.6	Mod	45	Severe	2	No
P21	18.79	F	12.02	NTBI	Stroke	12	95	WNL	63	Mod	2	No
P22a	32.81	M	99.93	TBI	TBI	18	92.6	Mild	55	Severe	2	No
M (SD)	24.24 (4.43)	M = 15 F = 7	52.00 (39.10)	NTBI = 12 TBI = 10	TBI = 10 Stroke = 7 Tumor = 4 Encephalitis = 1	13.64 (1.71)	78.78 (20.93)	WNL = 6 Mild = 8 Mod = 6 Severe = 1 Very severe = 1	55.09 (10.84)	WNL = 0 Mild = 3 Mod = 5 Severe = 14	3.27	No Hx = 17 MH = 2 LD = 3
Range	18.02–35.21		4.57–144.94			12–18	18.8–99.6		44–78		2–4
Deferred treatment/usual care control participants
C1	23.06	F	38.11	TBI	TBI	12	91.3	Mild	52	Severe	3	No
C2a	20.5	F	22.24	NTBI	Tumor/hemorrhage	13	65	Mod	46	Severe	2	No
C3	30.94	M	38.47	NTBI	Stroke	23	72.1	Mod	64	Mod	2	No
C4	31.53	F	59.76	NTBI	Stroke	14	84.3	Mild	71	Mild	3	No
C5	29.61	M	158.21	TBI	TBI	12	99.5	WNL	54	Severe	3	No
C6	22.35	M	48.55	TBI	TBI	12	92	Mild	55	Severe	3	No
C7a	21.89	M	4.57	NTBI	Tumor	15	94.2	WNL	52	Severe	2	Yes, LD
C8	24.95	F	42.68	TBI	TBI	14	97.6	WNL	79	Mild	2	Yes, MH
C9	21.1	F	17.45	NTBI	Stroke	13	72.3	Mod	51	Severe	2	No
C10a	20.45	M	53.88	NTBI	Stroke	13	92.8	Mild	74	Mild	2	Yes, LD
C11a	32.81	M	99.93	TBI	TBI	18	92.6	Mild	55	Severe	3	No
C12a	25.37	M	144.94	NTBI	Encephalitis	12	96.5	WNL	60	Mod	2	No
C13a	21.04	M	18.53	TBI	TBI	12	43.7	Severe	45	Severe	3	No
C14	38.23	F	61.46	TBI	TBI	16	84.2	Mild	53	Severe	2	No
M (SD)	25.99 (5.64)	M = 8 F = 6	57.77 (46.27)	NTBI = 7 TBI = 7	TBI = 7 Stroke = 4 Tumor = 2 Encephalitis = 1	14.21 (3.09)	84.15 (15.73)	WNL = 4 Mild = 6 Mod = 3 Severe = 1 Very severe = 0	57.93 (10.37)	WNL = 0 Mild = 3 Mod = 2 Severe = 9	2.43 (0.51)	No Hx = 11 MH = 1 LD = 2
Range	20.45–38.23		4.57–158.21			12–23	43.7–99.5		45–79		2–3

Note. Severity of language impairment was assigned as follows: within normal limits (WNL) > 93.8, mild = 93.8–76, moderate = 51–75, severe = 26–50, very severe = 0–25, based on the WAB-R manual. Severity of cognitive impairment was assigned as follows: WNL, < 1 SD below the mean; mild, ≥ 1 SD below the mean but < 2 SDs below the mean; moderate, ≥ 2 SDs below the mean but less than 3 SDs below the mean; severe, ≥ 3 SDs below the mean. Participants with “Yes” demarcation in the final column reported premorbid history (Hx) of mental health diagnosis (MH; e.g., attention-deficit disorder and depression) or learning disability/difficulty (LD; e.g., required individualized education program in school for reading). P13 and C17 were < 12 months postonset when they signed the consent form. They had an unexpected change in medical status after enrolling and thus started their deferred treatment control phase at 15 months postonset and treatment phase at 18 months postonset, respectively. MPO = months postonset of injury at time of enrollment; ABI = acquired brain injury; WAB-R-AQ = Western Aphasia Battery–Revised Aphasia Quotient (Kertesz, 2006; < 93.8 suggests presence of language impairment); RBANS-Total = Repeatable Battery for the Assessment of Neuropsychological Status (Randolph, 2012; M = 100, SD = 15); F = female; NTBI = nontraumatic brain injury; M = male; TBI = traumatic brain injury.

^a Participant started as a deferred treatment/usual care control and transitioned to the treatment group.

Assessment

All participants were administered a standardized assessment battery before and after each semester of the intervention. For participants who participated in multiple, consecutive semesters of the study, the posttreatment data from the previous semester was used as the pretreatment data for the subsequent semester. The following tests were selected from a larger battery of assessments administered as part of the intervention protocol: (a) the WAB-R to measure language function (e.g., verbal expression), (b) the RBANS to evaluate other cognitive function (e.g., memory), (c) the Scales of Cognitive and Communicative Ability for Neurorehabilitation (SCCAN; Holland & Milman, 2012) to assess language and other cognitive functions (e.g., attention and reading), and (d) the Discourse Comprehension Test (DCT; Brookshire & Nicholas, 1993) to evaluate narrative-level language function (i.e., auditory and reading comprehension at the multiparagraph level). See Supplemental Material S2 for pretreatment/deferred treatment subtest scores for the WAB-R, RBANS, SCCAN, and DCT for all participants (i.e., collected at the start of their first time point in the study).

Behavioral Intervention

ICCR involved classroom-style lectures, group and individual therapy, and computer- and application-based training (Gilmore, Ross, & Kiran, 2019). Participants attended ICCR 6 hours/day, 4 days/week for at least one 12-week semester (i.e., approximately 240 hours/semester). As demonstrated in Figure 2 and detailed in Table 2, participants were exposed to material from four different college-level courses per semester, alternating between two sets of courses daily (e.g., Monday/Thursday: Psychology and Statistics; Tuesday/Friday: Advanced Biology and English Literature). Daily treatment components included (a) watching a prerecorded lecture as a group (e.g., taxing attention and auditory comprehension), (b) reviewing lecture content as a group (e.g., targeting short-term memory, verbal expression, auditory comprehension, and problem solving), (c) answering practice quiz questions about the lecture as a group (e.g., recruiting short-term memory, problem solving, and reading), (d) participating in a discussion- or project-based course as a group (e.g., taxing verbal expression, reading, writing, and problem solving), (e) completing individualized technology training in a group context (e.g., focusing on various cognitive domains based on participants' clinical profile and needs), and (f) engaging in individual therapy with a speech-language pathologist (e.g., targeting various cognitive domains based on participants' clinical profiles and interests). Participants were able to take breaks as needed throughout the sessions. If they missed a session, they were provided instructions to access the material at home and/or during technology time on the next program day and any missed quizzes were made up. Of note, average attendance was 93%, suggesting good adherence to the treatment intensity and acceptability for participants.

Figure 2.

Sample Intensive Cognitive and Communication Rehabilitation program weekly schedule.

Table 2.

Detailed description of intensive cognitive and communication rehabilitation program components.

Component	Description	Materials	Common clinician support
Quiz on previous week's course content	Each question read aloud by speech-language pathologists	5 questions (i.e., multiple choice, yes/no, true/false, short answer)	Repeated the questions and provided extra time, as requested
AM lecture (e.g., biology)	Watch prerecorded video	45–60 min of open-source academic content Laptop and projector	Tactile, visual, and verbal cues to sustain attention and/or take notes
Lecture review	Discuss lecture content as a group	Whiteboards and markers Premade lecture notes for speech-language pathologists, support staff, and participants	Cues to support word retrieval, recall, and organization Modeled and supported participants to use metacognitive strategies
Practice quiz questions	Answer questions about the day's lecture content Share answers and discuss why correct/incorrect Locate answers in the notes Mark areas of notes to study	PowerPoint presentation with questions from the day's lecture and answers Laptop and projector Response sheets for students	Encouraged participants to use the activity to identify what they needed to study before the actual quiz the following week Modeled and supported participants to use metacognitive strategies (e.g., rehearse, imagine, take time, and activate)
PM seminar (e.g., personal finance)	Mix of discussion- and project-based activities Questions and discussion were interleaved with new content	Varied by course topic (e.g., book and chapter summaries for English literature)	Same cues as in lecture and lecture review (e.g., cues to sustain attention, support language comprehension/expression, promote recall/organization) Constructive feedback in verbal/written form (e.g., for public speaking, essay writing)
Individual therapy	Target therapy goals developed with the participant and focused on their needs	Varied by participant and/or domain Included impairment-based and functionally based approaches	Tactile, visual, and verbal cues, depending on the nature of the activity and participants' performance
Technology-based intervention	Range of therapeutic (e.g., Constant Therapy) and academic activities (e.g., making flashcards on Quizlet)	iPads laptops headphones cognitive–linguistic therapy applications Attention Process Training-3 Google Suite	Helped set-up the equipment Monitored participants' performance Provided direct intervention for challenging activities Cues to sustain attention to task

Note. During the COVID-19 pandemic in 2020, the lectures/seminars, review sessions, practice quiz questions, and individual therapy were delivered using the Zoom platform.

As detailed in Table 2, the majority of the intervention was group based and was delivered in a college classroom by the speech-language pathologist responsible for the classroom-based intervention and trained study support staff (i.e., graduate students in speech, language, and hearing sciences and/or research assistants from the Aphasia Research Laboratory). Courses were developed using open-source academic content, such as Khan Academy (Khan Academy, 2017) and Open Yale (Bloom, 2012). Trained study staff developed lecture notes and quiz questions independently and/or adapted from materials provided by the course's source. All new speech-language pathologists, graduate student clinicians, and research assistants were trained via a combination of in-person and hands-on experiences as well as a review of written protocols before implementing the intervention procedures.

Individualized speech-language therapy was provided by a graduate student or clinical fellow in speech-language pathology under the supervision of a licensed and certified speech-language pathologist. For each individual, therapy goals were established and targeted within weekly one-on-one treatment sessions. Treatment goals were generated via review of standardized assessment results, observation of client performance within the group setting, and collaboration with clients and/or their families to meet specific needs with respect to language and other cognitive domains. Individual treatment activities incorporated evidence-based cognitive rehabilitation approaches (Cicerone et al., 2019), such as semantic feature analysis (Gilmore et al., 2018), metacognitive strategy training (Kennedy, Coelho, et al., 2008), and copy and recall treatment (Beeson & Egnor, 2006).

One of the primary thrusts of the ICCR program is the benefit of “real-world” application, and thus, the ICCR program was delivered in a classroom setting (i.e., the same rooms used by Boston University students). Treatment participants' experiences were similar to those of students taking courses in typical college classrooms in several ways. For example, participants had to follow a schedule, including preparation for lunch. They traveled to different rooms for classes at times and for individual therapy. They also were responsible for remembering to bring school supplies and letting the clinician know if they would be out or had to leave early. The morning courses were generally cumulative in nature with each session's lecture content building on previous course material as is common in college. Participants watched course lectures as a group and took turns answering questions or explaining concepts to their peers. Similar to a “real-world” college course, participants inadvertently distracted one another during class (e.g., “searched through their bookbag for a pen” and “got up to use the restroom”).

Operations During the COVID-19 Pandemic

The program transitioned to remote delivery via Zoom during the spring 2020 semester and continued as such through fall 2020. There was no interruption of care in spring 2020 as the program transitioned during a natural break in the semester. The roles of the speech-language pathologists responsible for classroom-based and individually based treatment did not change, nor did those of the study staff trained to support these program components. The classroom speech-language pathologist led the prerecorded lecture viewing, the lecture review sessions, and the seminar course discussion via Zoom with “push-in” support from the graduate student clinicians. The individual therapy was also provided over Zoom with “real-time” feedback and support from the supervising speech-language pathologist. Finally, delivering ICCR remotely during the COVID-19 pandemic simulated the experiences of college students across the globe who also transitioned to online courses in accordance with safety guidelines. Although potential advantages (e.g., access to services outside the greater Boston area) and disadvantages (e.g., group dynamic changes) of remote ICCR delivery must be acknowledged and formally investigated in future work, extensive efforts were made to maintain the protocol delivery across in-person and remote means as detailed above, and thus, data from remote ICCR were included in the analyses.

Data Analysis

As shown in Figure 3, items from the WAB-R, the RBANS, the SCCAN, and the DCT were assigned to one of 10 subdomains based on how they were classified in the parent standardized assessment (i.e., auditory comprehension, reading comprehension, verbal expression, written expression, attention, memory, problem solving, orientation, upper limb/facial/instrumental apraxia, and visuospatial/constructional). This method worked well for the majority of the items, except when an item's subtest name did not clearly match one of the 10 subdomains. In those cases, items were assigned to the subdomain that reflected the primary nature of the item based on neuropsychological reference materials (Lezak et al., 2012) and clinical judgment. The reader is referred to Supplemental Material S3 for additional detail regarding the management of these items. Item accuracy was represented by a pair of columns in the analyses: (a) the number of points scored on an item and (b) the number of points missed on an item to capture the binary scoring system in which each point was either scored or missed by the participant.

Figure 3.

Item assignment to subdomains. Comp. = comprehension; Apraxia = items from WAB-R Apraxia subtest that measures upper limb, facial, instrumental, and complex actions; WAB-R = Western Aphasia Battery–Revised; RBANS = Repeatable Battery for the Assessment of Neuropsychological Status; SCCAN = Scales of Cognitive and Communicative Ability for Neurorehabilitation; DCT = Discourse Comprehension Test.

Growth Curves

A growth curve analysis approach was implemented to accomplish this study's specific aims for several reasons (Curran et al., 2010; Oleson et al., 2019). First, it captures longitudinal performance for the overall group while accounting for differences in baseline performance and change over time between participants—an important consideration given the known variability in recovery and treatment response in this population (Forkel et al., 2014; Lazar & Antoniello, 2008; Lazar et al., 2008; Millis et al., 2001). Second, it can predict outcomes given multiple repeated measurements for participants. Finally, it can manage missing data or unequal sample sizes over time—valuable advantages given this study's longitudinal design.

Model Building and Structure

Data were analyzed using generalized linear mixed-effects models (GLMMs), an extension of logistic regression that includes fixed and random effects and a common approach to growth curve analysis. In keeping with the recommendation to build a maximally complex random effects structure that is theoretically supported by the data set and research question (Barr et al., 2013), a GLMM was constructed to predict overall item accuracy with time point, group, and their interaction as fixed effects and etiology as a categorical covariate. Random effects included random intercepts for participant and item and by-participant random slopes of time point and group. The by-participant random slope of group was included to allow for differences in the slope for the deferred treatment/usual care and treatment phase for the six participants who contributed data to both groups. The full random effects model (i.e., with random slopes of time point and group) produced a singular fit, and so did a model without covariances between random slopes for group and the other by-participant random effects, suggesting that the random effects structure was overly complex for the data set. Thus, in keeping with best practice in mixed-effects modeling (Brauer & Curtin, 2018; Meteyard & Davies, 2020), the random slope of group was removed, and a model with random intercepts for participant and item and by-participant random slopes for time point was fit with the same fixed-effects structure. The model syntaxes for the between-groups GLMMs were subsequently constructed as follows:

BG1.  Overall effect of time point by group model:

cbind(points scored, points missed) ~ time point * group + etiology + (time point | participant) + (1 | item)

BG2.  Effect of time point by subdomain and group model (intermediate model with two-way interaction):

cbind(points scored, points missed) ~ time point * (subdomain + group) + etiology + (time point | participant) + (1 | item)

BG3.  Effect of time point by subdomain by group model (three-way interaction):

cbind(points scored, points missed) ~ time point * subdomain * group + etiology + (time point | participant) + (1 | item)

Within-group GLMMs were conducted separately for the treatment and deferred treatment/usual care groups with similar syntax (i.e., removed interaction term between group and the other predictor variables):

WG1.  Overall effect of time point model:

cbind(points scored, points missed) ~ time point + etiology + (time point | participant) + (1 | item)

WG2.  Effect of time point accounting for subdomain model (intermediate model with intercepts for subdomain):

cbind(points scored, points missed) ~ time point + subdomain + etiology + (time point | participant) + (1 | item)

WG3.  Effect of time point by subdomain model (intercepts and slopes for subdomain):

cbind(points scored, points missed) ~ time point * subdomain + etiology + (time point | participant) + (1 | item)

These model structures allowed for testing the effects of interest. In BG1, the Time Point × Group interaction term captured group differences in the effect of time on performance, that is, the effect of treatment (relative to control) on rate of improvement. BG2 included effects of subdomain and a Time Point × Subdomain interaction term to model differences between subdomains, and thus provided the comparison point for BG3, which also included the three-way Time Point × Subdomain × Group interaction to model subdomain differences in the effect of treatment (group differences in rate of change). Nested model fit was compared using likelihood ratio tests (implemented with the analysis of variance [ANOVA] function in R). A statistically significant improvement in model fit for BG3 compared to BG2 would indicate that the subdomains differentially responded to treatment, which can be further evaluated by estimating domain-specific intercepts and slopes from the between-groups subdomain model (BG3).

For within-group models, the time point effect in WG1 captured the rate of change over time for that group. WG2 included overall accuracy differences between subdomains, and WG3 included differences between subdomains in rate of change (Time Point × Subdomain interaction). As for the between-groups models, nested model fit comparisons based on likelihood ratio tests (implemented with the ANOVA function in R) were used to evaluate whether that interaction term in WG3 statistically significantly improved model fit compared to WG2. If it did, subdomain differences were further evaluated by estimating domain-specific intercepts and slopes from the within-group subdomain model (WG3).

In each model, item accuracy served as the dependent variable. One stipulation of a logistic mixed-effects regression model is that the outcome variable is expressed in integers. Standard scoring for six items of the WAB-R use half-points (i.e., five items from the dictated letters subtest were given a score of 0.5 or 0; the alphabet and numbers item: each letter or number correctly written was scored with 0.5 for a total score of 22.5 points), so these scores were scaled up by a factor of 2. Otherwise, traditional rounding rules were applied to all other decimal values in the points scored column (i.e., greater than or equal to 0.5 round up to the nearest integer; less than 0.5 round down to the nearest integer).

All of the models included ABI etiology as a dummy-coded categorical covariate with two levels (i.e., TBI and non-TBI), random intercepts for participant and item to allow for differences in starting accuracy across participants and items, and by-participant random slopes for time point to model individual differences in rate of accuracy change over time. Time point was coded as a numeric predictor (i.e., pre–time point = “0,” post–1 time point = “1,” post–2 time point = “2,” post–3 time point = “3”). As depicted in Figure 1, 22 participants contributed data to the pre– and post–1 treatment time point, 15 went on to complete another semester of treatment, contributing data to the post–2 treatment time point, and 13 went on to complete another semester of treatment, contributing data to the post–3 treatment time point. Fourteen deferred treatment control participants completed one semester as a control, contributing data to the pre– and post–1 control time point, and five went on to complete a second semester as a control, contributing data to the post–2 control time point (n = 5). GLMMs are robust to unequal sample sizes, which limited concern about the differences in sample size between the treatment and deferred treatment/usual care groups (Curran et al., 2010; Oleson et al., 2019). Furthermore, only time points that included at least five participants data were analyzed to minimize bias of the fixed-effects estimates that were of primary interest (i.e., time point and group; Brysbaert & Stevens, 2018; Heagerty & Kurland, 2001; Maas & Hox, 2005). Group was dummy-coded as a categorical variable with two levels (i.e., controls and treatment), with deferred treatment/usual care controls as the reference level. Subdomain was dummy-coded as a categorical predictor variable with 10 levels (i.e., auditory comprehension, verbal expression, written expression, reading comprehension, attention, memory, visuospatial/constructional, upper limb/facial/instrumental apraxia, orientation, and problem solving), with attention as the reference level.

To increase interpretability, log-odds estimates from the GLMMs were transformed to predicted probability in the plots, and both log-odds and predicted probability were reported in the tables (Heiss, 2020; Sauer, 2017). Both original and Benjamini-Hochberg–adjusted p values were reported for domain-specific slope estimates for the Time Point × Subdomain × Group (between-groups GLMM) and Time Point × Subdomain analyses (within-group GLMMs for treatment and deferred treatment/usual care control groups individually). Data management, visualization, and statistical analyses were completed in R (R Core Team, 2020) with the support of the following packages: lme4 (v1.1.26; Bates et al., 2015), lmertest (v3.1.3; Kuznetsova et al., 2017), tidyverse (v1.3.0; Wickham et al., 2019), broom (v0.7.6; Robinson & Hayes, 2020), patchwork (v1.1.1; Pedersen, 2020), and multcomp (v1.4.16; Hothorn et al., 2008).

Results

Between-Groups Analyses

Overall Effect of Time Point by Group

As reported in Table 3 and demonstrated in Figure 4, participants in the treatment group demonstrated significantly lower overall item accuracy than the deferred treatment/usual care control group at baseline (B = 0.19, SE = 0.04, predicted probability [Pred. Prob.] = .55, z = 4.76, p < .001). As the number of semesters in ICCR increased (i.e., time point), item accuracy in the treatment group increased at a significantly faster rate than in the deferred treatment/usual care control group (B = 0.09, SE = 0.04, Pred. Prob. = .52, z = 2.65, p < .01), suggesting an overall effect of treatment. Participants with TBI performed worse than participants with non-TBI at baseline, although this difference was not significant (B = −0.07, SE = 0.34, Pred. Prob. = .48, z = −0.21, p = .84).

Table 3.

Main results of the generalized linear mixed-effects regression analyses comparing the treatment group to the deferred treatment/usual care control group.

Model	Syntax	Term		Logit odds (SE)	Prob.	z	p (adj.)	Random effects: variance (SD)
								Intercept: ID	Intercept: Item	Slope: Time × ID; Corr
Overall	glmer(cbind(obs_score, (poss_score-obs_score)) ~ Time Point * Group + Etiology + (1 + Time Point | ID) + (1 | Item)	Intercept		1.81 (0.25)	.86	7.11	p < .001	0.91 (0.96)	3.48 (1.86)	0.01 (0.11); −.27
		Time point		0.03 (0.03)	.51	0.81	ns
		Group, ref. level = control		0.19 (0.04)	.55	4.76	p < .001
		Etiology, ref. level = non-TBI		−0.07 (0.34)	.48	−0.21	ns
		Time Point × Group		0.09 (0.04)	.52	2.65	p < .01
Subdomain	glmer(cbind(obs_score, (poss_score-obs_score)) ~ Time Point * Subdomain * Group + Etiology + (1 + Time Point | ID) + (1 | Item)	Auditory comprehension	Intercept	0.22 (0.07)	.55	3.01		0.91 (0.95)	2.05 (1.43)	0.01 (0.11); −.22
			Slope	−0.10 (0.07)	−.03	−1.48	ns (ns)
		Verbal expression	Intercept	0.04 (0.06)	.51	0.63
			Slope	0.18 (0.05)	.05	3.35	p < .001 (p < .01)
		Reading comprehension	Intercept	−0.10 (0.08)	.48	−1.19
			Slope	0.11 (0.07)	.03	1.44	ns (ns)
		Written expression	Intercept	0.02 (0.09)	.51	0.26
			Slope	0.20 (0.08)	.05	2.54	.011 (.056)
		Attention	Intercept	0.30 (0.08)	.58	4.00
			Slope	0.01 (0.07)	.00	0.12	ns (ns)
		Orientation	Intercept	0.22 (0.36)	.56	0.61
			Slope	−0.09 (0.38)	−.02	−0.24	ns (ns)
		Memory	Intercept	0.25 (0.07)	.56	3.62
			Slope	0.12 (0.06)	.03	1.93	.054 (.135)
		Problem solving	Intercept	0.44 (0.14)	.61	3.12
			Slope	−0.04 (0.13)	−.01	−0.29	ns (ns)
		Visuospatial/constructional	Intercept	0.74 (0.09)	.68	8.10
			Slope	0.19 (0.08)	.05	2.31	.021 (.069)
		Upper limb/facial/instrumental apraxia	Intercept	0.13 (0.16)	.53	0.82
			Slope	0.20 (0.15)	.05	1.38	ns (ns)

Note. Logit odds were converted to odds ratios and then to probability values (i.e., proportion of items correct [Prob.]) Time point was coded as a numeric predictor: Pre–time point = “0,” Post–time point 1 = “1,” Post–time point 2 = “2,” Post-time point 3 = “3.” Etiology was dummy-coded (i.e., traumatic brain injury [TBI] and non-TBI) with non-TBI as the reference level. Group control as the reference level. Subdomain was dummy-coded with attention as the reference level. The correlation value refers to the strength of association between the random slope of time point and the random intercept of participant. The negative value reflects that participants with lower baseline accuracy have steeper slopes. SE = standard error; adj. = p values for the domain-specific slopes were adjusted using the Benjamini–Hochberg method; SD = standard deviation; ID = participant; Corr = correlation; obs_score = score obtained for item; poss_score = maximum possible score for item; ref = reference; ns = nonsignificant.

Figure 4.

Overall effect of treatment by group. Plots reveal performance on overall items by group. Open circles = individual participant means; filled points = group means + SE; solid lines = model predicted group means.

Although etiology was not a significant predictor of overall item accuracy, a follow-up analysis was conducted to specifically test for differences in the intervention effect by etiology. The three-way interaction of time point, group (reference level = group), and etiology (reference level = non-TBI) was used to predict overall item accuracy with the same random effects structure as in the previous BG models (i.e., random intercepts for participant and item, by-participant random slopes for time point). The interaction term was not a significant predictor of overall item accuracy (B = −0.11, SE = 0.07, z = −1.50, p = .13), suggesting that the overall intervention benefits were similar for individuals with TBI and non-TBI. Full parameter estimates for this model are reported in Supplemental Material S4.

Effect of Time Point by Subdomain and Group

Adding the two-way interaction significantly improved model fit relative to the overall model, BG2 relative to BG1: χ²(18) = 360.08, p < .001, indicating that there were significant differences between subdomains. Expanding the model to include the three-way interaction term significantly improved model fit, BG3 relative to BG2: χ²(18) = 171.83, p < .001, indicating significant differences between the groups over time at the subdomain level. Full parameter estimates for BG2 and BG3 models are available in Supplemental Material S4.

Intercept and slope estimates for each subdomain are reported in Table 3 (see Supplemental Material S4 for code used to extract these values from the BG3 model). Item accuracy increased at a significantly faster rate over time for treatment participants than deferred treatment/usual care control participants in verbal expression (B = 0.18, SE = 0.05, Pred. Prob. = .05, z = 3.35, adjusted p < .01). Treatment participants also improved at a significantly faster rate over time than control participants in written expression (B = 0.20, SE =0.08, Pred. Prob. = .05, z = 2.54, p = .011, adjusted p < .056) and visuospatial/constructional (B = 0.19, SE = 0.08, Pred Prob. = 0.05, z = 2.31, p = .021, adjusted p = .069), although these did not survive multiple comparison correction.

Overall Effect of Treatment

As reported in Table 4, as the number of semesters in ICCR increased (i.e., time point), item accuracy significantly increased (B = 0.12, SE = 0.02, Pred. Prob. = .53, z = 5.04, p < .001), suggesting an overall effect of treatment. Participants with TBI performed slightly worse than participants with non-TBI, although this difference in item accuracy was not significant at baseline (B = −0.48, SE = 0.38, Pred. Prob. = .38, z = −1.28, p = .20).

Table 4.

Main results of the generalized linear mixed-effects regression analyses for the treatment group.

Model	Syntax	Term		Logit odds (SE)	Prob.	z	p (adj.)	Random effects: variance (SD)
								Intercept: ID	Intercept: Item	Slope: Time × ID; Corr
Overall model	glmer(cbind(obs_score, (poss_score-obs_score)) ~ Time Point + Etiology + (1 + Time Point | ID) + (1 | Item)	Intercept		2.01 (0.28)	.88	7.24	p < .001	0.94 (0.97)	3.52 (1.88)	0.01 (0.10); −.42
		Time point		0.12 (0.02)	.53	5.04	p < .001
		Etiology, ref. level = non-TBI		−0.48 (0.38)	.38	−1.28	ns
Subdomain	glmer(cbind(obs_score, (poss_score-obs_score)) ~ Time Point * Subdomain + Etiology + (1+Time Point | ID) + (1 | Item)	Auditory comprehension	Intercept	2.77 (0.30)	.94	9.21		0.92 (0.96)	2.11 (1.45)	0.01 (0.09); −.37
			Slope	0.06 (0.03)	.01	1.84	ns (ns)
		Verbal expression	Intercept	2.13 (0.32)	.89	6.73
			Slope	0.21 (0.03)	.05	7.07	p < .001 (p < .001)
		Reading comprehension	Intercept	2.10 (0.31)	.89	6.79
			Slope	0.10 (0.03)	.02	2.89	.03 (.06)
		Written expression	Intercept	2.33 (0.39)	.91	5.89
			Slope	0.15 (0.04)	.04	4.01	p < .01 (p < .01)
		Attention	Intercept	0.59 (0.48)	.64	1.23
			Slope	−0.01 (0.03)	.00	−0.31	ns (ns)
		Orientation	Intercept	3.27 (0.54)	.96	6.05
			Slope	0.28 (0.15)	.07	1.86	ns (ns)
		Memory	Intercept	−0.09 (0.30)	.48	−0.31
			Slope	0.15 (0.03)	.04	4.97	p < .001 (p < .001)
		Problem solving	Intercept	3.40 (0.34)	.97	10.00
			Slope	0.22 (0.06)	.05	3.69	p < .01 (p < .01)
		Visuospatial/constructional	Intercept	1.56 (0.38)	.83	4.10
			Slope	0.10 (0.04)	.03	2.56	.08 (ns)
		Upper limb/facial/instrumental apraxia	Intercept	3.10 (0.43)	.96	7.16
			Slope	0.07 (0.06)	.02	1.16	ns (ns)

Note. Logit odds were converted to odds ratios and then to probability values (i.e., proportion of items correct [Prob.]). Time point was coded as a numeric predictor: pretreatment = “0,” Posttreatment 1 = “1,” Posttreatment 2 = “2,” Posttreatment 3 = “3.” Etiology was dummy-coded (i.e., traumatic brain injury [TBI] and non-TBI) with non-TBI as the reference level. Subdomain was dummy-coded with attention as the reference level. The correlation value refers to the strength of the association between the random slope of time point and the random intercept of participant. The negative value reflects that participants with lower baseline accuracy have steeper slopes. SE = standard error; adj. = p values for the domain-specific slopes were adjusted using the Benjamini–Hochberg method; SD = standard deviation; ID = participant; Corr = correlation; obs_score = score obtained for item; poss_score = maximum possible score for item; ref = reference; ns = nonsignificant.

Effect of Treatment by Subdomain

Adding independent intercept terms for subdomains significantly improved model fit relative to the overall treatment model, WG2 relative to WG1: χ²(9) = 282.24, p < .001, indicating that there were significant differences in accuracy between subdomains. Adding independent slope terms for subdomains significantly improved model fit, WG3 relative to WG2: χ²(9) = 63.14, p < .001, indicating significant differences in treatment effects across the subdomains. Full parameter estimates for WG2 and WG3 models for the treatment group are available in Supplemental Material S5.

Intercept and slope estimates for each subdomain are reported in Table 4 and demonstrated in Figure 5a (see Supplemental Material S5 for code used to extract these values from the WG3 model). Item accuracy increased significantly over time in the verbal expression (B = 0.21, SE = 0.03, Pred. Prob. = .05, z = 7.07, adjusted p < .001), written expression (B = 0.15, SE = 0.04, Pred. Prob. = .04, z = 4.01, adjusted p < .01), memory (B = 0.15, SE = 0.03, Pred. Prob. = .04, z = 4.97, adjusted p < .001), and problem solving (B = 0.22, SE = 0.06, Pred. Prob. = .05, z = 3.69, adjusted p < .01) subdomains.

Figure 5.

(a) Effect of time point by subdomain for the treatment group. Plot reveals performance on individual subdomains over time for the treatment participants. (b) Effect of time point by subdomain for the deferred treatment control/usual care group. Plot reveals performance on individual subdomains over time for the control participants. Open circles = individual participant means; filled points = group means + SE; solid lines = model predicted group means; asterisks = significance after false discovery rate (FDR) correction (**p < .01, ***p < .001); crosses = significance at the original p value level (^† p < .05 uncorrected).

Overall Effect of Time

As reported in Table 5, as the number of semesters in the deferred treatment/usual care control group increased (i.e., time point), item accuracy did not significantly increase (B = 0.04, SE = 0.05, Pred. Prob. = .51, z = 0.74, p > .05). Participants with TBI performed slightly better than participants with non-TBI, although this difference in item accuracy was not significant at baseline (B = 0.37, SE = 0.39, Pred. Prob. = .59, z = 0.96, p > .05).

Table 5.

Main results of the generalized linear mixed-effects regression analyses for the deferred treatment/usual care control group only.

Model	Syntax	Term		Logit odds (SE)	Prob.	z	p (adj.)	Random effects: variance (SD)
								Intercept: ID	Intercept: Item	Slope: Time × ID; Corr
Overall	glmer(cbind(obs_score, (poss_score-obs_score)) ~ Time Point + Etiology + (1 + Time Point | ID) + (1 |Item)	Intercept		2.00 (0.30)	.88	6.684	p < .001	0.64 (0.80)	3.76 (1.94)	0.03 (0.18); −.57
		Time point		0.04 (0.05)	.51	0.735	ns
		Etiology, ref. level = non-TBI		0.37 (0.39)	.59	0.96	ns
Subdomain	glmer(cbind(obs_score, (poss_score-obs_score)) ~ Time Point * Subdomain + Etiology + (1 + Time point | ID) + (1 | Item)	Auditory comprehension	Intercept	2.69 (0.32)	.94	8.41		0.64 (0.80)	1.91 (1.38)	0.03 (0.17); −.54
			Slope	0.18 (0.08)	.04	2.30	p < .05 (ns)
		Verbal expression	Intercept	2.47 (0.34)	.92	7.37
			Slope	0.04 (0.07)	.01	0.64	ns (ns)
		Reading comprehension	Intercept	2.29 (0.33)	.91	6.97
			Slope	0.01 (0.08)	.00	0.14	ns (ns)
		Written expression	Intercept	2.52 (0.41)	.93	6.10
			Slope	−0.04 (0.08)	−.01	−0.44	ns (ns)
		Attention	Intercept	0.09 (0.50)	.52	0.18
			Slope	0.01 (0.07)	.00	0.09	ns (ns)
		Orientation	Intercept	3.17 (0.59)	.96	5.39
			Slope	0.37 (0.35)	.09	1.07	ns (ns)
		Memory	Intercept	−0.24 (0.32)	.44	−0.74
			Slope	0.06 (0.07)	.01	0.85	ns (ns)
		Problem solving	Intercept	2.81 (0.36)	.94	7.85
			Slope	0.24 (0.12)	.06	1.94	.05 (ns)
		Visuospatial/constructional	Intercept	0.83 (0.39)	.70	2.13
			Slope	−0.07 (0.08)	−.02	−0.88	ns (ns)
		Upper limb/facial/instrumental apraxia	Intercept	3.25 (0.45)	.96	7.20
			Slope	−0.11 (0.14)	−.03	−0.80	ns (ns)

Note. Logit odds were converted to odds ratios and then to probability values (i.e., proportion of items correct [Prob.]). Time point was coded as a numeric predictor: Pre–time point = “0,” Post–time point 1 = “1,” Post–time point 2 = “2.” Etiology was dummy-coded (i.e., traumatic brain injury [TBI] and non-TBI) with non-TBI as the reference level. Subdomain was dummy-coded with attention as the reference level. The correlation value refers to the strength of the association between the random slope of time point and the random intercept of participant. The negative value reflects participants with lower baseline accuracy have steeper slopes. SE = standard error; adj. = p values for the domain-specific slopes were adjusted using the Benjamini–Hochberg method; SD = standard deviation; ID = participant; Corr = correlation; obs_score = score obtained for item; poss_score = maximum possible score for item; ref = reference; ns = nonsignificant.

Overall Effect of Time by Subdomain

Adding independent intercept terms for subdomains significantly improved model fit, WG2 relative to WG1: χ²(9) = 318.15, p < .001, indicating that there were significant differences in accuracy between subdomains. Adding independent slope terms for subdomains did not significantly improve model fit, WG3 relative to WG2: χ²(9) = 15.30, p = .08, indicating only minimal rate of change differences across the subdomains. Full parameter estimates for WG2 and WG3 models for the deferred treatment/usual care control group are available in Supplemental Material S5. Although the Subdomain × Time Point interaction was only marginally statistically significant, to thoroughly assess for any evidence in support of the alternative hypothesis (i.e., controls performing significantly better over time in some domains), intercept and slope estimates for each subdomain are reported in Table 5 and demonstrated in Figure 5b (see Supplemental Material S5 for code used to extract these values from the WG3 model). The auditory comprehension subdomain was the only subdomain that showed significant improvement over time (B = 0.18, SE = 0.08, Pred. Prob = 0.04, z = 2.30, p = .02, adj. p = .19), although this difference did not survive correction.

Figure 6 shows the baseline accuracies and rates of change across subdomains for the treatment and deferred treatment control groups. In both groups, there was a moderate correlation with baseline accuracy and rate of change, treatment group: r = .30, t(8) = 0.90, p > .05, and deferred treatment/usual care control group: r = .26, t(8) = 0.77 p > .05, but neither of which were significant, and their interpretation is limited by restricted range (i.e., most subdomains had baseline accuracy around .90).

Figure 6.

Scatter plot showing the relationship between intercept and slope estimates for each subdomain for each of the groups. Treatment group scatter plot is in the left panel, and deferred treatment/usual care control group scatter plot is in the right panel. Horizontal line reflects the predicted improvement in accuracy after one semester of the intensive intervention and/or one semester of deferred treatment/usual care based on the slope estimate for that subdomain.

Discussion

There were several findings in this study. First, as hypothesized, there was an overall effect of treatment: As the number of semesters in ICCR increased, overall item accuracy increased at a significantly faster rate for the treatment group than the deferred treatment/usual care control group, irrespective of domain. These results support a cumulative benefit of ICCR on language and other cognitive function and extend findings of an initial efficacy study (Gilmore, Ross, & Kiran, 2019) to a larger sample of young adults with ABI. Second, individuals with TBI and non-TBI demonstrated similar overall benefits of the intervention, an important consideration given young adults with stroke may struggle to find a rehabilitation peer group. Third, item accuracy in the verbal expression subdomain improved at a significantly faster rate for the treatment group than the deferred treatment/usual care control group, suggesting some specificity to the intervention effect. Finally, within-group analyses revealed that the treatment group significantly improved in verbal expression, written expression, problem solving, and memory, whereas the deferred treatment control participants did not. Overall, these results emphasize that the integration of impairment-based retraining of language and other cognitive skills with “real-world” application in academically focused activities promoted gains in underlying cognitive processes (e.g., verbal expression) as measured via standardized assessment items—a central tenet of the ICCR program.

The between-groups subdomain analyses are promising in that treatment participants improved at a significantly faster rate than deferred treatment control/usual care control participants in the verbal expression subdomain with significant gains also being observed in the written expression and visuospatial/constructional subdomains that did not withstand multiple comparison correction. These subdomain-level results underscore the benefits of the Intensive Cognitive Communication Rehabilitation on specific cognitive domains that are (a) relied upon for academic success, (b) often impaired in individuals with ABI, and (c) reported to impact academic performance for individuals with ABI.

The faster rate of improvement in verbal expression for the treatment than control group is encouraging given the strong emphasis on the importance of oral communication for academic success in the broader education literature (Hargie, 2006; Mahmud, 2014; Morreale & Pearson, 2008; Rubin & Graham, 1988). Verbal expression activities in the classroom, such as giving presentations and participating in group discussions, have been associated with academic achievement in college, with studies showing that students with strong oral communication have higher GPAs (Mahmud, 2014). As a next step, it will be important to follow ICCR participants long term to assess for a relationship between gains in specific cognitive domains, such as verbal expression, and subsequent college enrollment and completion of a semester. It will also be valuable to investigate potential catalysts for the robust gains observed in verbal expression in the treatment group in order to support replicability of this finding in the future. On the other hand, it is wholly possible that the gains in verbal expression were driven by the sum of the intervention's parts as opposed to any one individual component of the intervention on its own, especially given that treatment was delivered in the context of a cohort. In fact, evidence from group process suggests (a) that bringing these young adults together into a peer group would lead to increased psychosocial support and increased confidence and (b) that these gains would, in turn, spur increased communication initiation and ultimately result in increased accuracy on items in the verbal expression subdomain (DeDe et al., 2019; Elman, 2006; Elman & Bernstein-Ellis, 1999; Fama et al., 2016; Griffin-Musick et al., 2020; van der Gaag et al., 2005). Given the potential benefit for verbal expression and psychosocial functioning, it is clear that future studies of the ICCR program's efficacy should explore quantitative and qualitative benefits of the group context on verbal expression at the impairment (e.g., discourse-level metrics examining peer-to-peer interaction during class discussion), activity/participation (e.g., frequency of self-initiated social interaction between classes), and quality of life levels (e.g., satisfaction and confidence during peer-to-peer communication in classroom discussion).

Despite these positive outcomes, there does appear to be some specificity to the effect of the intervention as treatment participants' item accuracy did not improve at a significantly faster rate than that of deferred treatment/usual care controls for all subdomains targeted by the intervention (i.e., auditory comprehension, reading comprehension, attention, orientation, problem solving, and upper limb/facial/instrumental apraxia). There are several potential reasons for this result. First, attention showed no change between or within groups as evidenced by flat slope estimates (between groups: B = 0.01, SE = 0.07; treatment group: B = −0.01, SE = 0.03; control group: B = 0.01, SE = 0.07). It is possible that attention may have required more domain-specific intervention to demonstrate improvement on neuropsychological assessments of this subdomain (e.g., direct attention training; Sohlberg et al., 2000). Second, deferred treatment/usual care control participants demonstrated larger slopes than treatment participants for several domains, leading to a negative between-groups estimate for that subdomain (i.e., orientation, problem solving, and auditory comprehension). Deferred treatment controls showed the steepest slope for the orientation subdomain (B = 0.37, SE = 0.35), but treatment participants also showed a positive slope in this domain (B = 0.28, SE = 0.15). Across groups, the orientation estimate was accompanied by the largest standard error (between groups: B = −0.09, SE = 0.38), indicating there was greater variability in predictions for this subdomain. The orientation subdomain had fewer items and points possible than the other subdomains (see Figure 3), which likely led to greater uncertainty in the predictions for this subdomain (i.e., logistic models take into consideration the number of trials—another advantage of the GLMM approach used in this study). Both groups showed positive slopes in auditory comprehension and problem solving, explaining the nonsignificant effect between groups for these domains. Finally, the intervention did appear to have some positive effect on the reading comprehension (between groups: B = 0.11, SE = 0.07; treatment group: B = 0.10, SE = 0.03; control group (B = 0.01, SE = 0.08) and upper limb/facial/instrumental apraxia function (between groups: B = 0.20, SE = 0.15; treatment group: B = 0.07, SE = 0.06; control group: B = −0.11, SE = 0.14), although these effects did not reach statistical significance. Overall, the subdomain analyses are promising and will serve to inform the development of future studies investigating the efficacy of the ICCR program.

Beyond the domain-specific gains, the overall treatment effect of this intensive, comprehensive cognitive rehabilitation program generates a larger question for neurorehabilitation: What is driving the significant recovery of language and other cognitive function in this chronic young ABI population? The neuroplasticity literature would suggest that the repetitive, intensive, salient training within the academic context was responsible for the gains (Kiran & Thompson, 2019; Kleim & Jones, 2008). Recent studies investigating intensity and dosage in cognitive rehabilitation (Brady et al., 2016; Königs et al., 2018) would support that it was the roughly 240 hours delivered per ICCR semester that spurred the participants' improvements in this study. The intervention incorporated evidence-based cognitive rehabilitation approaches, such as one-on-one language therapy and group-based metacognitive strategy training for executive function. It also employed academic material as the vehicle for therapy, which aligns with contextualized cognitive rehabilitation targeting language and other cognitive process in the context where the breakdown occurs (i.e., academic activities). Finally, ICCR is a comprehensive, multimodal cognitive rehabilitation program in line with a recent systematic review that recommended neuropsychological rehabilitation to reduce cognitive and functional disability after TBI or stroke (Cicerone et al., 2019). It was likely the integration of these components: (a) principles of neuroplasticity such as intensity, repetition, and salience; (b) evidence-based individual and group cognitive rehabilitation; and (c) contextualized skill/strategy training and application that promoted cognitive recovery for young adults with chronic ABI in this study. Further regarding the benefit of this last component of the program, ICCR targeted language and other cognitive domains within a functional setting (i.e., simulated college class), which transferred to item-level gains on standardized assessments of cognitive function proximal to the intervention. This finding is promising regarding the transfer of gains from integrated cognitive rehabilitation to other contexts. An important next step will be to assess the extent to which these gains in language and other cognitive function lead to successful enrollment in college for young adults with ABI in future work.

Within-group analyses were implemented in this study to identify subdomains that were responsive to change due to treatment and nontreatment factors. As shown in Figures 5 and 6, the verbal expression, written expression, memory, and reading comprehension subdomains significantly improved in the treatment group over time (at the original p < .05 level). The opposite pattern was observed in the deferred treatment/usual care control group for those subdomains. Taken together, this pattern underscores that verbal expression, written expression, memory, and reading comprehension domains were stimulable to the intervention. Future work should investigate what specific components of ICCR are associated with gains in these specific subdomains (e.g., weekly practice quiz scores and gains in the memory subdomain) and begin to identify the active ingredients of this multifaceted intervention. Additionally, problem solving and orientation showed a similar pattern of improvement in both groups, suggesting that these gains may have been associated with nontreatment factors (e.g., outpatient therapy, work, volunteer, and practice effects). The significant effect of time point for the problem solving domain observed in the treatment group should therefore be interpreted with some caution and would benefit from further investigation in future work.

Recall etiology was included as a covariate in all of the analyses but never significantly predicted item accuracy. To address any residual concern about including individuals with non-TBI and TBI in the same intervention—anticipated given the common separation of interventions for individuals with stroke and TBI in the speech-language pathology field—a final between-groups model was fit to the data using the interaction of Time Point × Group × Etiology to predict overall item accuracy. The interaction estimate was not significant, meaning the effect of the intervention was similar for individuals with non-TBI and TBI. This finding is important because young adults with poststroke aphasia often lack a rehabilitation peer group as most stroke survivors are older (Benjamin et al., 2019) and may have different long-term rehabilitation goals. Combining across these two ABI etiologies within an intervention would provide young stroke survivors a peer group with similar goals (e.g., post–secondary education) as young adults are a frequently affected age group to sustain TBI (i.e., ages 15–24 years; Taylor, 2017). Furthermore, similar effects of the intervention across etiologies may encourage clinical practice to move away from separating the cognitive impairments observed after stroke and TBI according to etiology (Coelho et al., 1996, 2005; Frey, 2020; Norman et al., 2013; Turkstra et al., 2005) and instead toward considering the cognitive deficit profile of patients regardless of etiology when planning assessment and intervention. Of note, no major limitations were observed in combining young adults with traumatic and nontraumatic etiologies in the same intervention. In fact, combining the etiologies provided natural teaching moments for clinicians to emphasize that all of the individuals in the program had areas of strength and areas for growth while also empowering individuals to support one another in ways that they could be successful. For example, participants with stroke-induced aphasia were able to support individuals with memory impairment after TBI by recalling a fact or showing them where to find information reviewed earlier. As a complement, individuals with memory impairment after TBI were able to support individuals with stroke-induced aphasia during times of word retrieval difficulty. Nevertheless, it will be important to add to these clinician-generated benefits by investigating the participants' perceptions and experiences of being grouped with others with different brain etiologies in future studies to ensure a complete perspective.

Unlike that of etiology, the influence of time postonset and ABI severity on treatment outcomes was not formally examined. In terms of time postonset, all participants in this study were in the chronic phase of recovery and thus outside the window when the majority of spontaneous recovery is believed to occur. Moreover, previous studies have not found treatment outcomes in the chronic phase to be influenced by time postonset (Doogan et al., 2018; Holland et al., 2017; Moss & Nicholas, 2006; Turner-Stokes, 2008), and thus, time postonset was not included as a covariate in this study. In regard to severity, ABI severity is not accurately captured through one single standardized assessment measure (e.g., WAB-AQ and RBANS-Total), and for reasons of multicollinearity, it would have been inappropriate to include multiple metrics. Thus, this study focused on first acquiring a robust understanding of impairments across a range of language and other cognitive processes (e.g., attention, memory) in young adults with ABI while accounting for variance in performance based on the nature of their injury. Furthermore, it is unlikely that severity played a role in the difference in overall treatment effect between the groups as (a) there were similar proportions of individuals with mild, moderate, and severe language and other cognitive impairment in both groups (see Table 1); (b) the treatment participant group demonstrated more severe impairment than the deferred treatment control group at the first time point but improved at faster rate (see Figure 4); and (c) subdomain-level results reveal the absence of a ceiling effect (e.g., some domains with high starting accuracy improved significantly over time, some domains with low starting accuracy remained stable over time; see Figure 6). Future studies of the ICCR program with larger participant samples will be better equipped to identify predictors of treatment success, which are likely to be multifactorial (e.g., severity, family support, and motivation) as opposed to unitary in nature (i.e., ABI severity).

Despite the encouraging results of this study and previous work (Gilmore, Ross, & Kiran, 2019), there is much about the ICCR program that requires further exploration. First, these gains in language and other cognitive function may have been supported by changes in the brain. Future studies should test this hypothesis by assessing to what extent there are brain changes before and after intervention and whether those changes are associated with gains in cognitive function. Longitudinal neuroimaging studies of this nature have the potential to inform future models of rehabilitation-induced recovery. Second, it will be important to conduct studies that elucidate the active ingredients of the ICCR program and begin to answer more fine-grained questions about intensity, dosage, and other principles of learning. Third, larger group studies will allow for the systematic investigation of factors (e.g., family support) that promote a positive response to the intervention, including eventual return to and success in college. Finally, to date, a number of treatment participants have enrolled in college postprogram (i.e., nine of 15 possible), but only one control participant (i.e., out of 10 possible) has pursued enrollment. It will be essential in future studies to more systematically investigate long-term outcomes by determining the extent to which ICCR participants enroll in college immediately postprogram and then go on to successfully complete a semester of college.

Conclusions

The results of this study revealed an overall effect of the ICCR program and, specifically, that treatment participants significantly improved in verbal expression at a faster rate than deferred treatment/usual care control participants. At the within-group level, treatment participants demonstrated significant longitudinal gains in memory, verbal expression, written expression, and problem solving, whereas deferred treatment/usual care control participants showed no significant longitudinal gains at the overall item or subdomain item level. These results emphasize the efficacy of this novel, intensive, comprehensive cognitive rehabilitation program in the largest participant sample to date. Furthermore, this study's findings provide strong evidence that integrating impairment-based retraining of language and other cognitive skills with “real-world” application in academically focused activities promotes change in underlying cognitive processes as measured by standardized assessment items—an impetus for a paradigm shift from typical rehabilitation for young adult ABI survivors.

Funding Statement

The Intensive Cognitive and Communication Rehabilitation (ICCR) program is funded internally through the office of the Dean of Sargent College of Health and Rehabilitation Sciences. Natalie Gilmore was funded by National Institute on Deafness and Other Communication Disorders Grants T32DC013017 (PI: Moore) and F31DC017892 (PI: Gilmore) over the course of the project.