The incidence of traumatic brain injury is higher than any other neurological diagnosis.1 Over 80% of all traumatic brain injuries are considered mild (mild traumatic brain injury (MTBI)), as the impact to the head results in only a relatively brief period of disturbed consciousness and amnesia. Mortality in these patients is low, and neurosurgical interventions are rarely needed (<1%). Nevertheless, MTBI is recognised as an important public health concern, as an estimated 5–15% of all patients suffer persistent symptoms and functional impairments for months to years after injury.2^–4 Given the high incidence of MTBI, and the good recovery in most patients, routine follow-up may not be feasible or needed. Unfortunately, the scientific foundations for reliable early identification of patients who are likely to recover well are weak.

In 2006, a review of the prognostic models for TBI identified only a few high quality studies concerning the prediction of MTBI outcome.5 These studies were mainly directed at calculating the risk of post acute complications, rather than at long term outcomes such as self-perceived symptoms or return to work. Although hundreds of studies report clinical risk factors for poor outcome, most of these are epidemiological or correlational in nature, using small or selected samples, considering only a limited set of predictors. Very few studies have addressed the validity of their models, or evaluated how to use these risk factors to guide clinical decision making (eg, regarding the necessity for outpatient follow-up). Furthermore, there are other important problems which hamper MTBI outcome studies, such as the lack of specificity of postconcussional symptoms (PCS), and the lack of generally accepted criteria for diagnosing pathological PCS.2 3 As a result, many studies use their own criteria, but often fail to discuss the rationale for their choices or the validity of their methods.

Although clinically usable prediction models are scarce, the existing literature does provide the ingredients for a potentially powerful prediction model. For example, it is recognised that traditional head injury severity indices have limited power to predict long term functioning.6 7 Rather, other injury characteristics (eg, early symptoms,8 9 presence of extracranial injuries10 11) as well as pre- and post-injury physical functioning (eg, pain, fatigue12 13) and psychological status (eg, emotional distress such as depressed mood,14 15 anxiety (especially acute post traumatic stress)16) or personality17 are considered essential for understanding and predicting individual outcome patterns.

In this prospective cohort study, our aim was to develop and internally validate a prediction rule for favourable recovery 6 months after sustaining MTBI, based on easily obtainable pre-, peri- and post-injury variables. For this purpose, we derived one rule to identify patients who reported the absence of PCS and a second for the prediction of full return to work (RTW). To facilitate clinical application, we derived a score chart from the models.

METHODS

Patients and procedure

The study was approved by the ethics committee of Radboud University Nijmegen Medical Centre, and all patients gave written informed consent. All consecutive patients with MTBI admitted to the emergency department of Radboud University Nijmegen Medical Centre, a level I trauma centre, between October 2004 and August 2006, were eligible to participate in the study if they were between 18 and 60 years of age, able to speak and write in Dutch and did not suffer from premorbid mental retardation or dementia. As soon as possible after admission to the emergency department, patients were informed about the study and asked to complete a questionnaire; when patients were admitted to hospital, they were visited by a researcher during their stay, otherwise they were sent a letter within days after injury containing information about the study and a questionnaire booklet with a request to return the completed forms together with written informed consent. Questionnaires that were completed more than 6 weeks after the injury were not included in the study. Consenting patients were sent a follow-up questionnaire 6 months later.

Definition of MTBI

In accordance with the criteria of the European Federation of Neurological Societies, MTBI was defined as a history of impact to the head with or without loss of consciousness (LOC) for ⩽30 min, and with or without post-traumatic amnesia (PTA), and a hospital admission Glasgow Coma Score (GCS) of 13–15.18

Outcomes

1. Postconcussional symptoms (PCS). PCS were measured using the Rivermead Post-Concussion Questionnaire (RPQ), a checklist assessing 16 common symptoms on a 5 point Likert Scale. Patients were asked to rate how problematic, if at all, each symptom was experienced compared with the situation before they sustained their head injury.19 Although the RPQ is often used to measure PCS, a gold standard for classification of “mild” versus “severe” PCS is lacking. We defined favourable outcome as a score of 0 (no problem), 1 (not a problem anymore) or 2 (mild problem, but not interfering with daily activities) on at least 13 of the 16 symptoms. To explore the sensitivity of this criterion for MTBI symptomatology, we applied the criterion to the 287 non-brain injured patients with a wrist or ankle distortion that served as controls in a previous report by our group.10 The results showed that 94% of these patients would meet this criterion for favourable outcome.

2. Return to work (RTW). At follow-up, patients were asked to state their current employment status, and indicate whether they experienced negative changes in their work situation because of the trauma. Patients were classified as having full RTW when they were not on sick leave at the time of follow-up, or reported no change in working status to partial or lower level employment because of the accident.

Prognostic factors

Clinical data were registered by the consulting resident of neurology in the emergency department, and thereafter collected by a research nurse and registered on prespecified forms. All data were manually entered into the electronic Radboud University Brain Injury Cohort Study (RUBICS) databank.

Pre-injury

In addition to age and gender, the following prognostic factors were included.

• Education: patients were categorised into three categories: low (10 years of formal education or less), middle (11–14 years of formal education) and high (14 years of formal education or more) level of education.

• Premorbid emotional problems: a self-reported history of treatment by a psychologist, social worker or psychiatrist, or current use of psychotropic medication, or both.

• Physical comorbidities: self-reported presence of one of the items listed in the questionnaire (asthma, chronic bronchitis, chronic obstructive pulmonary disease, severe cardiac disease, cardiac arrest, epilepsy, diabetes, chronic back problems, spinal disk herniation, osteoarthritis, rheumatoid arthritis, malignancies, cancer) or in the case of the presence of another health problem that can be expected to have great negative impact on daily functioning.

• Prior head injury: patients were asked if they had ever suffered an injury to the head or brain before and, if so, what and when. As even very minor cases of head injury may result in chronic symptoms, any report of “concussion” or “contusion” was considered to be relevant, as were non-traumatic causes of brain damage such as a brain tumour.

Peri-injury

• GCS: admission GCS scores were assessed to indicate a patient’s level of consciousness. In case of intubation or sedation, the pre-intubation/sedation GCS registered closest in time to admission was considered.

• LOC: the presence and duration of LOC was based on reports of witnesses of the accident or ambulance personnel.

• PTA duration: the presence and resolution of PTA was assessed by the consulting resident of neurology in the emergency department, using a standardised procedure which included the testing of anterograde memory (three word recall test) and orientation, and a behavioural screen of confusion or agitation. If PTA had already ended on admission, an estimate about the duration was made based on all information available (eg, the patient’s first memory after the accident, eyewitness reports). When PTA persisted in the emergency department, patients were admitted to the hospital for further monitoring. Duration of PTA was classified into three clinically meaningful categories: (1) no PTA, (2) 1–30 min and (3) >30 min.

• CT characteristics: a brain CT was performed according to international guidelines and scored using a predefined format.18 A CT was defined as abnormal if showing signs of contusion, oedema, subdural haematoma, epidural haematoma or subarachnoid haemorrhage.

• Early symptoms: based on previous reports, we included dizziness, nausea/vomiting and headache reported in the emergency department as predictors in the model.89

• Additional extracranial injuries: extracranial injuries were considered present if, in addition to an MTBI, patients had a score of 2 or more in one of the body regions of the Abbreviated Injury Score/Injury Severity Score.20

Early post-injury

• Postconcussional symptoms: severity of PCS was measured with the RPQ,21 and our criterion of “severe PCS”, as described above (see Outcomes), was used.

• Post-traumatic stress: the 15 item Impact of Events Scale measures intrusive symptoms (eg, intrusive thoughts) and avoidance symptoms (eg, numbing of responsiveness), and combined provide a total subjective stress score. In accordance with the guidelines, scores above 26 were classified as severe.22

• Fatigue: self-perceived fatigue severity was measured with the 4 item Abbreviated Fatigue Questionnaire. A cut-off value of 20 points was used to identify severe fatigue.2324

• Pain: patients rated the severity of current pain on a 4 point Likert scale (range from 0 (no pain) to 3 (severe pain)) on five body regions (head/skull, neck, arms/shoulders, chest/abdomen/back, pelvis/legs) and a total pain score was calculated by adding the scores on the five items. High levels of pain were defined as a total score >4. The questionnaire and scoring are provided in the appendix.

• Self-efficacy: the 10 item Generalised Self-Efficacy Scale was used to assess optimistic self-beliefs to cope with a variety of difficult demands in life. As suggested in the manual, a median split was applied.25

Most candidate variables were chosen from previous research results. The predictors fatigue, pain and self-efficacy have, to our knowledge, not been examined prospectively, but are nevertheless included, as they are repeatedly reported as potentially important in cross sectional studies.14 18 26 Questions on pre- and post-morbid functioning were included in the early questionnaire.

Statistics

The group with follow-up information and the group without follow-up were compared using the two sample t test in the case of continuous measures and χ2 tests for frequency data. A p value <0.05 was considered significant. To enhance clinical interpretability and reproducibility of the models, continuous measures were either dichotomised or split into categories based on published cut-off scores or, as described in the Methods section, in the case of the RPQ and the Brief Pain Questionnaire on criteria we developed ourselves. Logistic regression analysis with backward selection was used to develop the prognostic models. Variables that had an association with outcome (p value ⩽0.10) were included in the backward selection in the multivariable logistic regression analyses. As missing data were very scarce (<0.5% of all required values), missing values were not imputed. The area under the receiver operating characteristic curve (AUC) was calculated to assess the performance of the models in terms of accuracy of correct predictions. Prediction models usually provide too extreme estimates when no correction is applied in the development phase. We therefore used internal cross validation (bootstrap resampling) to correct for this optimism and to provide more realistic estimates.27. For the score charts, we simplified the prediction models by replacing the coefficients resulting from the logistic regressions by whole numbers that were proportional to those coefficients. This has no effect on the performance of the prediction rule, as only the ratios between the coefficients matter. Except for the bootstrap procedure, which was done with SAS 8.2, all statistical analyses were carried out using SPSS for Windows V.12.0.

RESULTS

Study population and follow-up

Of 1003 patients who attended the emergency department with an MTBI during the study period, 539 met the inclusion criteria. Of these, 452 patients were sent the early questionnaire; 87 were missed mostly for logistic reasons. Complete questionnaires were returned by 280 patients (62%). Only these patients received the 6 month outcome questionnaire, which was returned by 201 patients (72%). Compared with the total cohort, the final sample contained less men (n = 123; 62%) than the sample in which no follow-up data were available (n = 252; 75%, p = 0.009), and patients were somewhat younger (final sample 35.6 (SD 12.3) years vs rest of the cohort 38.2 (SD 12.5) years; p = 0.001). There were no differences between the groups for admission GCS, MTBI category, type of injury, presence of LOC and PTA, whether a CT of the brain was made and whether patients were admitted to hospital. Table 1 lists the baseline characteristics. The early questionnaire was completed, on average, 9 days after injury (SD 7.1, range 0–37 days), and the follow-up questionnaire on average 6.5 months after injury (SD 1.0, range 5.5–10.0 months). Low PCS at 6 months were reported by 152 patients (76%), and 153 (76%) patients reported full RTW. A total of 128 patients (64%) reported both the absence of PCS and full RTW.

Prognostic model

The univariate associations of the determinants with low PCS and full RTW at 6 months are presented in table 1. Dizziness was not included in the model as only three patients reported this complaint. Table 2 presents the variables for the final prediction models after backward stepwise analysis.

Model 1: Prediction of low PCS

The absence of comorbid physical problems, low levels of early PCS and low levels of early post-traumatic stress most strongly predicted low PCS. The odds of low PCS at 6 months in patients without premorbid physical comorbidities were 3.5 times the odds of when such comorbidities were present. Similarly, there was a 5.5-fold greater odds of low PCS at follow-up in patients who reported low PCS early after injury than in those patients who did experience severe symptoms at this stage. Lastly, there was a 10-fold greater odds of favourable outcome for patients without early post-traumatic stress compared with patients with severe early post-traumatic stress. The AUC for this model was 0.82. After a correction for optimism based on the bootstrap samples was made, the discriminatory ability of the model was decreased but still reasonable (AUC = 0.73).

Model 2: Prediction of full RTW

More than 11 years of education, absence of nausea or vomiting on admission, absence of additional extracranial injuries and no severe pain early after injury were associated with a greater chance of full RTW. The odds of full RTW at 6 months in patients with more than 11 years of formal education were 6.4 times the odds of favourable outcome in patients with less education. Similarly, when patients did not report nausea and did not vomit on admission, the odds of full RTW were 5.1 times the odds for favourable outcome compared with patients who experienced these problems. Thirdly, the odds for RTW were 3.4 times greater in patients with isolated MTBI than in patients that also suffered additional extracranial injuries. Lastly, the absence of pain shortly after injury was associated with a 2.3 greater odds of favourable outcome than when severe pain was present at this stage. The AUC was 0.79. Again, the discriminatory ability of the model decreased but was still fair (AUC = 0.70) after correction for optimism based on the bootstrap samples.

Considering the loss of power associated with the use of binary or continuous independent variables in regression analyses, we also explored which models would result when the predictor variables without published cut-off scores (pain, RPQ, general self-efficacy) were included as continuous variables. The resulting models were similar to those obtained in the original analysis. In tables 3 and 4, score charts with associated chances for recovery are provided for both models. For example, a patient without severe PCS and no post-traumatic stress early after injury (76% of the whole current sample) has a 90% chance of being free of debilitating PCS at 6 months. To illustrate the impact of internal validation on the power of a regression model, the predicted probabilities both before and after bootstrap correction for optimism are reported.

DISCUSSION

We have developed two models for the prediction of favourable 6 month recovery in a prospective unselected sample of patients with MTBI consecutively admitted to the emergency department of a level 1 trauma hospital. To enhance the reproducibility of these models in future studies, we internally validated the models by bootstrapping. The models, based on easily obtainable pre- peri- and early post-injury factors, identified a group of patients with 90% probability of low PCS or full RTW at 6 months. Calculation of a score chart enabled easy identification of risk scores and associated probabilities for favourable outcome.

In line with most MTBI outcome studies, the majority of patients in our study had recovered by 6 months after injury.4 28 However, one-third of the sample reported persisting PCS, incomplete return to work, or both. Although these numbers correspond with previous studies in unselected samples (including patients with extracranial injuries or psychiatric comorbidities), the prevalence of suboptimal outcome in the general MTBI population is expected to be lower, as the willingness to participate in research is thought to be less in those who fully recover.4 7 21 29^–32

The results of our study illustrate that in order to enable prediction of outcome after MTBI, factors unrelated to the head injury are of major importance. Regarding pre-injury characteristics, the chances of good outcome significantly increased with higher levels of education, especially in relation to return to work, which confirms earlier studies.33 34 Possibly, higher level jobs generally have better conditions for work resumption, such as greater decision making latitude and lower physical demands. Alternatively, more highly educated patients may have more adaptive coping skills. In turn, the absence of physical comorbidities predicted the absence of PCS, but not return to work. Possibly, people who are in suboptimal physical shape before the injury have less reserve to overcome the additional strain of an MTBI. Alternatively, symptoms that relate primarily to the comorbidity may falsely be attributed to the head injury.35 Factors that have incidentally found to predict outcome (although at different times since injury) such as age, gender or a history of emotional problems did not predict outcome in the present study.9 17 36 37

In accordance with many other studies, traditional injury characteristics such as LOC and PTA duration could not predict long term PCS, and we could not replicate the finding of others that acute symptoms had strong predictive value for the development of PCS.8 9 Absence of additional systemic injuries predicted full return to work. This confirms our earlier findings that 6 months is too early to determine final outcome, because many patients with multisystem injuries are still in the process of rehabilitation.10 Lastly, in the current study, the presence of nausea or vomiting in the emergency department was significantly related to incomplete RTW, rather than to PCS.8 9 This finding is less easily interpretable, and the mechanisms remain poorly understood.

Regarding early post-injury factors, the absence of early PCS and low levels of post-traumatic stress strongly predicted the absence of debilitating symptoms at 6 months.12 38 These results support the importance of considering emotional well being early after injury for long term outcome.14 21 39 Post-traumatic stress is closely related to other forms of emotional distress, such as depression and anxiety, that are known to negatively impact on outcome.40 In addition, it has been shown that the majority of patients meeting the criteria for acute stress disorder early after injury will have post-traumatic stress disorder after 2 years.39 Early interventions such as cognitive behaviour therapy may help to avoid this. Our findings also add to the literature by suggesting that less perceived competence to deal with difficult and unforeseen circumstances, and early levels of severe fatigue, are associated with greater likelihood of severe PCS at 6 months. As these factors were significant in the univariate, but not in the multivariate analysis, they may not be of key importance to MTBI outcome, but are still worth considering.

Several limitations of the study should be noted. Firstly, our sample size was relatively small for prediction modelling and, as only 37% of the whole cohort was included in the study, careful consideration of potential recruitment bias is needed. Response analyses showed that participants resembled a general emergency department admitted MTBI population; no differences were found on any of the injury characteristics, compared with non-participating patients, and age and gender did not contribute to outcome. Secondly, because of the lack of generally accepted criteria for diagnosing pathological PCS, we developed our own criterion for “low” and “high” PCS. Explorative analyses in control patients with mild orthopaedic injuries suggested that this criterion was sensitive for MTBI symptomatology. However, we should emphasise that this criterion, as in many other MTBI studies, is somewhat subjective and requires further validation. Thirdly, although we performed internal validation of the models by using bootstrap resampling, external validation remains necessary to confirm the predictive value of our models in future patients and other settings. Furthermore, the prediction models could strongly predict good outcome, even after correction for optimism, but were less favourable for the prediction of poor outcome. These results are in line with most previous studies that did not succeed in finding strong and reliable predictors for suboptimal outcome.5 28 In addition, the sensitivity and specificity of the models decreased substantially after internal validation, which may be due to the relatively small sample size, the many potential predictors, the low frequency of important predictors and the variance within the patient sample. Lastly, although we addressed many factors, there are other potentially relevant variables that we did not include, such as litigation or early cognitive testing.31 41 In addition, there are other methods available (eg, the Acute Stress Disorder Interview) to diagnose pathological stress in the acute phase that may be more sensitive and valid early after injury than the IES which we used in the present study.39 40

Apart from contributing to a greater understanding of factors influencing MTBI outcome, our findings may have important clinical implications. We showed that outcome could not be predicted based solely on pre- and peri-injury characteristics, but also required information regarding a patient’s emotional and physical functioning early after injury. As this information is relatively easy to obtain, an outpatient visit may not be necessary. Rather, consultation by phone or even through internet based questionnaires may suffice. The prediction models may then guide early clinical decision making by identifying patients in whom further follow-up is likely to be unnecessary. Interpolation of our own data to the general emergency department admitted MTBI population would suggest, with a high level of certainty, that about 75% of all patients are expected to recover well. Outpatient monitoring may be reserved for the remaining patients who have a higher risk of developing long term symptoms or occupational problems. This would potentially represent a substantial overall saving in cost and time. In addition, the score charts may help to easily inform patients more accurately regarding their prospects for recovery.

In conclusion, the present study supports the feasibility of early identification of patients with MTBI who are likely to have good 6 month recovery, on the basis of only a few factors. Patients who will recover well do not seem to simply have suffered milder head injuries but rather the results of our study illustrate that to enable prediction of outcome after MTBI, factors unrelated to the head injury are of major importance.