Global Outcomes Across 20 Years After Inpatient Rehabilitation for Traumatic Brain Injury

Abstract

Objective:

To explore and visually display differences in the distribution of the Glasgow Outcome Scale – Extended (GOS-E) over time after traumatic brain injury (TBI), focusing on variations in outcome distributions based on GOS-E at Year 1 postinjury and age at injury.

Setting:

Community.

Participants:

14 010 individuals who received inpatient rehabilitation in the TBI Model Systems.

Design:

Cross-sectional analysis of a prospectively collected longitudinal database.

Main Measures:

GOS-E scores at 1, 2, 5, 10, 15, and 20 years postinjury, and age at injury.

Results:

The proportions of cases in each GOS-E category are displayed using 100% stacked bar graphs for each follow-up period. These graphs reveal that GOS-E at Year 1 and age at injury clearly influence outcomes over time. Trends include decreasing good recovery and increasing severe disability as Year 1 GOS-E worsens, along with rising mortality rates as age at injury increases.

Conclusion:

The study introduces a novel approach for visually representing patterns of change in GOS-E outcomes, emphasizing differences across strata defined by GOS-E at Year 1 and age at injury. The figures provide a valuable tool for communicating potential outcomes, particularly when GOS-E at Year 1 and age are known. Evaluating the visual interpretability of these graphs among persons with brain injury, family members, healthcare providers, and other stakeholders will help determine their broader usability and value.

THERE IS GROWING recognition that traumatic brain injury (TBI) can be a chronic condition with outcomes that are more dynamic than stable even many years postinjury.1,2 Long-term recovery trajectories, particularly following complicated mild to severe TBI, have been extensively studied using data from the TBI Model Systems National Data Base (TBIMS NDB).3 This cohort includes individuals who received inpatient rehabilitation after TBI. Over the first 10 years postinjury, TBIMS NDB studies reveal significant improvement in functional outcomes during the initial 5 years,4,5 followed by declines in functional independence1,6,7 and global outcomes,1 beginning around the 10-year mark. Declines are more pronounced when mortality is included as an outcome,1 as in measures such as the Glasgow Outcome Scale – Extended (GOS-E)8 and Disability Rating Scale (DRS).9 Key predictors of average change include age at injury and severity of the initial injury,1,10 along with the presence of co-morbidities, which exacerbate declines over time.7 When shorter time intervals are examined (eg, 5- to 10-year change), outcomes at earlier time points are highly predictive of subsequent functioning.10 Among older adults, declines in global outcome measured by the GOS-E are particularly steep for those aged 60 years and older, with an even greater decline among individuals 70 years and older.11

International studies, such as those conducted in Australia by Ponsford and colleagues, have also examined long-term recovery. Their findings suggest that deficits apparent at 2 years postinjury often persisted to 10 years.12 Unlike the US studies, they reported that older age at injury did not significantly affect 10-year outcomes, except in employment status.12 Another study found that shorter duration of posttraumatic amnesia and younger age at injury were associated with higher employment rates.13 Similarly, research from Norway identified younger age and less severe injury as key prognostic factors for global outcome,14 consistent with findings from the US and Australian cohorts.

Fewer studies have addressed outcomes beyond 10-year postinjury. TBIMS NDB studies indicate that global outcome, as measured by the GOS-E, plateaus around the 10-year mark before declining, with this trend accelerating after 15 years.4,15 Mortality further accentuated these declines.4 Significant predictors of decline include older age at injury and greater initial TBI severity,16,17 as was male sex, Black race, and lower education levels.16 As time postinjury increases, factors such as aging and cognitive reserve appear to play a larger role in functional decline than initial injury severity.18 European studies also highlight persistent effects of TBI over decades, with some evidence of improvement many years postinjury.19,20

Individual growth curve models (IGCMs) have been widely employed to analyze longitudinal TBI outcomes, producing mathematical models that predict average outcome trajectories based on demographic and injury characteristics.1,4,7,15,21 IGCMs effectively incorporate multiple individual factors to predict outcomes over time and produce easily interpretable trajectory plots. However, while IGCMs provide valuable insights into the single predicted trajectory, they do not estimate the probability of the predicted trajectory occurring compared to the wide variety of other possible outcome trajectories for a given individual. This limitation highlights the need for complementary approaches to capture the uncertainty and variability in individual outcomes over time.

Using the TBIMS NDB, Corrigan and colleagues calculated the percentage of individuals improving, remaining stable, or declining across specific time periods.2 Their analysis of GOS-E scores across 7 intervals (Years 1-2, 2-5, and every subsequent 5-year period out to 30 years postinjury), revealed that improvement was more common than decline in all intervals except Years 1-2, with only 41%-49% of individuals maintaining stable outcomes. However, this analysis did not examine change over longer periods or stratify results by individual factors such as age or early global outcomes.

The present study built on these efforts by producing simple, interpretable graphics that illustrate the percentage distributions of GOS-E categories (including death) over the first 20 years postinjury using data from the TBIMS NDB, which is publicly available. This descriptive study specifically examines the influence of early global outcome, age at injury, and their combination on GOS-E distributions, informed by a review of the literature and preliminary analyses.

METHODS

All research conducted within the TBIMS is approved by each center’s local Institutional Review Board. ChatGPT was used for grammatical editing of the manuscript. The study included a subset of the TBIMS NDB participants who were alive with a valid GOS-E score at their Year 1 follow-up, and were due for their Year 2 follow-up before June 30, 2024. Participants in the TBIMS NDB had sustained a TBI caused by an external mechanical force, meeting at least one of the following criteria: loss of consciousness exceeding 30 minutes, a Glasgow Comma Scale score below 13 in the Emergency Department, posttraumatic amnesia lasting more than 24 hours, or intracranial neuroimaging abnormalities. Eligible participants were at least 16 years old at injury, admitted to a TBIMS acute care hospital within 72 hours, receive inpatient rehabilitation at a TBIMS center, and provided informed consent (or had consent provided by a family member or legal representative, if incapacitated). Extensive data on patient demographics and hospitalization details are collected at enrollment, with participants followed longitudinally at 1, 2, and 5 years postinjury, and every 5 years thereafter, collecting information about their current status, health, and outcomes. Follow-up interviews, typically conducted by telephone with the participant or a proxy, were scheduled within specific time windows: ±2 months for the Year 1 follow-up, ±3 months for the Year 2 follow-up, and ±6 months for the 5-year follow-up and subsequent follow-ups. Data from the 1, 2, 5, 10, 15, and 20-year follow-up periods were included on all eligible cases provided the follow-up window closed before the analytic cutoff date of June 30, 2024. TBIMS data is publicly available at tbindsc.org/Researchers.aspx.

Primary outcome: GOS-E

The primary outcome variable for this study was the GOS-E, a widely used measure of global functional outcome.8 The GOS-E evaluates the full spectrum of impairment, functional activity, and societal participation and is the most frequently used variable used in prior studies of change in TBI outcome over time. The GOS-E structured interview consists of 19 questions (16 with yes/no responses and 3 with a choice of 2 or 3 simple responses) that classifies outcome into 1 of 8 categories: upper good recovery (UGR, score = 8), lower good recovery (LGR, score = 7), upper moderate disability (UMD, score = 6), lower moderate disability (LMD, score = 5), upper severe disability (USD, score = 4), lower severe disability (LSD, score = 3), vegetative state (VS, score = 2), and death (D, score = 1). The GOS-E is both reliable (kappa = 0.85)8 and valid, with strong associations to established measures like the Disability Rating Scale, the Beck Depression Inventory, the General Health Questionnaire, the SF-36, and the Neurobehavioral Function Inventory.22 The follow-up year was captured for each participant, and age at injury was calculated as the difference between the participant’s birth date and injury date. Data collected for analysis include the GOS-E (which was added to the NDB in April 1998, but is not collected at rehabilitation discharge) and demographic and injury characteristics.

Covariate selection and preliminary analysis

To identify key stratifying variables for GOS-E outcomes over 2 to 20 years postinjury, a preliminary analysis was conducted. This analysis assessed GOS-E as a continuous outcome (range 1-8) using repeated measures mixed-effects models to evaluate the relative importance of select covariates. Covariates included demographic characteristics (age group, sex, race, ethnicity, and Year 1 values for education, marital status, companionship, household income, and substance problem) and injury characteristics (GOS-E at Year 1 and cause of injury), along with their interactions with time postinjury. Results identified age group at injury and Year 1 GOS-E as the strongest predictors of GOS-E, with these variables showing the largest partial eta squared ( ${\eta }_p^2$) values for both main effects and interactions with time. While other sociodemographic and injury-related factors showed statistically significant associations with GOS-E outcomes, their potential multicollinearity would have added complexity to the model that couldn’t easily be visualized and their effect sizes were modest compared to those of age groups and Year 1 GOS-E (see Supplemental Digital Content Part 1 for detailed analyses, available at: http://links.lww.com/JHTR/A995).

Rationale for stratifying variables

Previous research has consistently demonstrated that injury severity and age are among the strongest predictors of long-term TBI outcomes. Injury severity is strongly related to early global outcomes. Early outcome assessment, particularly after the initial recovery period and rehabilitation (eg, GOSE at Year 1), logically remain the strongest predictors of the same outcome assessment at later time points. Age is also a consistent predictor of outcome, as it influences recovery potential and mortality risk. The findings from the preliminary analysis align with the broader existing literature, supporting the selection of age group at injury and GOS-E at Year 1 as the primary stratifying variables.

Analysis

Follow-up data were transformed from a long format (1 record per follow-up) to a wide format (1 record per participant). Variables collected during the initial hospitalization were linked and appended to the wide-format dataset. For participants who died during follow-up (after Year 1), the GOS-E for the first follow-up post-death was assigned the valid score of “1” and labeled “dead.” Subsequent follow-ups occurring before June 30, 2024 were also coded “1” and labeled “dead.” Follow-ups scheduled for future dates (after June 30, 2024) were coded “system missing” and excluded from the analysis. Cases where participants were lost to follow-up, refused participation, were incarcerated, or withdrew from the study were coded “99,” labeled “unknown,” and excluded from the analysis.

While participants were followed longitudinally, we conducted a cross-sectional analysis to determine the distribution of GOS-E outcomes at the 1-, 2-, 5-, 10-, 15-, and 20-year follow-up periods postinjury. This approach was selected to highlight differences in the distribution of outcomes across 20 years and to maximize use of available data at each time point without requiring the assumptions inherent in longitudinal models. By focusing on descriptive cross-sectional methods, we aimed to produce simple, interpretable visualizations of outcome variability that are accessible by clinicians and TBI stakeholders.

The main analyses tabulated the number of cases in each GOS-E category at each follow-up period. Percentages for each category were calculated and displayed as a 100% stacked bar graph to illustrate the distribution of GOS-E categories over time. To explore the influence of key predictors, we stratified the stacked bar charts by GOS-E at Year 1 and by age group at injury. Stratifying by Year 1 GOS-E scores provided a way to indirectly examine change over time by fixing initial outcome and observing subsequent distributions. Stratifying by age at injury, using age groups 16-24, 25-39, 40–59, and 60+ years, highlighted potential age-related trends in recovery and mortality across the follow-up periods. Finally, combined stratification by Year 1 GOS-E scores and age group at injury provided additional insights into how these factors jointly influence long-term outcomes.

RESULTS

Analyzing data available after the last data submission of 2024, a total of 14 827 participants were alive with a valid GOS-E score at their Year 1 follow-up and had a window closing date before June 30, 2024, for their Year 2 follow-up. Of these, 12 433 participants had valid GOS-E scores at Year 2, 10 255 at Year 5, 7095 at Year 10, 4225 at Year 15, and 1540 at Year 20. The distribution of GOS-E scores at 1, 2, 5, 10, 15, and 20 years postinjury is shown in Figure 1 using stacked bar charts. The general pattern is early stability followed by decline in the GOS-E severity categories (scores 2-8) due to an expected increase in mortality over time. Rates of good recovery (UGR plus LGR) increases from 35% at Year 1 to 38% at Years 2 and 5 before declining to 28% at Year 20. The rate of moderate disability (UMD plus LMD) is 32% at Years 1 and 2 before slowly declining to 27% by Year 20. The rate of severe disability (USD plus LSD) declines more rapidly from 33% in Year 1 to 10% in Year 20; VS declines from 0.5% to 0.1%; while death increases from 3% at Year 2 to 36% at Year 20. Detailed sample sizes for each bar, frequencies, and percentages for all GOS-E categories across follow-up periods, along with larger graphical representations are presented in Supplemental Digital Content Part 2, available at: http://links.lww.com/JHTR/A995.

Figure 1.

GOS-E distribution by time postinjury among individuals with TBI, alive with valid GOS-E at 1 year postinjury. GOS-E: Glasgow Outcome Scale – Extended; UGR: Upper Good Recovery; LGR: Lower Good Recovery; UMD: Upper Moderate Disability; LMD: Lower Moderate Disability; USD: Upper Severe Disability; LSD: Lower Severe Disability; VS: Vegetative State; D: Death.

The distribution of GOS-E across time postinjury and stratified by Year 1 GOS-E scores is shown in Figure 2, demonstrating how Year 1 GOS-E influences outcome from Years 2 to 20. Each graph begins with participants grouped by their Year 1 GOS-E category, with subsequent bars showing the distribution of GOS-E scores for the same GOS-E Year 1 group over time. These 7 graphs illustrate those individuals with better GOS-E scores at Year 1 are more likely to maintain or improve their outcomes over time compared to those with lower scores at Year 1. Across all Year 1 GOS-E strata, increasing mortality is observed, with the rate of increase being greatest for individuals who had severe disability or were in a vegetative state at Year 1. Among participants who remain alive, the general pattern is one of stability or improvement rather than decline, regardless of Year 1 GOS-E category. Some participants with moderate or severe disability at Year 1 show some improvement to levels of good recovery over the 20 years, but the likelihood of such improvement diminishes with greater Year 1 disability. For example, among those with Upper Moderate Disability at Year 1, about a third are in a Good Recovery category over the 20 years. Notably, the vegetative state category at Year 1 includes far fewer participants (n = 77) than other categories, leading to less reliable percentage distribution for this subgroup. Detailed sample sizes for each bar, frequencies, percentages, and larger graphs are provided in Supplemental Digital Content Part 3, available at: http://links.lww.com/JHTR/A995.

Figure 2.

GOS-E distribution by time postinjury, stratified by Year 1 GOS-E categories. GOS-E: Glasgow Outcome Scale – Extended; UGR: Upper Good Recovery; LGR: Lower Good Recovery; UMD: Upper Moderate Disability; LMD: Lower Moderate Disability; USD: Upper Severe Disability; LSD: Lower Severe Disability; VS: Vegetative State; D: Death. *Fewer than 50 participants contributed data at this time point. Interpret with caution.

Figure 3 presents 4 graphs illustrating how age at injury differentiates GOS-E outcomes from Years 2 to 20. The graphs show a clear trend: younger age at injury is associated with better GOS-E outcomes over time. The most notable difference across the age groups is observed in the GOS-E category of death. While the percentage of expired cases increases over the 20 years in all age groups, as expected, the rate of death increases significantly with age at injury. By Year 20, only 12% of participants in the youngest age group (16-24) had died, compared to 91% in the oldest age group (60+). Among participants who remain alive, the distribution of GOS-E categories tended to remain relatively stable over time, with slight improvement observed between Years 1 and 5. Detailed sample sizes for each bar, frequencies, percentages, and larger graphs are provided in Supplemental Digital Content Part 4, available at: http://links.lww.com/JHTR/A995.

Figure 3.

GOS-E distribution by time postinjury, stratified by age group at injury. GOS-E: Glasgow Outcome Scale – Extended; UGR: Upper Good Recovery; LGR: Lower Good Recovery; UMD: Upper Moderate Disability; LMD: Lower Moderate Disability; USD: Upper Severe Disability; LSD: Lower Severe Disability; VS: Vegetative State; D: Death.

Figure 4 presents 24 graphs illustrating how the combination of GOS-E at Year 1 and age group at injury influence GOS-E from Years 2 to 20. The figure is organized into 6 rows, each representing a GOS-E category at Year 1 (from UGR to LSD), and 4 columns, each corresponding to an age group at injury (16-24, 25-39, 40-59, and 60+ years old). The VS category at Year 1 is excluded due to an insufficient number of cases to reliably divide into 4 age groups. The 24 graphs highlight the combined effect of better outcomes at Year 1 and younger age at injury predicting more favorable outcomes over time. The upper left graph (GOS-E of UGR among the 16-24 age group) depicts the very best outcomes, with at least 75% in good recovery and a maximum of 5% mortality over time. In contrast, the lower right graph (GOS-E of LSD among the 60+ age group) shows the very worst outcomes, with good recovery not exceeding 5% and mortality rising to 95% by Year 20. Two predominate patterns emerge from the data. Within each age group, good recovery decreases and severe disability increases as Year 1 GOS-E worsens from UGR to LSD. Similarly, within each Year 1 GOS-E category, mortality increases as age at injury increases. Detailed sample sizes for each bar, frequencies, percentages, and larger graphs are provided in Supplemental Digital Content Part 5, available at: http://links.lww.com/JHTR/A995; outcomes for the Year 1 VS category and age groups at injury are also provided in Supplemental Digital Content Part 5, available at: http://links.lww.com/JHTR/A995.

Figure 4.

GOS-E distribution by time postinjury, stratified by year 1 GOS-E and age group at injury. GOS-E: Glasgow Outcome Scale – Extended; UGR: Upper Good Recovery; LGR: Lower Good Recovery; UMD: Upper Moderate Disability; LMD: Lower Moderate Disability; USD: Upper Severe Disability; LSD: Lower Severe Disability; VS: Vegetative State; D: Death. *Fewer than 50 participants contributed data at this time point. Interpret with caution.

DISCUSSION

This paper presents a novel approach to visually displaying the distribution of global outcomes, as measured by GOS-E, during the first 2 decades after inpatient rehabilitation for TBI. It highlights the influence of Year 1 GOS-E and age at injury on GOS-E outcomes over time. The graphic figures, derived from the TBIMS NDB, are easily interpretable by healthcare providers and TBI stakeholders, including family members. The graphs make it clear that there are multiple outcome paths for people with TBI and they provide the probability of each GOS-E outcome category at each follow-up time period. The visuals clarify the multiple outcome paths for individuals with TBI and provide the probability of each GOS-E category at specific follow-up periods. By stratifying outcomes by Year 1 GOS-E and age group at injury, these graphs offer insights into the likely outcome trajectories while still emphasizing the variety of possible outcomes. They provide a glimpse into the future for TBI patients but underscore the inherent uncertainty.

Individuals with good recovery at Year 1 are more likely to maintain their recovery, until death, which depends heavily on age group at injury and years postinjury. Those with moderate disability at Year 1 tend to remain in this category rather than improve or decline, again influenced by aging. For individuals with severe disability at Year 1, age plays a critical role: those under 60 are more likely to improve, while those 60 or older are more likely to remain severely disabled, with mortality increasing as they age.

This study has several limitations. While the TBIMS NDB is the largest longitudinal TBI database in the world, it only includes individuals who received inpatient rehabilitation and excludes those with mild TBI or less severe injuries. The study focuses solely on GOS-E as the global outcome; future research could explore additional outcomes. The graphics are directly interpretable when considering only 2 influencing factors; analyzing more factors would require different methods and compromise simplicity. This analysis is descriptive rather than attempting to model outcome trajectories based on multiple predictor variables; it is primarily cross-sectional rather than longitudinal. While all cases included in later follow-ups are included in the Year 1 data, the reverse is not true. As such, these results reflect cross-sectional estimates and should not be interpreted as individual trajectories. Although stratification by Year 1 GOS-E scores and age group at injury offers insights into aggregate change over time, the cross-sectional nature of this analysis limits the ability to fully capture individual-level dynamics. Additionally, the number of cases decreases with each follow-up period, and the VS category is particularly small, limiting statistical reliability. The set of covariates used in the preliminary analysis to identify the 2 most important predictors of GOS-E outcomes does not include all the potential variables that might influence outcomes. Finally, the lack of a control group without TBI prevents comparisons with the natural aging process.

Despite these limitations, this graphic approach effectively presents the broad range of outcome trajectories in an accessible and interpretable format. While other methods, such as IGCMs, offer the ability to model individual trajectories and include multiple predictors, they focus on a single trajectory. Similarly, analysis of percentages improving, remaining stable, or declining between follow-up periods is straightforward buts fail to capture changes over longer periods. However, determining the differences in the characteristics of those who improve, remain stable, or decline may be fruitful in future research.

The figures in this paper can serve as a starting point for healthcare professionals to communicate likely outcome possibilities, particularly when GOS-E at Year 1 and age are known. However, these probabilities are derived from the TBIMS sample and may vary when applied to other populations. Evaluating the visual interpretability of these graphs among diverse TBI stakeholders will help determine their broader usability and value.