If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Neuromedicine and Movement Science, Norwegian University of Science and Technology (NTNU), Trondheim, NorwayDepartment of Physical Medicine and Rehabilitation, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
Department of Neuromedicine and Movement Science, Norwegian University of Science and Technology (NTNU), Trondheim, NorwayDepartment of Neurosurgery, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
Department of Psychology, Norwegian University of Science and Technology (NTNU), Trondheim, NorwayDepartment of Neurology and Clinical Neurophysiology, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
Department of Physical Medicine and Rehabilitation, St. Olavs Hospital, Trondheim University Hospital, Trondheim, NorwayDepartment of Psychology, Norwegian University of Science and Technology (NTNU), Trondheim, Norway
Department of Neuromedicine and Movement Science, Norwegian University of Science and Technology (NTNU), Trondheim, NorwayDepartment of Physical Medicine and Rehabilitation, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
Department of Neuromedicine and Movement Science, Norwegian University of Science and Technology (NTNU), Trondheim, NorwayDepartment of Radiology and Nuclear Medicine, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
Department of Neuromedicine and Movement Science, Norwegian University of Science and Technology (NTNU), Trondheim, NorwayDepartment of Neurosurgery, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway
Department of Physical Medicine and Rehabilitation, St. Olavs Hospital, Trondheim University Hospital, Trondheim, NorwayDepartment of Psychology, Norwegian University of Science and Technology (NTNU), Trondheim, Norway
Department of Physical Medicine and Rehabilitation, Harvard Medical School, Boston, MassachusettsSpaulding Rehabilitation Hospital and Spaulding Research Institute, Charlestown, MassachusettsHome Base, A Red Sox Foundation and Massachusetts General Hospital Program, Charlestown, Massachusetts
To describe personal factors in patients with mild traumatic brain injury (MTBI) and 2 control groups and to explore how such factors were associated with postconcussion symptoms (PCSs).
Design
Prospective cohort study.
Setting
Level 1 trauma center and outpatient clinic.
Participants
Participants (N=541) included patients with MTBI (n=378), trauma controls (n=82), and community controls (n=81).
Main Outcome Measures
Data on preinjury health and work status, personality, resilience, attention deficit/hyperactivity, and substance use. Computed tomography (CT) findings and posttraumatic amnesia were recorded. Symptoms were assessed at 3 months with the British Columbia Postconcussion Symptom Inventory and labeled as PCS+ if ≥3 symptoms were reported or the total score was ≥13. Predictive models were fitted with penalized logistic regression using the least absolute shrinkage and selection operator (lasso) in the MTBI group, and model fit was assessed with optimism-corrected area under the curve (AUC) of the receiver operating characteristic curve.
Results
There were few differences in personal factors between the MTBI group and the 2 control groups without MTBI. Rates of PCS+ were 20.8% for the MTBI group, 8.0% for trauma controls, and 1.3% for community controls. In the MTBI group, there were differences between the PCS+ and PCS− group on most personal factors and injury-related variables in univariable comparisons. In the lasso models, the optimism-corrected AUC for the full model was 0.79, 0.73 for the model only including personal factors, and 0.63 for the model only including injury variables. Working less than full time before injury, having preinjury pain and poor sleep quality, and being female were among the selected predictors, but also resilience and some personality traits contributed in the model. Intracranial abnormalities on CT were also a risk factor for PCS.
Conclusions
Personal factors convey important prognostic information in patients with MTBI. A vulnerable work status and preinjury health problems might indicate a need for follow-up and targeted interventions.
After mild traumatic brain injury (MTBI), a significant minority experiences a range of persistent somatic, cognitive, and emotional symptoms over months or even years, a condition here referred to as postconcussion symptoms (PCSs).
and the disability some people experience can therefore seem out of proportion to a TBI that is categorized as mild. The heterogeneity in MTBI outcome, together with the inconsistent findings on the role of injury severity and neuroimaging findings on outcome, suggest that the biopsychosocial model may serve as a framework to understand and treat PCSs.
Systematic review of self-reported prognosis in adults after mild traumatic brain injury: results of the International Collaboration on Mild Traumatic Brain Injury Prognosis.
Is sex an indicator of prognosis after mild traumatic brain injury: a systematic analysis of the findings of the World Health Organization Collaborating Centre Task Force on Mild Traumatic Brain Injury and the International Collaboration on Mild Traumatic Brain Injury Prognosis.
In the Trondheim MTBI follow-up study, we acquired comprehensive information in patients and control participants without MTBI about a range of personal factors such as personality, resilience, optimism, preinjury somatic and psychological health, experienced life events, and substance use, in addition to measures of injury severity. Our aims were to describe these in patients with MTBI and 2 control groups, and to explore whether such factors were associated with MTBI and PCSs. We built prediction models and explored which of the personal factors were most consistently associated with PCSs after MTBI and to what degree the injury-related variables added to model performance.
Methods
Participants
The patients with MTBI were part of the Trondheim MTBI follow-up study (n=378), a study consisting of patients 16-59 years old presenting from April 2014 to December 2015 at 2 emergency departments: St. Olavs Hospital, Trondheim University Hospital, a level 1 trauma center and the Trondheim Municipal Emergency clinic, a general practitioner-run, outpatient clinic.
In this cohort, around 60% of all eligible patients were enrolled, and the cohort has been shown to be representative of all young adult and middle-aged patients with MTBI in the catchment area.
categorized as mild according to the World Health Organization Collaborating Centre Task Force on Mild Traumatic Brain Injury criteria. All had Glasgow Coma Scale scores of 13-15 at presentation in the emergency department and either witnessed loss of consciousness <30 minutes, confusion, or posttraumatic amnesia (PTA) <24 hours. Patients that met these criteria and had traumatic lesions on computed tomography (CT) scans were included in the study if the intracranial lesion did not require surgery.
Methodological issues and research recommendations for mild traumatic brain injury: whe WHO Collaborating Centre Task Force on Mild Traumatic Brain Injury.
Exclusion criteria were (1) nonfluency in the Norwegian language; (2) pre-existing severe neurologic disorder (eg, stroke, multiple sclerosis) or a prior history of a complicated mild, moderate, or severe TBI; or (3) ongoing psychiatric (eg, psychotic or bipolar disorder), health (eg, cancer), or substance abuse problems determined by the researcher responsible for inclusion and considered to be severe enough to likely interfere with follow-up.
The control groups were (1) trauma controls (TCs): 82 patients with minor orthopedic injuries, matched on the group level on age and sex, using a procedure ensuring similar numbers of men and women in each 5-year age interval and (2) community controls (CCs): 81 persons recruited among staff, family, and acquaintances of patients and staff, matched on age, sex, and education, using a procedure also ensuring similar years of education in each 5-year age interval. The exclusion criteria applied to the MTBI group were used for the controls, but in addition, the TCs did not have trauma affecting the head, neck, or dominant arm, and the CCs were not receiving treatment for psychiatric disorders.
The regional committee for research ethics approved the study. Participants and parents of participants younger than 18 years in the cohort study gave informed consent.
Study procedures
Patients were consecutively identified based on lists of emergency department visits and CT referrals, and they were approached and enrolled as previously described.
Data on preinjury characteristics (personal factors) were collected during the first week after injury, considering the situation as it was before their injury. Both a structured interview and questionnaires were used. At 3 months after injury, the patients underwent an outcome evaluation by a telephone interview.
Study variables
PTA was defined as the self-reported time after injury for which the patient had no memory. From a pilot study, we knew that many participants would not be able to provide a valid estimate in minutes; therefore, a predefined option was to record PTA as either <1 hour or 1-24 hours. Intracranial traumatic findings were obtained from the radiology report of the acute head CT, here recorded as yes/no. Fractures of the cranium, face, or other bones were obtained from the radiology report and recorded as yes/no.
Personal factors assessed were all preinjury: age, sex, education, school marks, reading difficulties, work status, previous MTBI, headache, other pain, psychiatric problems, substance use, alcohol use, sleep quality, attention deficit hyperactivity disorder (ADHD) symptoms, personality traits, life orientation (optimism/pessimism), threatening events, and resilience. Measures, definitions, and cutoff criteria are shown in table 1.
Table 1Personal factors in the study describing preinjury status
Variables
Measures and Categorizations
Collection Methods and Measure Details
Age
Years
Medical records
Sex
Man or woman
Medical records
Education
Years of completed education. Starting from the first year of school, at 6 y of age.
Self-report, interview
School marks
Average high school marks on a 1-6 scale.
Self-report, interview
Reading difficulties
Recorded as “yes” if the person had been diagnosed as having reading difficulties.
Self-report, interview
Reduced work
Recorded as “yes” if the person was working or studying <80% (of a 37.5-hr wk).
Self-report, interview
Previous MTBI
Recorded as “yes” if the person had sustained ≥1 head traumas likely to have fulfilled the same criteria for MTBI as applied in this study.
Self-report, interview
Preinjury pain
Recorded as “yes” if the person had nonheadache pain in any part of the body graded ≥3 on a 0-10 NRS
Pain map and NRS
Preinjury headache
Recorded as “yes” or “no”
Item from self-report questionnaire. “Have you suffered from headache during the last year?”
Psychiatric history
Recorded “yes” if the person reported a history of psychiatric illness.
Self-report, interview
Substance use
Recorded “yes” if the person reported using drugs other than alcohol.
the first 3 items: difficulties falling asleep, staying asleep, and waking up to early.
ISI is a questionnaire, but the first 3 questions were administered as an interview. Self-report questionnaire. A 5-point Likert scale, higher scores indicate greater sleep problems/poor sleep quality.
Individual scores for inattention, hyperactivity, and total symptom burden. The scores were calculated both for the screening part of the questionnaire (the first 6 items) and the full scale (all 18 items). The total score was calculated for each scale. A likely diagnosis of ADHD was defined as scoring at or above a threshold (“sometimes” on questions 1-3 and “often” on questions 4-6) on at least 4 of the first 6 items.
Development of the Alcohol Use Disorders Identification Test (AUDIT): WHO collaborative project on early detection of persons with harmful alcohol consumption--II.
Development of the Alcohol Use Disorders Identification Test (AUDIT): WHO collaborative project on early detection of persons with harmful alcohol consumption--II.
a short form of the BFI including 44 items. The mean score for each scale was calculated.
Self-report questionnaire yielding individual scores on extroversion, agreeableness, conscientiousness, neuroticism, and openness. Higher scores indicate higher levels of that personality trait. At least 50% of the items on each personality domain had to be answered for that scale to be calculated.
Self-report questionnaire with ten items, 6 of them measuring optimism/pessimism and 4 fillers (not scored items). Lower scores indicate higher optimism and the variable is therefore referred to as “pessimism.” Two missing items were accepted.
Validation of the Long-term Difficulties Inventory (LDI) and the List of Threatening Experiences (LTE) as measures of stress in epidemiological population-based cohort studies.
Self-report questionnaire measuring experience of environmental stressful events during the last year. The Norwegian version comprised 13 items. Two missing items were accepted on LTE-Q.
The mean score was calculated for all dimensions separately, and total resilience score was the mean of all item scores.
Self-report questionnaire with 33 items measuring 6 dimensions (perception of self, planned future, social competence, family cohesion, social resources, structured style) and a score of total resilience. Higher scores indicate higher resilience. Three missing items were accepted on RSA.
Abbreviations: ISI, Insomnia Severity Index; LTE-Q, List of Threatening Events Questionnaire; NRS, Numeric Rating Scale; RSA, Resilience Scale for Adults.
The British Columbia Postconcussion Symptom Inventory consists of 13 core symptoms distributed over 4 symptom categories and 3 life problems. In the current study, the 13 core symptoms were used to calculate the total score. PCSs were defined as having at least 3 core symptoms rated as at least moderate (score ≥3) or a total score ≥13. PCSs were assessed both in the MTBI and control groups and refer to a high level of postconcussion/postconcussion-like symptoms. The MTBI group was accordingly divided into a PCS+ and a PCS− group.
Statistical analyses
Missing values on single questionnaire items were replaced by that individual’s mean of the answered questions on the scale/questionnaire. The proportion of missing items across questionnaires ranged from 0.12%-0.44%. Very few participants had more than 1 item missing on a returned questionnaire: Adult ADHD Self-Report Scale, respondents with ≥1 missing items 3.10% and respondents with 2 missing items 0.48%; the Alcohol Use Disorders Identification Test, respondents with 1 missing item 1.17% and none had ≥2 missing items; the Big Five Inventory, respondents with ≥1 missing items 10.88% and respondents with ≥2 missing items 3.98%; Life Orientation Test-Revised, respondents with 1 missing item 1.64% and none had ≥2 missing items; List of Threatening Events Questionnaire, respondents with ≥1 missing item 2.36% and respondents with 2 missing items 0.23%; the Resilience Scale for Adults, respondents with ≥1 missing items 4.00% and respondents with ≥2 missing items 0.24%.
Group differences between the MTBI group, the TC group, and the CC group were evaluated with Kruskal-Wallis tests and chi-square tests. Whether personal and injury-related factors predicted PCSs in patients with MTBI was evaluated with logistic regression models. First, univariable analyses for each variable of interest were conducted. To reduce the number of comparisons, we only analyzed 1 measure from most questionnaires (eg, from the Adult ADHD Self-Report Scale only the total score was analyzed). Odds ratios with 95% CIs are reported. For the group comparisons and the univariable analyses, P values uncorrected for multiple comparisons are reported as well as differences that remained statistically significant after Bonferroni correction. Second, models were fitted by penalized logistic regression using the least absolute shrinkage and selection operator (lasso) as implemented in the Stata command lasso logit. To account for many predictors, lasso shrinks the values of the coefficients to less extreme values, thereby improving the external validity of the model. For variables with low predictive value, the coefficients could be shrunk (set) to 0 and be left out of the final model. The degree of shrinkage was determined by 10-fold cross-validation. In effect, the method performs estimation of the coefficients and variable selection simultaneously and provides estimates of overall fit rather than statistical significance of the individual predictors. Stata handles factor variables by including all dummy variables (ie, none is initially left out as reference category), and the lasso procedure then shrinks 1 or more of the dummy variables to 0. The uncertainty of the coefficients from the lasso was assessed by repeating the penalized regression procedure in 1000 bootstrap samples. The uncertainty for each of the variables was assessed by the proportion of the 1000 bootstrap samples when its coefficient was set to 0. The lower proportion, the higher is the probability that a variable contributes to outcome prediction. The area under the curve (AUC) of the receiver operating characteristic curves was used to assess performance of the models. Optimism-corrected AUCs with 95% CIs were obtained from bootstrapping with 1000 replications. In this internal validation procedure, the model is estimated in the bootstrap samples and then tested in the original sample. The mean difference between the AUCs obtained in the bootstrap samples and the original sample is referred to as the “optimism,” which is subtracted from the AUC obtained in the original model.
In the univariable analyses, we allowed for a different number of observations between analyses (ie, all available data were used), while in the lasso models only complete cases were included.
Results
Overall characteristics of the MTBI group and the control groups
There were small and mostly statistically nonsignificant differences between the MTBI, TC, and CC groups (table 2). More people in the MTBI group reported prior MTBI, reading difficulties, and a psychiatric history, and the higher frequency of prior MTBI remained statistically significant after Bonferroni correction.
Table 2Comparisons between the MTBI group, the trauma control group, and the community control group
Not working or studying at the time of injury at all (“full time not working”), or working or studying to some extent but <80% (“part time not working.”)
The comparisons remained statistically significant after Bonferroni correction for multiple comparisons (critical P value=.002).
Preinjury pain, yes, n (%)
372
71 (19.1)
81
9 (11.1)
74
14 (18.9)
.228
Preinjury headache, yes, n (%)
272
83 (30.5)
76
21 (27.6)
76
29 (38.2)
.331
Psychiatric history, yes, n (%)
376
61 (16.2)
81
9 (11.1)
81
5 (6.2)
.044
Substance use, yes, n (%)
369
29 (7.9)
81
5 (6.2)
80
12 (15.0)
.083
Poor preinjury sleep quality (ISI), median (IQR)
370
0.00 (0.00-0.33)
81
0.00 (0.00-0.33)
81
0.00 (0.00-0.67)
.538
ADHD symptoms (ASRS), median (IQR)
264
80
76
Screener inattention
5.00 (3.00-5.00)
5.00 (3.00-6.00)
5.00 (4.00-6.75)
.315
Screener hyperactivity
3.00 (2.00-5.00)
3.00 (2.00-5.00)
3.00 (2.00-4.00)
.726
Screener total
8.00 (6.00-10.00)
7.00 (5.00-10.00)
8.00 (6.00-10.75)
.392
Full scale inattention
12.00 (9.00-14.00)
11.00 (8.25-14.75)
12.00 (9.25-15.00)
.406
Full scale hyperactivity
10.00 (7.00-14.00)
9.00 (7.00-13.75)
11.00 (7.25-13.75)
.425
Full scale
22.50 (17.00-27.75)
21.00 (16.00-26.00)
23.50 (18.25-28.00)
.345
Probable ADHD diagnosis, n (%)
24 (9.1)
7 (8.8)
10 (13.2)
.543
Alcohol use (AUDIT)
272
80
76
Full scale, median (IQR)
6.00 (3.00-10.00)
5.00 (4.00-8.00)
6.00 (4.00-9.00)
.371
≥8, n (%)
107 (39.3)
24 (30.0)
26 (34.2)
.277
Personality (BFI), median (IQR)
271
80
76
Extroversion
4.75 (4.00-5.38)
4.63 (3.88-5.25)
4.63 (4.00-5.25)
.469
Agreeableness
5.33 (4.89-5.89)
5.44 (4.89-6.11)
5.39 (4.78-6.00)
.722
Conscientiousness
5.00 (4.44-5.67)
5.11 (4.33-5.78)
5.06 (4.11-5.53)
.389
Neuroticism
3.00 (2.38-3.75)
3.12 (2.13-3.75)
3.00 (2.41-3.63)
.812
Openness
4.60 (4.00-5.30)
4.70 (4.03-5.20)
4.65 (4.03-5.30)
.875
Pessimism (LOT-R), median (IQR)
270
1.17 (0.83-1.67)
80
1.17 (0.83-1.83)
76
1.17 (0.80-1.67)
.796
Threatening Life events (LTE-Q), median (IQR)
269
1.00 (0.00-2.00)
79
1.00 (0.00-2.00)
76
1.00 (0.00-2.00)
.355
Resilience (RSA), median (IQR)
269
80
76
Perception of self
5.33 (4.36-6.00)
5.50 (4.21-6.00)
5.33 (4.38-5.83)
.907
Planned future
5.75 (4.37-6.00)
5.75 (4.75-6.25)
5.50 (4.83-6.33)
.757
Social competence
5.33 (4.50-6.00)
5.33 (4.50-6.17)
5.50 (4.83-6.33)
.215
Family cohesion
5.67 (4.83-6.17)
5.83 (5.17-6.17)
5.67 (4.83-6.33)
.822
Social resources
6.14 (5.71-6.71)
6.28 (5.75-6.71)
6.07 (5.57-6.68)
.464
Structured style
5.00 (4.25-5.75)
5.25 (4.50-6.00)
5.00 (4.06-5.75)
.498
Total resilience
5.51 (5.00-5.94)
5.66 (5.08-5.96)
5.56 (4.89-6.06)
.765
Abbreviations: ASRS, Adult ADHD Self-Report Scale; AUDIT, Alcohol Use Disorders Identification Test; BFI, Big Five Inventory; IQR, interquartile range; ISI, mean of the first 3 items of Insomnia Severity Index; LOT-R, Life Orientation Test-Revised; LTE-Q, List of Threatening Events Questionnaire; RSA, Resilience Scale for Adult.
∗ The comparisons remained statistically significant after Bonferroni correction for multiple comparisons (critical P value=.002).
† Not working or studying at the time of injury at all (“full time not working”), or working or studying to some extent but <80% (“part time not working.”)
There were 70 patients (20.8%) in the MTBI group with PCSs (the PCS+ group) 3 months after injury. In the control groups, 6 (8%) of the TCs and 1 (1.3%) of the CCs met PCS criteria.
Correlations between personal factors
A bivariate correlation matrix illustrating associations between the preinjury personal characteristics is presented in fig 1. There were very small correlations for many of the variables, such as age, previous MTBI, pain, headache, alcohol use, and experienced life events. Resilience correlated positively with extroversion, agreeableness, and conscientiousness and negatively with pessimism, ADHD symptoms, and neuroticism. School marks correlated positively with education, psychiatric history correlated with reduced employment, ADHD symptoms correlated positively with neuroticism and negatively with conscientiousness, and female sex correlated positively with neuroticism.
Fig 1Spearman correlations (continuous-continuous associations), rank-biserial correlations (nominal-continuous associations), and phi coefficients (nominal-nominal associations) between the personal factors in the MTBI group.; Abbreviations: ADHD symptoms, full scale of the Adult ADHD Self-Report Scale; alcohol use, full scale of the Alcohol Use Disorders Identification Test; Extroversion, Agreeableness, Conscientiousness, Neuroticism, and Openness, subscales from the Big Five Inventory; pessimism, full scale of the Life Orientation Test-Revised; poor preinjury sleep quality, mean of the 3 first items in Insomnia Severity Index; threatening life events, List of Threatening Events Questionnaire, total nos. of events; resilience, full scale from the Resilience Scale for Adults.
Comparison of the PCS+ and PCS− groups in univariable analyses
More individuals in the PCS+ group had intracranial lesions on CT and long PTA duration. Regarding the personal factors, many statistically significant differences were revealed. In the PCS+ group, there were more women; more worked less than full time before the injury; and more reported preinjury pain, psychiatric problems, poor sleep quality, and headache. They had higher scores on the measure of ADHD symptoms and neuroticism, and they had lower scores on extroversion. The PCS+ group also reported lower resilience, higher pessimism, and more threatening life events (table 3).
Table 3Associations between PCSs and personal and injury-related factors at 3 mo in patients with MTBI (univariate logistic analyses)
Statistically significant after Bonferroni correction for multiple comparisons (critical P value=.002).
Abbreviations: ASRS, full scale of the Adult ADHD Self-Report Scale; AUDIT, full scale of the Alcohol Use Disorders Identification Test; BFI, Big Five Inventory; Extroversion, Agreeableness, Conscientiousness, Neuroticism, and Openness, subscales from the Big Five Inventory; IQR, interquartile range; ISI, mean of the 3 first items in Insomnia Severity Index; LOT-R, full scale of the Life Orientation Test-Revised; LTE-Q, List of Threatening Events Questionnaire, total nos. of events; OR, odds ratio; RSA, full scale from the Resilience Scale for Adults.
∗ Statistically significant after Bonferroni correction for multiple comparisons (critical P value=.002).
Table 4 shows the predictors selected and their penalized coefficients from the full lasso model. The AUC for the selected model was 0.864 (optimism-corrected, 0.790; 95% CI, 0.724-0.857). In the 1000 bootstrap samples, working less than full time, preinjury pain, CT findings, not being triaged to CT, female sex, preinjury headache, poor preinjury sleep quality, openness, long PTA, and total resilience were the 10 predictors most often not set to 0 (ie, the most important variables in predicting PCSs) (fig 2).
Table 4Penalized coefficients from a lasso model predicting PCSs (complete case analysis: PCS+ n=42; PCS− n=192)
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
−0.33
0.72 (0.33-1.00)
28.7
Abbreviations: ASRS, full scale of the Adult ADHD Self-Report Scale; AUDIT, full scale of the Alcohol Use Disorders Identification Test; BFI, Big Five Inventory; Extroversion, Agreeableness, Conscientiousness, Neuroticism, and Openness, subscales from the Big Five Inventory; ISI, mean of the 3 first items in Insomnia Severity Index; LOT-R, full scale of the Life Orientation Test-Revised; LTE-Q, List of Threatening Events Questionnaire, total nos. of events; OR, odds ratio; Resilience, full scale from the Resilience Scale for Adults; RSA, Resilience Scale for Adults.
∗ Variables selected by lasso. A coefficient of 0 means that the variable was not selected by lasso. 95% CIs and the proportion of times the variables were set to 0 and were obtained from bootstrapping with 1000 replications. CT-no findings was not selected by lasso and is therefore reference category for this factor variable.
Fig 2Factors most often set to 0. NOTE. The histogram shows the percentage of the 1000 bootstrap samples that gave coefficients equal to 0 for each variable. Proportions closer to 0 indicate greater probability for the variable to be included in the model. Abbreviations: ASRS, full scale of the Adult ADHD Self-Report Scale; AUDIT, full scale of the Alcohol Use Disorders Identification Test; Extroversion, Agreeableness, Conscientiousness, Neuroticism, and Openness, subscales from the Big Five Inventory; ISI, mean of the 3 first items in Insomnia Severity Index; LOT-R, full scale of the Life Orientation Test-Revised; LTE-Q, List of Threatening Events Questionnaire, total nos. of events; resilience, full scale from the Resilience Scale for Adults.
In a model excluding the injury-related variables, female sex, working less than full time, preinjury pain, poor sleep quality, headache, ADHD symptoms, higher neuroticism and openness, and lower extroversion and resilience were selected as predictive of PCSs, and the AUC was 0.805 (optimism-corrected, 0.734; 95% CI, 0.657-0.814). In a model that only included the injury-related variables, selected predictors for PCSs were intracranial lesions on CT, not being triaged to CT, and longer PTA duration, and the AUC was lower at 0.661 (optimism-corrected, 0.631; 95% CI, 0.551-0.700).
In a subgroup of patients without intracranial lesions on CT, female sex, working less than full time, preinjury pain, poor sleep quality, headache, ADHD symptoms, openness, and resilience were predictors, and the AUC was 0.818 (optimism-corrected, 0.734; 95% CI, 0.647-0.814).
Discussion
This study illustrates that a diverse range of personal factors are associated with having persistent symptoms 3 months after an MTBI. Women were more likely than men to have persistent symptoms, and being unemployed or working less than full time before the injury was a predictor for having persistent symptoms, as were preinjury health problems and personality characteristics. Having intracranial traumatic lesions on head CT was also associated with PCSs but was less important than the personal factors in the multivariable model.
Female sex was one of the factors most often included in the multivariable PCS prediction model. Thus, sex was a predictor for PCSs, even when some possibly confounding variables were controlled for. Previous reviews have concluded that the literature is inconclusive when it comes to female sex being a risk factor for PCSs
Is sex an indicator of prognosis after mild traumatic brain injury: a systematic analysis of the findings of the World Health Organization Collaborating Centre Task Force on Mild Traumatic Brain Injury and the International Collaboration on Mild Traumatic Brain Injury Prognosis.
and called for further research. We consider our study to add important evidence in this respect, given that we were able to control for a broad range of personal factors that could otherwise confound the association between sex and PCSs.
Importantly, we found that reduced employment was a strong and unique predictor of PCSs, even when possibly confounding variables were controlled for in the multivariable lasso model (eg, poor sleep quality and psychiatric history, which correlated moderately with reduced employment). Unemployment has previously been found to predict PCS after MTBI
Predicting long-term global outcome after traumatic brain injury: development of a practical prognostic tool using the Traumatic Brain Injury Model Systems national database.
Reduced employment might be a vulnerability factor or full-time employment might be a protective factor. For example, having a structured daily activity to return to after MTBI may be motivating and colleagues at the workplace may provide vital support. Efforts to promote an effective and durable return to work are likely important during early rehabilitation after MTBI.
Preinjury health factors, such as having a psychiatric history, symptoms of ADHD, headaches, and sleep problems, were all significant predictors. Having preinjury bodily pain emerged as a particularly strong predictor. Preinjury health problems as a risk factor for PCSs have also been reported in other studies,
Pre-injury comorbidities are associated with functional impairment and post-concussive symptoms at 3- and 6-months after mild traumatic brain injury: a TRACK-TBI study.
It was expected that preinjury health factors would be associated with PCSs. In fact, some of the preinjury factors assessed in the present study also constitute symptoms of PCSs (eg, headache). This illustrates the complexity of assessing PCSs because the symptoms are nonspecific for MTBI and they are common in the general population.
Thus, it is challenging to separate symptoms related to preinjury health factors from MTBI-related symptoms. However, the prevalence of PCSs was considerable higher in patients with MTBI than in TCs and CCs, despite very small differences in preinjury health factors between these groups (see table 2), suggesting that the combined effect of preinjury health factors and MTBI is associated with the development of PCSs.
Personality characteristics were also related to persistent symptoms. Not surprisingly, neuroticism, a personality trait linked to an increased tendency of experiencing negative emotions and distress, was associated with PCSs, in line with other studies of patients with MTBI
Stress Management and Resilience Training (SMART) program to decrease stress and enhance resilience among breast cancer survivors: a pilot randomized clinical trial.
Interestingly, this was not one of the variables most often included in our lasso multivariable model. The reason for this might be that other correlated variables, such as low resilience, high neuroticism, and reduced employment, have greater explanatory power.
Some injury severity indicators were associated with outcome. In the present study, head CT had been performed in 79% of the sample, and of those, only 6% had intracranial traumatic lesions (ie, complicated MTBI). Patients with complicated MTBI had higher risk for PCSs, which is in line with 1 previous study
PTA duration >1 hour was also associated with PCSs, albeit weakly and not in the uncomplicated MTBI group. Interestingly, not being triaged for CT, which probably can be considered a proxy for having fewer symptoms and an overall better clinical presentation in the emergency department, was associated with a reduced risk of PCSs.
One aim of this study was to report the discriminative ability of different models using the AUC of the receiver operating characteristic curve and to explore how a broad range of personal factors contributed to model performance. In the full model, we found an AUC value of 0.79 after optimism correction, which is high compared with other reported multivariable models in MTBI.
When injury-related variables were removed, the AUC was only modestly reduced to 0.73. In contrast, a model with only injury severity variables performed poorly (AUC, 0.63). In the subsample with uncomplicated MTBI, the full model also performed adequately (AUC, 0.73), and interestingly, PTA duration was not selected in this model. Positive CT findings were clearly related to having persistent symptoms, but because it is uncommon to have an abnormal head CT after MTBI, combining this and other injury severity variables did not produce a useful prediction model.
Study limitations
This study has several limitations. First, preinjury personal factors were all based on self-report, and they were all collected after the injury, making them susceptible to some degree of reporting bias or inaccuracy. This limitation is, for the most part, unavoidable. Second, there was a low rate of complicated MTBIs in this study, which reduces the statistical power of that variable for predicting persistent symptoms. Importantly, however, we consider that a natural consequence of recruiting a more representative cohort of people with MTBIs from both ambulatory clinics and the emergency department. Third, although our sample size was large (N=378), the prevalence of some potentially important preinjury predictor variables in the cohort was small, and thus, power was reduced in analyses of the influence of those predictors (eg, reading difficulties, reduced employment, possible ADHD). Fourth, the number of people with PCSs was modest, and a larger sample of people with PCSs would have allowed for analyses on how prognostic factors moderate each other. Fifth, in the multivariable models, only complete cases were included. Using multiple imputation in lasso models is less straightforward than in ordinary regression models because lasso performs variable selection. However, in the univariable analyses, we could use all data available (ie, the number of observations differed between analyses), and the most important factors for predicting PCSs in the lasso models were also strongly associated with PCSs in the univariable models, further supporting their role in the development of PCSs. Sixth, the CC group was not population-based, and the rate of PCSs in the CC group should be interpreted with caution. However, the groups were matched, not only on age and sex but also on education, and there were only small differences between the MTBI group, the TC group, and the CC group on the factors that predicted PCSs. Seventh, litigation is associated with PCSs in other studies and countries, and we did not collect information on this variable.
However, in our study, conducted in a country where health care is free and people have access to sickness and disability benefits from the government, a litigation process is seldom initiated in the first few months, when our study was conducted. Also, results from this study were solely for research purposes, not available for anyone else; hence, no financial gain could be obtained through participation. Finally, the relative strength of risk factors may differ in other settings and in other countries, as illustrated in a study of moderate TBI in Norway and the Netherlands,
We collected a wider range of personal factors than prior studies and found that several of these factors contributed in large, important, and unique ways to predicting persistent symptoms after MTBI. Particularly important preinjury personal factors relating to persistent symptoms were reduced employment, bodily pain, and headaches, and women were more likely than men to have persistent symptoms. Other important preinjury characteristics included poor sleep quality, symptoms of ADHD, a tendency toward neuroticism, and lower resilience. The common injury-related variables, CT abnormalities, and longer duration of PTA were also predictive of worse outcome but much less so than a combination of personal factors. The results of this study, illustrated in fig 3, highlight the value of taking a broad approach and using the biopsychosocial model as a framework to understand and treat PCSs.
Systematic review of self-reported prognosis in adults after mild traumatic brain injury: results of the International Collaboration on Mild Traumatic Brain Injury Prognosis.
The study represents an important step in the development of practical and feasible prognostic models for adults with MTBI, with the ultimate goal of using these models to identify at-risk individuals and refer them for earlier evidence-informed treatment and rehabilitation.
Fig 3Biopsychosocial factors related to symptom reporting at 3 months after MTBI.
Systematic review of self-reported prognosis in adults after mild traumatic brain injury: results of the International Collaboration on Mild Traumatic Brain Injury Prognosis.
Is sex an indicator of prognosis after mild traumatic brain injury: a systematic analysis of the findings of the World Health Organization Collaborating Centre Task Force on Mild Traumatic Brain Injury and the International Collaboration on Mild Traumatic Brain Injury Prognosis.
Methodological issues and research recommendations for mild traumatic brain injury: whe WHO Collaborating Centre Task Force on Mild Traumatic Brain Injury.
Development of the Alcohol Use Disorders Identification Test (AUDIT): WHO collaborative project on early detection of persons with harmful alcohol consumption--II.
Validation of the Long-term Difficulties Inventory (LDI) and the List of Threatening Experiences (LTE) as measures of stress in epidemiological population-based cohort studies.
Predicting long-term global outcome after traumatic brain injury: development of a practical prognostic tool using the Traumatic Brain Injury Model Systems national database.
Pre-injury comorbidities are associated with functional impairment and post-concussive symptoms at 3- and 6-months after mild traumatic brain injury: a TRACK-TBI study.
Stress Management and Resilience Training (SMART) program to decrease stress and enhance resilience among breast cancer survivors: a pilot randomized clinical trial.
Supported by the Liaison Committee between the Central Norway Regional Health Authority and the Norwegian University of Science and Technology (project nos. 90157700 and 46060918).
Disclosures: Grant Iverson, PhD, has been reimbursed by the government, professional scientific bodies, and commercial organizations for discussing or presenting research relating to mild TBI and sport-related concussion at meetings, scientific conferences, and symposiums. He has a clinical and consulting practice in forensic neuropsychology, including expert testimony, involving individuals who have sustained mild TBIs. He has received research funding from several test publishing companies, including ImPACT Applications Inc, CNS Vital Signs, and Psychological Assessment Resources (PAR Inc). He receives royalties from 1 neuropsychological test (WCST-64). He acknowledges unrestricted philanthropic support from ImPACT Applications Inc, the Heinz Family Foundation, the Mooney-Reed Charitable Foundation, and the Spaulding Research Institute. The other authors have nothing to disclose.
31154-0&id=gr1.jpg){kind=link}
31154-0&id=gr2.jpg){kind=link}
31154-0&id=gr3.jpg){kind=link}