Effects of inpatient rehabilitation after acute care on functional and quality-of-life outcomes in children with severe traumatic brain injury

ABSTRACT Objective Few studies have assessed the effectiveness of the rehabilitation process in children surviving severe traumatic brain injury (TBI). We evaluated whether receiving inpatient rehabilitation after acute hospitalization was associated with better functional outcomes compared to receiving only non-inpatient rehabilitation in children with severe TBI and explored an effect modification for Glasgow Coma Scale (GCS) score at hospital discharge. Methods We included 254 children who received rehabilitation following severe TBI from a multinational observational study. The Pediatric Glasgow Outcome Scale – Extended (GOS-E Peds), parent/guardian-reported and child-reported Pediatric Quality of Life Inventory (PedsQL) at 12 months post-injury were assessed and described using summary statistics. Unadjusted and propensity score-weighted linear/ordinal logistic regression modeling were also performed. Results 180 children received inpatient rehabilitation and 74 children received only non-inpatient rehabilitation after acute hospitalization. Among children with a GCS<13 at discharge, those receiving inpatient rehabilitation had a more favorable GOS-E Peds score (OR = 0.12, p = 0.045). However, no such association was observed in children with a higher GCS. We found no differences in PedsQL scores between rehabilitation groups. Conclusions Future studies are warranted to confirm the benefits of inpatient rehabilitation for children with more severely impaired consciousness when medically stable.


Introduction
Children surviving a severe traumatic brain injury (TBI) are at increased risks of physical and neurobehavioral impairments, and therefore long-term functional disabilities (1)(2)(3). Rehabilitation services which are designed to reduce disability and improve quality of life, can be an important component of recovery (4), yet there are no comprehensive guidelines that specify optimal care, including rehabilitation services, after acute medical management for children with severe TBI (5). Inpatient rehabilitation, which generally refers to therapy services and continued nursing care received during an inpatient hospital stay (6), is a commonly used treatment option after acute care for pediatric patients. Other options for children who require rehabilitation include non-inpatient rehabilitation services such as those provided at home or in outpatient settings (6).
There is significant variability in the use of rehabilitation services following TBI in pediatric patients (7)(8)(9)(10)(11). Studies of children experiencing a TBI of any severity have shown that younger age, less severe brain injury, and unfavorable family socioeconomic status were associated with a reduced likelihood of receiving inpatient and non-inpatient rehabilitation (8,10,11). Our previous work demonstrated that, among children with severe TBI, those in the UK were less likely to receive inpatient rehabilitation after acute hospitalization compared to those in the US (7). This could be explained by multiple factors, such as differences in clinical decision-making practices, availability of inpatient rehabilitation in patient-preferred geographic areas, eligibility criteria for inpatient rehabilitation admission, and related health-care policies.
Despite growing clinician and family recognition of rehabilitation services, there is a paucity of evidence on the contribution of rehabilitation process, including the use of inpatient rehabilitation and/or non-inpatient rehabilitation after acute care, to children's long-term outcomes after severe TBI. Studies in adults with TBI have demonstrated that early initiation of inpatient rehabilitation after acute care is associated with better long-term functional outcomes compared to no inpatient rehabilitation at all, or late initiation of inpatient rehabilitation (12,13). Similarly, one study of children who required inpatient rehabilitation following severe TBI indicated that delayed transfer from acute care to rehabilitation adversely affected functional recovery (14). However, no previous pediatric study has examined whether receiving inpatient rehabilitation after acute care, regardless of the need for additional noninpatient rehabilitation, is associated with better long-term outcomes compared with receiving only non-inpatient rehabilitation services.
Given that children with rehabilitation needs after severe TBI are unlikely to receive no rehabilitation at all (11), we aimed to determine if those receiving inpatient rehabilitation (with or without additional non-inpatient rehabilitation) have better global function and health-related quality of life (HRQOL) compared to those receiving only non-inpatient rehabilitation in a multisite, multinational, longitudinal cohort, the Approaches and Decisions in Acute Pediatric TBI (ADAPT) trial. Although patients with severely impaired consciousness are often denied access to inpatient rehabilitation due to uncertain prognosis, studies have suggested that they can achieve significant functional recovery during and after inpatient rehabilitation (15,16). Therefore, we explored the potential effect modification for Glasgow Coma Scale (GCS) score at hospital discharge on the associations between rehabilitation process and outcomes.

Study participants
The ADAPT trial was an observational study to evaluate the effectiveness of multiple acute management strategies in children with severe TBI. The study design has been described in detail elsewhere (17)(18)(19). Briefly, 1000 children meeting the inclusion/exclusion criteria (inclusion: age <18 years, diagnosis of severe TBI [GCS ≤ 8], placement of intracranial pressure [ICP] monitor at study site; exclusion: pregnancy) were enrolled from 51 sites within the US and abroad during 2014 to 2016. All sites received Institutional Review Board approval (or equivalent) to conduct the study and collect data regarding the acute hospitalization. This analysis focused on different patterns of rehabilitation process after acute hospitalization and their associations with long-term global function and HRQOL outcomes. Informed consent for the collection of data on rehabilitation use and follow-up outcomes were obtained from study participants or their parents/guardians at the time of discharge from the clinical site. For this study, children who did not have English as the primary language in home were excluded, because their HRQOL were not assessed due to the concerns of the reliability and validity of the translated version of the instrument. Thus, children included in this study were enrolled from the US, UK, Australia, and New Zealand.

Rehabilitation process
Rehabilitation process after acute hospitalization was defined based on the child's initial discharge destination (per the medical record) and parent/guardian-reported use of rehabilitation within the 12-month follow-up period (7). At the 12-month assessment, parents/guardians were asked to recall if their child received any rehabilitation services because of the TBI since hospital discharge. Of those receiving rehabilitation after acute hospitalization, children were classified into the "inpatient rehabilitation" group if their initial discharge destination was indicated as an inpatient rehabilitation facility (IRF) or skilled nursing facility (SNF) in the medical records (20)(21)(22); "noninpatient rehabilitation" group if they were directly discharged home and if their parents/guardians reported receipt of rehabilitation within the 12-month follow-up period. Children who had an initial discharge destination of another hospital were not included in the comparison due to the lack of information regarding whether they received inpatient rehabilitation or not.

Outcomes
GOS-E Peds scores range from 1 to 8, with higher scores indicating a poorer outcome (23). The GOS-E Peds is a developmentally appropriate instrument to assess the impacts of TBI on function in major areas of life in children (23). For this study, trained research personnel completed the GOS-E Peds by interviewing parents/guardians in person or by phone and having age-appropriate interactions with children whenever possible at 12 months after their enrollment into the ADAPT trial.
HRQOL was measured using the PedsQL at 12 months. The PedsQL parent proxy-report form is validated for children ≥2 years of age, and the child self-report form is validated for children ≥5 years of age; therefore, analyses of the parent/ guardian-reported and child self-reported HRQOL were limited to these age ranges, respectively. The PedsQL instrument is comprised of 23 items (or 21 items when administered to parents/guardians of children under 5) that assesses physical (8 items), emotional (5 items), social (5 items), and school function (5 items or 3 items for children under 5), and generates scores for each domain, as well as a psychosocial summary score and a total score for overall HRQOL (24). The PedsQL scores range from 0 to 100 with higher scores indicating better HRQOL (24). Favorable reliability and validity of the PedsQL to measure HRQOL have been demonstrated in children with TBI (25). For comparisons, minimal clinically important differences (MCIDs) in domain specific or summary PedsQL scores have been established (24). In the ADAPT trial, parents/guardians of all children aged 2 years or older at the time of assessment self-administered the PedsQL parent proxyreport form questionnaire after instructions from site research personnel. Children who were 8 years or older also completed the PedsQL self-report form questionnaire per PedsQL administration guidelines, and children aged 5 to 7 years were interviewed by site research personnel who documented responses using the Young Child self-report form.

Covariates
We collected data on children's sociodemographic and clinical characteristics including functional level at discharge from acute care hospitals. These covariates were chosen based on the existing literature suggesting their potential associations with rehabilitation services use and outcomes (2,10,26-29).
Children's demographic variables included age at injury, sex, country/region, and race. Injury-related variables included likelihood of abusive head trauma (AHT), mechanism and cause of injury. Injury or illness severity measures included Glasgow Coma Scale (GCS) score at the time of intracranial pressure (ICP) monitor placement, GCS score at hospital discharge, Abbreviated Injury Scale (AIS) scores, Injury Severity Score (ISS), and Pediatric Risk of Mortality [PRISM] III score. Data were also collected on pupil response at the time of ICP monitor placement, pre-hospital and resuscitation events, such as cardiac arrest, hypoxia, and hypotension, as well as acute computed tomography (CT) findings. Other collected clinical information included length of hospital stay, systemic complications, and neurological complications during acute hospitalization. Data on demographic, injury-related and clinical characteristics were collected by abstracting children's medical records. Data on social and family-related factors, including the highest educational level achieved by primary caregiver, number of employed family members and overall family functioning, were collected from self-administered questionnaires completed by parents/guardians at hospital discharge. Parents/ guardians also reported children's preinjury conditions. Preinjury conditions were classified as neurodevelopmental disability, attention, or learning problems and other conditions. Children's functional level at discharge was evaluated by parents/guardians using the functional status scale (FSS) with detailed instructions.

Statistical analysis
The GOS-E Peds and PedsQL scores were summarized using frequencies and percentages or means and standard deviations and compared between inpatient rehabilitation and noninpatient rehabilitation groups using Chi-square or twosample T tests. Unadjusted ordinal logistic and linear regression models were used to examine the associations between rehabilitation process and GOS-E Peds and PedsQL scores, respectively. The proportional odds assumption was assessed and found to hold. Propensity score (PS) methods were applied to adjust for confounding. We estimated the PS using the generalized booted regression approach (30,31). All potential confounding variables described in the previous section were included in the PS model. From this approach, a PS indicating the probability of receiving inpatient rehabilitation after acute hospitalization given confounding variables was calculated for each child. We applied PS weighted regression models to assess the associations of rehabilitation process with GOS-E Peds and PedsQL scores (30,31). If, after weighting by the inverse probability of treatment, the absolute standardized difference for a potential confounding variable exceeded 0.25 (32,33), this confounding variable was regarded not to be sufficiently balanced by the inverse probability of treatment weighting (IPTW) and further included in the regression models for the outcome analyses.
To address the missing values for some confounding variables, we conducted multiple imputations using the fully conditional specification method based on the assumption that data were missing at random (M = 15 imputations). All covariates considered as potential confounding characteristics, the rehabilitation process variable, and the outcome of interest were included in the imputation models. In each imputed dataset, PS weighted regression was performed to estimate the associations between rehabilitation process and outcomes. Estimates from 15 imputed datasets were combined using Rubin's rules (34).
The extent that GCS at discharge modified the relationships between rehabilitation process and outcomes were first tested by including an interaction term of the rehabilitation variable by GCS in adjusted outcome models. A p value <0.05 for the interaction term was suggestive of the effect modification for GCS. Then, we dichotomized GCS at the mean (13.1 ± 2.5) and presented the estimated effects of inpatient rehabilitation versus non-inpatient rehabilitation on that outcome in two GCS strata (GCS 3-12 and GCS 13-15). Such GCS dichotomization has also been used to evaluate short-term neurological status in prior research (35).
To account for the effects of weighting on standard errors, we performed all inverse probability weighted regression analyses using the survey sampling technique. All statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC). Results were considered significant at p value < 0.05.

Results
Of children with rehabilitation data, 180 received inpatient rehabilitation and 74 received only non-inpatient rehabilitation after acute care for severe TBI (Figure 1). Baseline sociodemographic and clinical characteristics of these 254 children included in the analysis of GOS-E Peds outcome have been presented in our previous work (7). Baseline characteristics of children included in the analysis of parent/guardian-reported and child self-reported PedsQL scores are shown in Supplemental Tables 1 and 2, respectively. Table 1 summarizes the GOS-E Peds, parent/guardian-reported, and child selfreported PedsQL scores for children in the two rehabilitation groups. At the 12-month follow-up, 46.7% of children in the inpatient rehabilitation group and 35.1% of children in the non-inpatient rehabilitation group showed severe disability or were at vegetative status (GOS-E Peds > 4). Among 181 children with parent/guardian-reported PedsQL scores, the mean scores for physical function were 65.9 (SD 28. Without adjustment for confounding, the inpatient rehabilitation group had 29% (95% confidence interval [CI], 0.80-2.09) increased odds for higher GOS-E Peds scores (less favorable global function) compared with the non-inpatient rehabilitation group, though results were not statistically significant ( Table 2). After adjustment for confounding through IPTW, the inpatient rehabilitation group had 25% (95% CI, 0.37-1.53) reduced odds for higher GOS-E Peds scores compared to the non-inpatient rehabilitation group, though this result was not statistically significant.
The effect modification for GCS score at hospital discharge on the association between rehabilitation process and GOS-E Peds was statistically significant in both the unadjusted (P = 0.006) and the confounding-adjusted analyses (P = 0.049) ( Table 2). In the analysis stratified by GCS < vs ≥ 13 at discharge (n = 84 and n = 148, respectively, in the unimputed data) and adjusted for confounding, rehabilitation process was not associated with GOS-E Peds among those with a GCS ≥ 13 (OR [95% CI]: 1.12 [0.51-2.47]). However, in children with a GCS < 13 at hospital discharge, inpatient rehabilitation was associated with reduced odds for higher GOS-E Peds scores (less favorable global function) (OR [95% CI]: 0.12 [0.02-0.95]), though a wide confidence interval was observed for this result. Table 3 summarizes the estimated associations of rehabilitation process after acute care with PedsQL raw scores reported by parents/guardians and by children themselves at 12 months. In the unadjusted analysis, child self-reported physical PedsQL score was 6.92 (95% CI, −15.40-1.55) points lower in the inpatient rehabilitation group when compared to the non-inpatient rehabilitation group, which exceeded the MCID but did not reach statistical significance. After adjustment for confounding through IPTW, the difference in child self-reported physical PedsQL score between rehabilitation groups diminished to 5.81 (95% CI, −16.50-4.88), which was less than the MCID. For other child self-reported PedsQL and parent/guardian-reported PedsQL measures, none of the differences in summary or domain-specific scores between two rehabilitation groups approached the MCIDs or statistical significance in the unadjusted or confoundingadjusted analyses. No significant effect modification was found for GCS at hospital discharge on the association of rehabilitation process with parent/guardian-reported or child self-reported summary or domain-specific PedsQL scores.

Discussion
This is the first study to our knowledge to compare the effects of different patterns of care continuum with regard to rehabilitation process on long-term functional and quality-of-life outcomes in children with severe TBI. In the confoundingadjusted analysis in children with a GCS < 13 at hospital discharge, those who received inpatient rehabilitation showed more favorable global function at 12 months compared to children who received only non-inpatient rehabilitation. No differences were found in global function between rehabilitation groups in children with a GCS ≥ 13 at hospital discharge. None of the confounding-adjusted differences in parent/guardian-reported or child self-reported PedsQL scores between two rehabilitation groups reached the MCIDs or statistical significance. In the unstratified analysis of GOS-E Peds, the OR for unfavorable global function associated with inpatient rehabilitation compared with only non-inpatient rehabilitation reversed from 1.29 to 0.75 after adjustment for confounding, though these results were not statistically significant. Such a reversal of the effects of inpatient rehabilitation may result from strong confounding factors such as injury severity and functional level at hospital discharge, which could influence clinical decisions for rehabilitation disposition and independently associate with long-term outcomes. After controlling for confounding, we observed a trend toward beneficial effects of inpatient rehabilitation in the entire cohort.
In agreement with previous literature that suggested a benefit of early onset inpatient rehabilitation after acute care in adult and pediatric patients with severe TBI (12,14,36), we observed a trend indicating that receipt of inpatient rehabilitation after acute hospitalization was associated with more favorable global function in children who had a GCS < 13 at hospital discharge, though such association was not found in those with a higher GCS. Our findings suggested that children with more severely impaired consciousness when medically stable may benefit more from receiving inpatient rehabilitation than receiving noninpatient rehabilitation after acute care. This is consistent with previous research in adults which showed that patients with prolonged impaired consciousness after TBI can regain consciousness and obtain at least partial functional independence during inpatient rehabilitation (16). Patients with prolonged impaired consciousness often experience medical complications that may impede recovery (37). It has been suggested that these patients can achieve better recovery when receiving care in a specialized setting managed by professionals who are aware of the risks related with impaired consciousness and able to initiate timely medical interventions (38). We observed no differences in global function between inpatient and non-inpatient rehabilitation groups among children who had less severely impaired consciousness or normal consciousness at hospital discharge. One possible explanation for this lack of association is that among children with less severe impairment, beneficial effects of early inpatient rehabilitation failed to be maintained by 12 months due to gradual functional improvements in the noninpatient rehabilitation group. Moreover, as the GOS-E Peds only provides one composite score to determine global function, it may not adequately represent the range of impairment (39). Future studies are underway to incorporate multiple outcome measures including cognitive tests to detect subtle differences in Table 1. Pediatric Glasgow Outcome Scale Extended (GOS-E Peds) and Pediatric Quality of Life Inventory (PedsQL) scores at 12-month follow-up of children with severe TBI in the inpatient rehabilitation and non-inpatient rehabilitation groups. outcome between children receiving inpatient rehabilitation and children who received only non-inpatient rehabilitation.
The non-significant effect modification for GCS at hospital discharge on associations between rehabilitation process and HRQOL was likely due to limited power. There were fewer children with a GCS < 13 at discharge included in the analysis of parent/guardian-reported and child self-reported HRQOL (n = 55 and n = 41, respectively, in the unimputed data) compared to the number included in the global function analysis, especially among those who only received non-inpatient rehabilitation. Consistent with the findings on global function among children with a GCS ≥ 13 at hospital discharge, we did not observe significant associations between rehabilitation process and HRQOL.
Several limitations of this study should be noted. First, rehabilitation disposition may have been misclassified for some children as it was only based on their discharge destination from acute care; no data were collected from an inpatient rehabilitation admission. This limitation may be particularly relevant to the small number (n = 4) of children discharged to SNFs, as the provision of rehabilitation services in such settings may vary. Children who were referred to but eventually failed to receive inpatient rehabilitation may have been misclassified in the inpatient rehabilitation group. Children who were directly discharged home after acute care but later admitted to IRFs may have been misclassified in the non-inpatient rehabilitation group. Such misclassification may have resulted in more similar rehabilitation groups and biased the findings toward the null. Second, the relatively small sample size in the analysis of children with a GCS < 13 at discharge (n = 84 in the unimputed data) likely led to an underpowered analysis and the wide confidence intervals of results regarding the association between rehabilitation process and GOS-E Peds, which gives us caution against definitive conclusions. These results need to be confirmed in future studies with larger sample sizes. Third, we cannot rule out the possibility of residual confounding effects on results. Nevertheless, our study is the first to evaluate the effectiveness of rehabilitation process in a large cohort of pediatric patients with severe TBI and focus on multiple outcomes including functional and both parent/guardian-perceived and child selfperceived quality-of-life outcomes.

Conclusion
Our study suggests that, among children with severe TBI, those with more severely impaired consciousness after acute medical stabilization benefited from receiving inpatient rehabilitation, particularly in terms of global function at 12 months after injury. Parent/guardian-reported or child self-reported HRQOL at 12 months were not different between children who received inpatient rehabilitation and children who received only noninpatient rehabilitation after acute hospitalization. Our findings warrant further research investigating the effects of different components of the rehabilitation process such as the types, amounts, intensity, and duration of rehabilitation services, the rehabilitation settings, and the timing of the initiation of rehabilitation on long-term outcomes of children sustaining severe TBI. Additionally, studies are needed to confirm the effectiveness of inpatient rehabilitation in children with prolonged impaired consciousness after severe TBI.

Disclosure statement
No potential conflict of interest was reported by the author(s). Table 3. Unadjusted and adjusted associations of inpatient rehabilitation versus non-inpatient rehabilitation with parent/guardian-reported and child self-reported Pediatric Quality of Life Inventory (PedsQL) scores in children with severe TBI (M = 15 imputations).