Stability and Change in Biopsychosocial Factors Associated With Fatigue 6 and 12 Months After Traumatic Brain Injury: An Exploratory Multilevel Study

Abstract

Objective: 

To explore factors associated with stability and change in fatigue from 6 to 12 months following traumatic brain injury (TBI).

Setting: 

Combined in- and outpatient acute care and postacute rehabilitation settings.

Participants: 

A total of 103 patients with confirmed intracranial injury were assessed 6 and/or 12 months following TBI.

Design: 

A prospective observational study with repeated measures at 2 time points, analyzed with a hybrid mixed-effects model.

Main Measures: 

Primary outcomes were the fatigue factor derived from items from several fatigue patient-reported outcome measures (PROMs; Fatigue Severity Scale, Chalder Fatigue Scale, Giessen Subjective Complaints List–fatigue subscale, and Rivermead Post-Concussion Symptoms Questionnaire–fatigue item) Secondary outcomes were PROMs relating to pain, somatic and psychological distress, insomnia, sleepiness, personality traits, optimism, resilience, behavioral activation and inhibition, and loneliness, as well as neuropsychological measures. Demographic variables and injury severity characteristics were included as covariates.

Results: 

In multilevel regression, female sex, years of education, and 3 factors related to injury severity, somatic vulnerability, and psychosocial robustness were all significantly associated with variation in fatigue between subjects, and explained 61% of the variance in fatigue that was due to stable between-subject differences. Fatigue levels declined significantly over time. Changes in pain severity, somatic symptom burden, psychological distress, and behavioral inhibition were positively associated with changes in fatigue, explaining 22% of the variance in fatigue within subjects.

Conclusions: 

The study demonstrated that several previously implicated factors show robust effects in distinguishing individuals with TBI on levels of fatigue, but only a few show additional within-subject associations across time. Pain severity, somatic symptom burden, psychological distress, and behavioral inhibition correlated with fatigue across time, implicating these factors as crucial targets for rehabilitation of patients with TBI who suffer from persistent fatigue.

PERSISTENT FATIGUE is common after traumatic brain injury (TBI),^1,2 and associated with functional impairment and reduced quality of life even when controlling for injury severity.³ Estimates of prevalence vary between 7% and 80%, dependent on injury severity, the patient-reported outcome measure (PROM) employed, cut-off values applied, and interval investigated.^2,4 Fatigue remains one of the most troublesome symptoms in chronic TBI,⁵ and there is need for interventions that ameliorate fatigue in earlier phases.⁶ Although some TBI-specific mechanisms have been suggested, such as cognitive impairment^7–10 and endocrine disturbances,¹⁰ fatigue is common in many chronic illnesses, and relatively normally distributed in the general population.^11,12 There is also overlap in several predisposing and exacerbating factors for fatigue across disorders,^13,14 such as pre- or comorbid pain, sleep disorders, and psychological distress. The biopsychosocial nature of fatigue necessitates a multifactorial approach in studying changes in fatigue and related factors across time.

In a previous cross-sectional analysis of the first wave (6 months post-injury) of the current study,¹⁵ we found 3 factors underlying biopsychosocial correlates of fatigue. Pain, somatic symptoms, daytime sleepiness, and insomnia were related to fatigue through a factor termed “somatic vulnerability.” Psychological distress, personality traits (neuroticism, extraversion, conscientiousness, and optimism), behavioral inhibition, and loneliness were associated with fatigue through a factor termed “psychosocial robustness.” Third, we demonstrated injury severity and neuropsychological variables to be associated with fatigue, although the effects were small. Together, these factors explained 44.2% of variance in fatigue 6 months after injury. Cross-sectional designs do, however, not inform us regarding directional influences and potential confounders in the relationships between fatigue and associated factors.

There is currently a scarcity of longitudinal TBI studies examining temporal dynamics between fatigue and associated biopsychosocial variables. In a study of 88 patients with complicated mild to severe TBI, Schönberger et al¹⁶ found that early fatigue predicted more depression 6 months post-injury, but that early depression and daytime sleepiness did not predict later fatigue. Beaulieu-Bonneau and Ouellet¹⁷ found injury severity-dependent trajectories of fatigue 4, 8, and 12 months following TBI, and that depression and insomnia were associated with fatigue at all time points, while pain was associated with fatigue only 4 and 8 months post-injury. More recently, Rakers et al¹⁸ examined symptom clusters implicated in trajectories of fatigue following mild TBI with latent class growth models throughout the first 6 months following injury. The 2 clusters with persistent levels of fatigue were characterized by an overrepresentation of females, presence of pain, pre- and comorbid sleep complaints, and passive coping compared with those experiencing decreases in fatigue.

Studies focusing on identifying changes in fatigue over time seldom investigate the relationship with changes in other variables, as the aim is typically to predict which individuals develop persistent fatigue, and do not take into account that there may be considerable stability in both fatigue outcome and its predictors, as has been demonstrated in patients with persistent musculoskeletal pain.¹⁹ However, to ascertain whether associated variables are related to changes in fatigue across time, within-subject associations must be examined. For example, shared genetic susceptibility for both fatigue and several of the implicated factors such as pain and psychological distress has been demonstrated in musculoskeletal research,²⁰ which may complicate our understanding of their relationships. Exposure to risk factors prior to the brain injury, such as premorbid mental or physical illness, might predispose someone to both fatigue and depression following TBI, and explain their co-occurrence. Confounders such as these cannot be controlled for when examining only between-subject variability. Studying within-subject change and associations across variables over time provides a reliable method for identifying targets crucial for intervention, by taking into account stable trait-like propensities for fatigue and related factors.¹⁹

AIMS

The aim of this study was to investigate characteristics of stable between-subject levels of fatigue 6 and 12 months following TBI, and to evaluate synchronous changes in fatigue and associated factors within subjects.

METHODS

Recruitment

Participants were identified prospectively between January 2018 and March 2020 at the Neurosurgical Department Oslo University Hospital (OUH), Ullevål. They underwent assessments approximately 6 and 12 months post-injury. Inclusion criteria were being between 18 and 65 years, admitted with TBI (International Classification of Diseases, Tenth Revision [ICD-10] diagnoses S06.1-S06.9) with intracranial injuries (verified by CT or MRI). Exclusion criteria were severe pre- or comorbid psychiatric illness (schizophrenia or bipolar spectrum disorders, or ongoing suicidal ideation), ongoing treatment-dependent substance or alcohol abuse, and severe physical and/or cognitive functional impairment hindering the completion of the study protocol. Patients were identified following admission to the acute hospital and recruited through routine follow-up consultations or by invitations through mail. Injury severity indices were retrieved from the Oslo TBI Registry–Neurosurgery, a quality database at OUH.²¹

Primary outcome measures

Our study employed several fatigue measures, and in an earlier study we found a single, reliable factor underlying items from these PROMs in our sample.¹⁵ Fatigue was therefore measured with a factor analytic regression score, with factor loadings constrained across time (ie, assuming no time-related measurement invariance), using items from several fatigue PROMs. The Fatigue Severity Scale²² has been extensively used in the measurement of fatigue following TBI, and consists of 9 items pertaining to perceived impact of fatigue on various functional domains, of which only items 3 to 9 were included in the factor due to nonsalient loadings from items 1 and 2. Chalder Fatigue Questionnaire has been used in studies of chronic fatigue syndrome/myalgic encephalomyelitis and other neurological illness,²³ with 11 questions of various fatigue symptoms related to habitual functioning. Items relating to daytime sleepiness and common cognitive symptoms were not included in the fatigue factor due to overlap with independent variables in the study. The Giessen Subjective Complaints List measures somatic symptom burden, with one subscale pertaining specifically to fatigue.²⁴ The subscale consists of 6 items, of which 3 were not included in the fatigue factor due to overlap with independent variables. Finally, one fatigue item from Rivermead Post-Concussion Symptoms Questionnaire²⁵ was included.

Secondary outcome measures

The study included several PROMs of factors potentially associated with fatigue, such as pain severity and dispersion, somatic symptom burden, psychological distress, daytime sleepiness, and insomnia severity, 5-factor personality traits (neuroticism, extraversion, conscientiousness, agreeableness, and openness), trait optimism, behavioral inhibition, loneliness, and facets of resilience. Furthermore, several injury severity indices from the acute phase were included, e.g., Abbreviated Injury Scale–Head (AIS-head).²⁶ A variable was calculated for those patients discharged directly to rehabilitation from the neurosurgical department (direct pathway to rehabilitation), which was associated with fatigue in an earlier cross-sectional study on data from the first measurement.¹⁵ Finally, performance-based assessment with neuropsychological subtests from the Delis-Kaplan Executive Function System²⁷ (Trail Making Test & Color Word Interference Test), Wechsler's Adult Intelligence Scale IV²⁸ (Digit Span), Wechsler's Abbreviated Scale of Intelligence²⁹ (Matrix Reasoning & Similarities), and Conners Continuous Performance Task III (CPT-III) were included. Further details on specific instruments and measures used can be found in Supplemental Digital Content Table S1.1 (available at: https://links.lww.com/JHTR/A636).

Analyses

All analyses were conducted in Stata, Version 16.³⁰ Multilevel modeling is one way of investigating within-subject associations, as the within-subject stability (ie, between-subject variance) in fatigue and its correlates are segregated from within-subject changes, and can be investigated separately. To assess the data for clustering effects, all time-varying variables (ie, measured at both occasions) were assessed for intraclass correlations within individuals.

Using principles from the hybrid fixed-random effects model proposed by Allison,³¹ all time-varying secondary outcome variables were segregated into person-mean variables, and within-subject deviations from the person-mean at each time point. Mean scores averaged across both time points within individuals were thus generated for all time-varying variables to create between-subject components for all independent variables, and change scores were generated as each observation's deviation from the individual's mean to create within-subject components for all independent variables (see Figure 1 and Supplemental Digital Content Section S3, available at: https://links.lww.com/JHTR/A636). Person-mean centering is a commonly employed technique in multilevel modeling for segregation of between- and within-subjects effects.³²

Figure 1.: A visual illustration of the variance compartmentalization, for ease of comprehension of the multilevel approach. Rather than calculating variance as the deviation of each measurement from the total sample mean as in traditional regression analyses, the multilevel approach separates variance components into deviations of each individual's mean from the total sample mean (between-subject variance), and the deviations of each measurement from the individual's mean (within-subject variance). The between-subject variance provides an estimate of the degree of total variance due to differences between individuals rather than between measurements, while the within-subject variance provides an estimate of the degree of total variance due to differences in measurements within individuals. The hybrid mixed model uses the same approach on independent variables, to evaluate separate between- and within-subject effects of time-varying predictors.

Correlation matrices were generated between individual mean scores (level 2) for fatigue and all level 2 variables. Next, correlation matrices were observed between the change scores (centered level 1) for fatigue and other time-varying variables. Pearson correlations were used for associations with continuous variables, and Spearman correlations for dichotomous variables.

Exploratory multilevel factor analysis was conducted by performing separate principal axis factor analyses on (1) all associated between-subject variables (including mean scores of time-varying variables) and (2) all associated within-subject variables (change scores from individual mean), respectively. The first analysis aimed to identify clustering of variables between patients, while the second analysis aimed to identify clustering of changes in variables across time. Factor retention was decided on the basis of eigenvalue thresholds from parallel analyses of 100 random correlation matrices (95% threshold values),³³ and oblimin rotations were performed so as to allow factors to correlate. Loadings were deemed salient above |0.40|, and factor scores were generated through regression.

Finally, linear multilevel regression was performed with the fatigue factor as a primary outcome variable, to evaluate the relative contributions to fatigue by multilevel factors derived from secondary outcome variables, demographics and time since injury. Observations (level 1) were parameterized as clustered within individuals (level 2). See Figure 1 for an illustration of how multilevel models compartmentalize variance components based on clustering.

The baseline variance component model included no fixed effects. For the final regression model, time-invariant variables were added (ie, demographics and resulting factors from the level 2 factor analyses), along with time-varying variables (ie, time and variables and factors from the level 1 factor analyses). Akaike information criterion and Bayesian information criterion for the models were reported to determine model fit and parsimony. Changes in variance at both levels were calculated to determine to what degree the model predicted between-subject variance and within-subject variance in fatigue, and effect sizes were calculated as the explained variance contributed to the model at each level by each variable. See Supplemental Digital Content Section S5 (available at: https://links.lww.com/JHTR/A636) for further details on analyses, including a script with explanations of the procedures (Section S3).

RESULTS

Sample demographics and injury severity

A schematic presentation of eligible and included patients is shown in Figure 2.

Figure 2.: Flowchart outlining the inclusion and recruitment process, along with an overview of the contribution of each patient to either one or both of the time points.

A total of 103 patients were included, with some variation in their contribution to either one or both time points. See Table 1 for a sample overview. Ninety-six patients were assessed at the first assessment (T1), and 98 at the second assessment (T2), with an average interval between measurements of 220 days (SD = 59.0). There was greater variability in the time point for the second measurement due to restrictions imposed by COVID-19 (see Supplemental Digital Content Section S2, available at: https://links.lww.com/JHTR/A636). In multilevel modeling this issue is handled by allowing time since injury to vary within individuals, and the effects of time can be evaluated despite differing measurement intervals.

TABLE 1 - Distribution of central sample characteristics and the various measures of fatigue at both time points (T1 and T2)

	n (%)	T1	T2
Variable	Level 2	Level 1	Level 1
Total N	103	96	98
Head Injury Severity Scale
Mild	23 (22.3)
Moderate	51 (48.5)
Severe	29 (28.2)
Anatomical injury severity (AIS-head)
2—moderate	3 (2.9)
3—serious	16 (15.5)
4—severe	32 (31.1)
5—critical	52 (50.5)
Cause of Injury
Falls	47 (45.6)
Traffic (including bicycle)	37 (35.9)
Sports-related	6 (5.8)
Violent crime	5 (4.9)
Other or unknown	8 (7.8)
Direct discharge to rehabilitation services	61 (49.5)
Age, median (IQR)	48 (34, 58)
Male	83 (80.6)
Education, mean (SD), y	13.6 (2.4)
Months since injury, mean (SD)		6.9 (1.0)	14.0 (2.1)
FSS, mean (SD)	3.7 (1.5)	3.7 (1.4)	3.8 (1.5)
CFQ, mean (SD)	15.9 (5.7)	16.2 (5.4)	15.5 (6.0)
GSCL fatigue, mean (SD)	1.0 (0.9)	1.0 (0.9)	1.0 (0.9)
RPQ fatigue item		n (%)	n (%)
0 = not a problem		38 (39.6)	38 (38.8)
1 = no longer a problem		11 (11.5)	16 (16.3)
2 = a mild problem		20 (20.8)	18 (18.4)
3 = a moderate problem		20 (20.8)	17 (17.4)
4 = a severe problem		7 (7.3)	9 (9.2)

Abbreviations: AIS, Abbreviated Injury Scale; CFQ, Chalder Fatigue Questionnaire; FSS, Fatigue Severity Scale; GSCL, Giessen Subjective Complaints List; HISS, Head Injury Severity Scale; IQR, interquartile range; RPQ, Rivermead Post-Concussion Symptoms Questionnaire; SD, standard deviation.

Preliminary analyses

Fatigue demonstrated a considerable within-subject stability (intraclass correlation coefficient) of 0.78, indicating that most of the variance in fatigue was due to differences between people, rather than changes within people. Similar patterns indicating trait-like stability were observed for all the included predictors, supporting the use of a hybrid mixed-effects model, which deconstructs variables into between- and within-subject-components.

Multilevel factor analysis (MFA)

Level 2 MFA—between-subject variables

Factor analysis of level 2 between-subject variables confirmed the 3 factors that were found in our previous publication from the first wave only.¹⁵ Factor 1 was termed “psychosocial robustness,” with positive loadings from facets of resilience, trait extraversion, conscientiousness and optimism, and negative loadings from anxiety, depression, loneliness, behavioral inhibition, and trait neuroticism. Factor 2 was termed “somatic vulnerability,” with positive loadings from all measures of pain and somatic symptom burden, as well as daytime sleepiness and insomnia severity. Factor 3 was termed “injury severity,” with positive loadings from AIS-head, direct discharge to rehabilitation, and 3 neuropsychological measures of processing speed, mental flexibility, and intraindividual variability of reaction times/sustained attention (CPT-III Coefficient of Variation). For an overview of variables included in these factors, see Figure 3. Due to missing neuropsychological data in 5 participants at both time points due to color blindness on subtests in the injury severity factor (Color-Word Interference Test subtests 2 and 4), the analyses presented in this article were conducted with an injury severity factor comprised of only the AIS-head and the direct pathway to rehabilitation variables. The same sequence of analyses was conducted in the complete case sample with an injury severity factor incorporating the 3 neuropsychological measures as a sensitivity analysis, with minimal increases in the factor's contributions to later regression models (data not shown).

Figure 3.: A graphical presentation of the findings, with separate between-subjects effects (level 2) and within-subjects effects (level 1).

Level 1 MFA—within-subject variables

Factor analysis of level 1 within-subject change variables supported only one factor, indicating correlated change between several of the independent variables, specifically numerical rating scales of strongest and average pain within the last week, somatic symptom burden, anxiety, depression, and behavioral inhibition. While changes in performance on several neuropsychological measures were correlated with changes in fatigue, these changes did not load saliently on any single factor, indicating that there was no common factor underlying neuropsychological improvement. See Figure 3 for an overview of included variables. Factor loadings from both level 1 and 2 analyses can be inspected in Supplemental Digital Content Section S4 (available at: https://links.lww.com/JHTR/A636).

Linear multilevel regression

Results from the linear multilevel regression are shown in Table 2, and the proportion of explained variance for each variable in the model can be inspected in Supplemental Digital Content Table S5.1 (available at: https://links.lww.com/JHTR/A636). Female sex and education were significantly positively associated with fatigue, each contributing approximately 2% explained variance to the between-subject level in the final regression model. All level 2 factors (ie, somatic vulnerability, psychosocial robustness, and injury severity) were significantly associated with fatigue. Somatic vulnerability uniquely explained 36% of the variance in random intercepts for fatigue, while psychosocial robustness and injury severity uniquely explained 4.5% and 3.9%, respectively. Months since injury explained 6.4% of variance within subjects, while the correlated change factor uniquely explained 17.7% of variance within subjects. In total, the final regression model explained 61.1% of variance between subjects in fatigue, and 21.7% of variance within subjects, summing up to 52.3% variance in total. See Supplemental Digital Content Section S6 (available at: https://links.lww.com/JHTR/A636) for comments on post hoc analyses.

TABLE 2 - Fixed regression coefficients with standard errors, 95% confidence intervals, along with random effects at baseline and for the complete model, with level-wise percentage of explained variance, and fit indices

			95% Confidence interval
Fixed effects (level)	Coefficient	SE	Lowest	Highest
Constant	−0.63	0.28	−1.17	−0.08
Age—centered (2)	0.00	0.00	−0.01	0.00
Sex (2)	0.35a	0.16	0.05	0.66
Years of education—centered (2)	−0.06a	0.03	−0.11	−0.00
Psychosocial robustness (2)	−0.21b	0.07	−0.34	−0.08
Somatic vulnerability (2)	0.58c	0.07	0.44	0.72
Injury severity (2)	0.34b	0.12	0.10	0.57
Correlated change factor (1)	0.15c	0.03	0.08	0.21
Months since injury (1)	−0.02a	0.00	−0.04	−0.01
Random effects (level)	Baseline model	Full model	Explained, %
Between-subject variance (2)	0.75	0.30	61.1
Within-subject variance (1)	0.21	0.17	21.7
Total variance	0.96	0.46	52.3
Fit indices
Log likelihood	−228.89	−175.21
AIC	463.78	372.42
BIC	473.58	408.31
Observations	193	193
Groups	102	102
df	3	11

Abbreviations: AIC, Akaike information criterion; BIC, Bayesian information criterion; SE, standard error.

^aP < .05 for significance of fixed effects.

^bP < .01 for significance of fixed effects.

^cP < .001 for significance of fixed effects

DISCUSSION

Fatigue is associated with a wide range of biopsychosocial factors in the first year following TBI, and the present study examined between-subject associations with fatigue, and furthermore evaluated within-subject changes associated with changes in fatigue from 6 to 12 months. The findings highlight several biopsychosocial determinants for identifying patients at high risk for developing fatigue following TBI, and also factors associated with increases or decreases in fatigue within individuals.

Between-subject effects

Factor analyses replicated similar underlying dimensions in trait-like stability of predictors as reported in previous cross-sectional analysis of the first wave of this study.¹⁵ However, these between-subject factors demonstrated more robust effects in multilevel regressions than in our previous cross-sectional design, now explaining 61% of the variance in fatigue between individuals. Female sex also demonstrated a small, but significant positive association with fatigue, along with a slight negative association between years of education and fatigue.

These findings emphasize that through the first year post-injury, knowing the sex of the patient, their educational level, their degree of somatic vulnerability and psychosocial robustness, and initial injury severity allowed us to distinguish significantly between individuals regarding their risk for fatigue after TBI.

Between-subject effects are, however, prone to confounding from potentially shared causes. Shared genetic susceptibility for fatigue, pain, and psychological distress has been demonstrated earlier,^20,34,35 and shared risk for and resilience to fatigue and associated factors might additionally be accumulated through an individual's idiosyncratic life experiences prior to and following injury. Thus, while demonstrating that individuals with higher psychosocial robustness (ie, lower levels of trait neuroticism, behavioral inhibition, loneliness and psychological distress, and higher levels of conscientiousness, extraversion, resilient coping, and optimism) have significantly lower levels of fatigue, this does not automatically imply any of these factors as crucial to the within-subject process of increasing or decreasing fatigue across time. For this, an evaluation of longitudinal within-subject effects is necessary.

Within-subject effects

Within-subject changes from 6 to 12 months in pain and somatic symptom burden, depression, anxiety, and behavioral inhibition were correlated with one another, and loaded on a single factor, which was positively associated with changes in fatigue. Thus, increases or decreases in fatigue from 6 to 12 months demonstrate significant synchronous changes in pain, somatic symptoms, psychological distress, and behavioral inhibition. These within-subject changes, along with the time-dependent decrease in fatigue, explained approximately 22% of within-subject variance in fatigue. This is in line with the findings by Rakers et al,¹⁸ linking persistence and recovery of fatigue with the presence or absence of emotional distress, pain, and active or passive coping styles. Behavioral inhibition, which demonstrated a significant within-subject association with fatigue in our study, can be conceptualized as a propensity for avoidance of unpleasant and novel sensations, and is linked with passive coping. The correlated change between behavioral inhibition and fatigue thus aligns well with the results from Rakers et al,¹⁸ in demonstrating that increases or decreases in behavioral inhibition are associated with increases or decreases in fatigue, respectively, which could imply compensatory changes in coping strategies for the management of fatigue.

Implications for research and rehabilitation

The findings support the notion of fatigue following TBI as a multifactorial phenomenon, associated with a wide range of biopsychosocial factors 6 to 12 months following injury. While there is a significant reduction in fatigue in this period, our study demonstrates that some factors only help to distinguish levels of fatigue between individuals, while others inform us of characteristics of those who experience reductions or increases in fatigue from 6 to 12 months. The 2 levels of analyses inform us of different, but clinically important ways to understand the development of fatigue following injury. See Figure 3 for a graphical overview of the results, for ease of comprehension in the following discussion.

Between-subject factors are important for clinicians and researchers preoccupied with the question of which patients might be at risk for experiencing fatigue. Thus, if a female patient with severe TBI initially reports pain and daytime sleepiness (indicators of somatic vulnerability), depressive symptoms, and high levels of trait neuroticism (negative indicators of psychosocial robustness), there is a considerable likelihood that she will also experience fatigue. The underlying cause for the co-occurrence between fatigue and these factors cannot be delineated in this study, but one may presume that their co-occurrence may be due to shared genetic, injury-related, and environmental causes,³⁵ as well as potential reciprocal pathways between them over time.

Within-subject factors are important for clinicians and researchers preoccupied with the question of which mechanisms drive changes in fatigue. Therefore, if the aforementioned patient did suffer from fatigue, a reduction or increase in fatigue to a follow-up 6 months later would in part be dictated by synchronous changes in pain and psychological distress, but not daytime sleepiness or trait neuroticism, according to our findings. Thus, individualized rehabilitation aimed at ameliorating fatigue should focus on the simultaneous treatment of pain and psychological distress, as their development is correlated with changes in fatigue. For instance, self-reported personality traits would not be expected to change significantly over a 6-month interval due to their relatively stable nature,³⁶ but they are nevertheless linked with fatigue between subjects. Pain, somatic symptoms, behavioral inhibition, and psychological distress, on the other hand, fluctuate within individuals in association with fatigue. Further studies examining the development of fatigue would be well served in also tracking changes in these associated factors, to unravel potential causal pathways among them.

Limitations

The study has some limitations. It did not include patients without intracranial pathology, and the results may not generalize to this cohort of the TBI population. While the large battery of instruments employed allows exploration of overlap between variables commonly associated with fatigue, the exploratory approach combined with a relatively small sample size might affect generalizability. As with all dimension reduction techniques, a parsimonious structure is sought at the cost of complexity. There is, however, reason to trust the general pattern of findings, and the within-subject associations partially replicate previous findings from an unrelated, nonclinical sample.³⁵ Directional causality cannot, however, be concluded, as the synchronous changes might imply several potential directional pathways between fatigue and associated factors. As alternative editions do not exist and were not employed for the neuropsychological measures, retest effects cannot be segregated from cognitive improvement, which might be one reason we did not find a within-subject neuropsychological change factor. Finally, longitudinal studies with more than 2 measurements could potentially capture more within-subject variability and map individual differences in trajectories across time, and further studies are warranted.

CONCLUSION

The study used a multilevel approach to explore stable and time-varying relationships between fatigue and commonly implicated biopsychosocial factors. While sex, preinjury educational attainment, injury severity, and neuropsychological function explained significant variance in fatigue, the vast majority of explained variance was due to self-reported biopsychosocial constructs. Furthermore, the multilevel approach allowed us to disentangle between-subject risk and protective factors, and to single out within-subject factors crucial to changes in fatigue from 6 to 12 months following injury.