Traumatic brain injury (TBI) is a worldwide public health problem and can result in long-term disability (1–3) with the need for extensive and highly specialized initial health care provision, followed by comprehensive rehabilitation efforts (4). Physical, cognitive and emotional problems, including inability to return to full- or part-time work, as well as diminished quality of life, are frequent long-term consequences of TBI. Effective delivery of rehabilitation services and integration of medical perspectives, as well as vocational, educational and community support, are deemed necessary to meet the complex needs of this population. Services derive from the act of serving and refer to the provision of intangible products offered to persons with health conditions. Rehabilitation services in particular imply strategies targeting subsequent disability (5). Service delivery can be viewed from societal, institutional and individual perspectives. These different levels are often referred to as macro-level, including policy and financial aspects, meso-level, including organization and availability of services, and micro-level, including accessibility and content of services provided to an individual patient (6). Donobedian (7) described the quality of services as a causal relationship among the attributes of setting, the process of care, and the outcome. Evaluating the quality of rehabilitation services is important at every level, but the complexity of services and hence the challenges of evaluation can increase when moving from the micro- to the meso- and macro-levels.

A wide variety of rehabilitation interventions have been developed and evaluated with respect to content and outcomes for different functional problems after TBI (8). Service delivery to patients with severe TBI has focused on the acute phase, underpinning the importance of early initiated and well-organized delivery (9–11). Less is known about effective rehabilitation service delivery in the postacute and later phases, at least for the general TBI population (12, 13) and reviews focusing on this issue are warranted (14). Furthermore, the structure and process of care are seldom described, although they clearly impact the outcomes of TBI (15). The lack of a framework for depicting differences in service delivery could contribute to the scarce knowledge regarding optimal rehabilitation service delivery. Recently, Gutenbrunner et al. (16) proposed a classification for rehabilitation services, the International Classification System for Service Organization in Health-related Rehabilitation (ICSO-R), describing the meso-level of heath care. The ICSO-R is based on three dimensions Service provision,” Funding and Delivery each of which has a more extensive list of categories and subcategories that characterize rehabilitation services (17). The classification builds on the conceptual framework by Meyer et al. (5) describing health-related rehabilitation services according to their organizational setting including technical and human resources in addition to their goals.

The classification was developed in order to cover the gap between classifications at the micro and macro level of health care exemplified by the International Classification of Functioning, Disability and Health (ICF) (18) and the International Classification of Health Accounts (ICHA) (19). The ICSO-R intended to provide tools for analyzing provision and delivery of rehabilitation services. Based on these assumptions the classification might also be useful for contrasting and comparing rehabilitation services across different care facilities at the local, regional and country levels. Thus far, the lack of framework has hampered a systematic approach to the service aspects in the existing rehabilitation literature and how these aspects influence outcomes (20). There is an urgent need for prognostic models in TBI facilitating comparative audits of services among hospitals, other health care settings and countries (21). The Service provider dimension of ICSO-R describes the framework of the institution, organization, the resources and quality assurance and could be applied to evaluate where, by whom and in which context the service is delivered. The Funding dimension describes the main sources of income and funding of the services (i.e., diagnosis-related groups, per-day payment or other forms of services refund.). Finally, the Delivery dimension contains the main strategies (i.e., preventive, curative, rehabilitation, supportive or other strategies) delivered to the users, aspects of intensity and duration of intervention and the way the service is organized, and can be used in order to evaluate what, for what and how the services are delivered (17). Hence, the ICSO-R may serve as a tool for such comparative audit.

Using the ICSO-R as a framework, the current review aims to provide an overview of randomized, controlled trials (RCTs) with rehabilitation service relevance provided to patients with moderate and severe TBI in the post-acute phase and to evaluate the extent to which organizing, funding and providing of rehabilitation services were addressed.

A systematic literature search was conducted to identify controlled trials evaluating the effects of rehabilitation services or rehabilitation interventions with service implications.

Studies targeting adults (>17 years old) with moderate or severe TBI and providing rehabilitation following the acute phase were included. TBI was defined as “an alteration in brain function, or other evidence of brain pathology, caused by an external force” (22). “Following the acute phase” was defined as rehabilitation occurring after discharge from the trauma center/acute-care hospital. A multidisciplinary approach was defined as at least two professions involved directly in the delivery of the intervention. Service relevance was operationalized to differences in the delivery, funding, or provision of the services between the intervention arms in the studies. A librarian was consulted to elaborate a thorough search strategy. Potential articles of interest in the English language were identified through a systematic search of the Medline (OVID), EMBASE, Cumulative Index of Nursing and Allied Health Literature (CINAHL), PsycINFO, and Cochrane Central Register of Controlled Trials (CENTRAL) databases (November 2016) with updated searches for 2016 through July 3, 2018, revealing a total of 2,970 hits (Appendix 1). In order to retrieve the highest possible number of relevant articles two filters were applied: the Cochrane Highly Sensitive Search Strategy for identifying randomized trials in MEDLINE: sensitivity-maximizing version (2008 revision) (https://training.cochrane.org/handbook); and the best balance of sensitivity and specificity filter for Therapy. These strategies build on optimizing the thesaurus terms, and adapted to the individual databases as recommended http://handbook1.cochrane.org. http://hiru.mcmaster.ca/hiru/HIRU_Hedges_home.aspx#Hedges. These filters were combined with “OR” in the search strategy in order to increase the number of hits. The filters were applied to the Medline and EMBASE searchs, but not in the search in the other databases. Hence these filters are assumed to increase and not limit the relevant hits.

The titles of the 2,970 studies were screened for eligibility, with supplementary evaluation of the abstracts when necessary (238 studies). A total of 88 studies were identified as candidates. These studies were evaluated independently by two of the authors regarding fulfillment of the inclusion criteria, and when in doubt, full-text manuscripts were assessed. Based on consensus 46 studies were found to meet the inclusion criteria. In 23 of the studies the sample size was lower than 60. Inter individual variations in functional outcomes vary across the specific outcome measures and so do the power estimates for needed number of patients. However, power around 80% and with detection of group differences around half the standard deviation is often recommended (23). Thus, studies with n<60 were excluded post-hoc in order to avoid studies with too low power. Hence, 23 studies were included in the final analyses (Figure 1).

The quality of the studies was evaluated by three researchers according to their adherence to the CONSORT guidelines (24) and was scored according to the Cochrane recommendations (25). Risk of bias in the included studies was assessed in 12 domains: adequate randomization method; concealed allocation; blinding of participants; blinding of care providers; blinding of outcome assessors; dropout rate described and <20% for short-term and 30% for long-term outcomes; intention to treat, i.e., all participants were analyzed according to their randomized group allocation; unselective reporting, i.e., results were provided from all prespecified outcomes; groups similar at baseline regarding demographics and other important clinical characteristics; similar or absent co-interventions; adequate compliance with the interventions; and similar timing of primary outcomes for the intervention and control groups. The score of 1 was given if the domain fulfilled the Cochrane recommendations and 0 if not. Hence, the total score had a possible range from 12 (highest level of quality) to 0 (lowest level of quality). Two pairs of raters evaluated the studies. Each rater conducted independent evaluations, and agreement between raters in each pair was provided. When in doubt, the domain was scored as 0. A consensus-based total score was subsequently elaborated. A data extraction sheet was elaborated capturing the type of randomization, number of involved study centers, sample size and age of the participants, as well as outcome measures.

The data were summarized descriptively, and agreement between researchers was assessed by Pearson's correlation coefficient (r). The frequency of described dimensions and categories of service provision according to ICSO-R (15), with the predefined selection of descriptors suggested by Røe et al. (26), are reported. Service provision was categorized regarding Location (i.e., country), Organization (public or private), Context (hospital or community), Facility (rehabilitation or general medical) and Profit orientation (yes/no). The Funding dimension was assessed according to public, private or insurance-based sources of money. For the Delivery dimension Strategy of the intervention (diagnostic/therapeutic/management/prevention), Target group (TBI specific or not), Service goals (improvement of body function/activities and participation/adaptation to environmental factors), Team structure (interdisciplinary/multidisciplinary/single disciplines) and Mode of production, categorized as inpatient or outpatient delivery (inpatient or not), were assessed. We also attempted to disentangle whether these dimensions and categories varied between the intervention arms in the studies and summarized the main dimension/category differences. For the categories of Intensity, Aspects of time and Other, we could not apply the predefined terms from Røe et al. (26) we did not find applicable equivalents and used the descriptive approach by Kiekens et al. (17). Main outcome measurements from the selected studies are reported, along with the dimensions of the International Classification of Functioning, Disability and Health (ICF) (Body function, Activity and Participation, Environmental factors).

The present review provides an overview of randomized, rehabilitation trials with service provision relevance in the post-acute phase after moderate and severe TBI. Half of the studies reported statistically significant differences between interventions in one or more of the outcome measurements. That most of the studies focused on the rehabilitation strategy i.e., content, with implicit, more than explicit, variations in the service provision and delivery, was a challenge. Furthermore, the lack of universal terminology and reporting standards for the service aspects, as well as the diversity of interventions and outcome measures, prohibited analysis of the effects of service provision across studies, as well as metaanalytic approaches.

Rehabilitation service provision is complex and varies across health care settings and countries, with a lack of synthesized information regarding effective organization of services based on randomized trials (50). Successful outcomes at the patient level are dependent on the organization, capacity and quality of rehabilitation services at the macro-, meso- and micro-levels (51). Several reviews have been conducted regarding the effects of different interventions and treatment modalities targeting physical, cognitive and emotional problems after TBI (8, 52–59). However, very few evaluated directly the effects of differences in service provision and delivery supporting the gap in knowledge regarding post-acute services for TBI at the meso-level. Service provision and delivery related factors may thus influence outcome across reported significance. The review also illustrated that, when applying a structured framework, differences in service provision and delivery could be deduced from intervention studies, primarily evaluating programs at the micro-level.

The implicit components of services included in the treatment and interventions compared in clinical studies are an enormous challenge regarding the evaluation of effective service models. Thus, synthesizing evidence regarding effective components in service provision is also difficult. ICSO-R provided a tool for systematizing important elements of services across intervention arms. Describing the studies according to ICSO-R did, however, indicate that important elements of services varied across intervention arms and could influence the outcomes. One might argue that the majority of the studies were conducted before the ICSO-R was published. However, the aspects of services addressed in the ICSO-R have been relevant to service provision for decades (7). Laver et al. (14) conducted a systematic review regarding evidence for organizing health care for people with acquired brain injury, identifying 8 studies of TBI. When excluding studies with mainly mild TBI and those conducted in the acute phase, the studies included in Lavers' review overlapped with the present review. However, a main limitation is the current lack of subcategories in the ICSO-R, overlapping categories, and a lack of definitions. We applied some predefined subcategories suggested by Røe et al. (26). These predefined categories clearly failed to capture the main differences between studies regarding the intensity and timing of the intervention, as well as regarding team structure. These subcategories were developed to reduce overlap between categories, which is inherent to the original ICSO-R. This adjustment might have biased our results with over reporting of context differences and underreporting of organizational and facility differences. The new version of the ICSO-R that is being elaborated might provide a better tool for analyzing the effects of service provision in the future. The present review also illustrated that improving aspects of better information regarding service provision and delivery could be gained from the existing literature.

We categorized the studies according to the main service dimension and categories differing between the intervention arms (Tables 1, 2). Based on this approach, we identified aspects of service provision and delivery that clearly could impact evaluations of the effects in these studies. Very few studies clearly stated whether the services were private or publicly organized. These aspects could impact patient selection beyond the socio-demographic characteristics reported in the studies. We did not identify any studies focusing on the funding dimension, although both public and private organization of services and type of funding are very important aspects for policy makers and stakeholders (60). Patients' payments and refunding of the services influence outcomes, but they were poorly described in the included studies. Future studies should address this aspect more directly because resources constitute a barrier to the implementation of services (61).

Neither the dimensions nor the categories in the ICSO-R are mutually independent. In the studies comparing specified interventions with waiting lists or usual care treatment (Table 1), the differences in the service provider dimension were evident, often with different contexts and more resources in the active interventions, compared to treatment as usual care and waiting lists. However, covariance with delivery aspects was inevitable in these studies. Some of the studies included in the present review rather explicitly targeted the mode of production (51–54), but comparing in- and outpatient services generally also implies differences in provision, i.e., context and facilities.

Covariance among categories within each dimension of the ICSO-R was even more evident. For example, in Bowen et al. (42), the mode of production with pre- and post-discharge comparisons of rehabilitation interventions co-varied with differences in the timing category (Table 2). In the studies with several intervention arms, the nature of differences could vary between the intervention arms, rendering the classification challenging. In the study by Bowen et al. (42), two of the intervention arms varied regarding the mode of production, whereas the third arm (treatment as usual) could be evaluated as having different service provision aspects (Context and Resources).

The ICSO-R was specifically developed to cover rehabilitation services at the meso-level (15). The primary goal of the majority of included studies in this review focused on the content of the interventions, i.e., the micro-level of services (5). The distinction between service delivery at the meso-level and content at the micro-level might not always be clear cut (22). Several of the studies identified in our literature search evaluated the effects of different neuropsychological approaches. Since intensity is a category in the ICSO-R, studies with slightly different intensities of rehabilitation were included, although intensity could be evaluated as an important aspect of the content. Difficulties in clear-cut distinctions between the meso- and micro-levels of services are accompanied by a lack of invariance across such aspects. Content of treatment is not included in the ICSO-R but is needed to assess the effects of rehabilitation. To address the effects of different service provision components on outcomes, a more specified, detailed and universally applied system for service provision and delivery is needed. To some extent, the needed process could be compared with the development of ICF (18). Hence, 17 years after the launching of ICF, its application as a framework for systematizing outcome evaluation is increasing (62). A revised version of the ICSO-R represents one step toward this goal. However, a universal and not too complicated taxonomy for the content of effective ingredients is also urgently needed (63).

Consequently, the effects of differences in rehabilitation services cannot be directly determined from the present review. Nevertheless, it is reasonable to assume that resources and contextual factors did contribute to the documented differences between the intervention arms. It was an important aspect of the aim to evaluate in-hospital vs. at-home services in several of the studies, i.e., differences in mode of production and context (42, 43). These studies failed to document major differences in outcomes, except for better patient satisfaction with at-home services (Table 2). This finding is in contrast to the experience of patients with stroke, in whom early supported discharge showed superior efficacy over in-hospital services (64). This difference might be caused by greater variability in the needs and goals of patients with TBI and also methodological limitations in the studies included in the present review. In contrast, Winter et al. and Moriarty et al. (25) documented improvements in individually targeted outcomes for patients, as well as their relatives in people centered in home care, compared to “treatment as usual.” However, in these studies, the intensity/amount of rehabilitation could also have been different and impacted the outcome (Table 2). Variable influence in outcomes was indicated by variations in team structure, intensity and aspects of time, while Hanks et al. (49) study supported improved outcomes by peer involvement in the interventions.

The present review also underpins that replication of interventions across service providers, and delivery aspects might be important. The study by Bell et al. (28) indicated positive results of a telephone follow-up in a single center study, but it was not replicated with a multicenter design (29). Underpinning the need for validation studies, Wade et al. (48) documented significant results in their replicated study with early intervention (aspects of time). All, except one of the included studies targeted symptoms or functional problems as outcome. Increased use of common data elements and linking approaches between measurements may facilitate better comparison between studies in the future (65, 66).

The quality of the included studies varied, and reaching a maximum score might not be possible due to the lack of possibility of blinding patients and rehabilitation providers. We applied the CONSORT guidelines and scored the quality according to Furlan et al. (25) Rehabilitation might require an adapted scoring system acknowledging the special challenges in this field (67). For example, although not blinded to the intervention itself, blinding to its aims and mechanisms could be the best possible choice and could be acknowledged. The quality evaluation clearly revealed that improvement is needed in describing the concealment of randomization. Possibly more important for the results and their interpretation are better assessment and description of co-interventions. To evaluate the components related to the differences and effects in randomized rehabilitation trials, improved description of the provision and delivery of the services, along with the content of the interventions, is needed and should be included in quality evaluation systems.

The main limitations of the present review are the lack of inclusion and exclusion criteria for the ICSO-R categories and the lack of common descriptors of the services in the studies. This is an obstacle for the inclusion process of studies as well as for analyzing and reporting the influence on outcome in the studies.

A lack of uniformity of data and collection methods, the heterogeneity of structures and processes of rehabilitation services, and the lack of common outcome measurements made the study results less generalizable and the comparison between studies difficult. Standardized descriptions of services, including provider, funding and delivery dimensions, could improve the service relevance of rehabilitation studies and give valuable information to many different stakeholders. A shorter version of ICSO-R with value sets may be needed for inclusion in rehabilitation studies description.

CR, CT, EH, OT, PA, and NA contributed to the theoretical background, framework for analysis and writing process. CR, CT, and EH conducted the data analysis and, together with NA, elaborated the first manuscript draft.