The recent increase in the number of individuals presenting to emergency departments with neck pain resulting from whiplash injury has led it to be described as a modern epidemic.¹ In the UK alone it has been estimated that 250 000 cases of whiplash injury occur per annum following motor vehicle collisions.² Recovery occurs in the majority of cases: Pearce reports that 55–86% of patients are free from pain after six months³ although symptoms persist in a sizeable minority, and there are currently no proven therapies for alleviating persistent symptoms.^4–6 In economic terms, significant costs are associated with the persistence of symptoms: one study estimated that nearly half of all compensation payments for whiplash injury were made to the 12% of compensation claimants who received payments for more than six months.⁷

A number of studies have demonstrated that initial injury severity is associated with a poor prognosis,^8,9 and others have reported an association between specific symptoms occurring soon after the collision and the future persistence of neck pain.^8,10,11 Previous studies in this field have tended to focus primarily on one area of potential aetiological interest—most frequently, collision factors and/or the nature of the acute injury. Non-recovery has been associated with a number of collision factors including direction, speed, and lack of awareness of the impending collision,^12,13 although there is no consistent evidence to suggest that any one factor is aetiologically more important than any other.

More recently, psychological factors have also been associated with poor prognosis. A number of authors have demonstrated the important role of psychological and psychosocial factors in determining the outcome of whiplash injury including perceived pain interference,¹⁴ general psychological distress¹⁵ and emotional problems.¹⁶ Further, Hendriks et al report that increasing somatisation score is associated with lack of functional recovery at 12 months and is more important aetiologically than are collision specific factors.¹²

Only by studying psychological factors in the context of collision factors, and aspects of the initial injury, can an understanding of the aetiological mechanism of persistent neck pain be gained. From this it may be possible to identify subgroups of patients with neck pain at high risk of persistent symptoms. However, there are few large scale epidemiological studies in this area (Hendriks et al¹² only provide outcome data on 119 individuals). The aims of current study were:

1. to examine the relative contribution of pre-collision health and psychosocial factors, mechanical (that is, collision) factors, and the psychological response to the collision to the development of persistent neck pain

2. to identify those at high risk of persistent symptoms by using information on these factors and the initial post-collision clinical information.

Thus, we aimed to provide a model whereby individuals with the highest risk of a poor outcome can be easily identified in the emergency department.

METHODS

Design

We conducted a prospective cohort study in a group of patients attending emergency departments with neck pain following a motor vehicle collision.

Subjects

Participants were recruited from four emergency departments within Greater Manchester, UK, between 1 February 2002 and 30 June 2003. Eligible participants were aged 17–70 years, attending with neck pain within 24 hours of a motor vehicle collision in which they were a driver or passenger of a car, van, or heavy goods vehicle. We defined neck pain as pain since the collision, lasting one day or longer, in the area identified in fig 1. Those with a fracture or dislocation of the neck, distracting injury or suspected alcohol or drug intoxication were excluded from the study. Additionally, patients who reported an episode of neck pain in the month prior to their collision were also excluded.

The North West Multi-centre Research Ethics Committee gave ethical approval for the study. Approval was also obtained from the appropriate local research ethics committees and NHS trusts.

Baseline data collection

A postal questionnaire was mailed to all consenting patients within 24 hours of their attendance at the emergency department. This collected basic demographic information, plus data on a number of potential risk factors for the development of persistent symptoms.

Pre-collision health and psychosocial factors

Participants were asked to rate their general health prior to the collision using a Likert scale (response: excellent to poor) and to report the number of general practitioner (GP) consultations they had made in the previous 12 months. Lifetime experience of prior neck pain (ever or never) was ascertained using the pre-shaded manikin (fig 1). We also enquired about the presence of any other bodily pain during the month preceding the collision using a blank pain manikin as described elsewhere.¹⁷ Those who were in full or part-time employment were asked to rate eight aspects of their psychosocial work environment based on Karasek’s demand-support-control model of workplace strain¹⁸ using questions previously used in a large occupational cohort study that had demonstrated their construct validity.¹⁹

Collision factors

A number of aspects of the collision were assessed, including the speed and direction of both vehicles. In addition, participants rated the severity of the collision using a 100 mm visual analogue scale (VAS), reported their position in the vehicle, their use of airbag, seatbelt, and headrest, and whether they had been aware of the impending collision.

Psychological response to collision

Psychological response to the collision was assessed using the General Health Questionnaire (GHQ)²⁰ and the Modified Somatic Perceptions Questionnaire (MSPQ).²¹ These were included in the questionnaire with reference to the period “since your accident”.

Initial injury/clinical factors

Participants were asked to rate the severity of their initial injury using a 100 mm VAS. They were also asked to rate any concomitant disability using the Neck Disability Index²² and to report the occurrence of a number of other whiplash associated disorder (WAD) symptoms: dizziness, tinnitus, problems with vision, etc. In addition, participants were asked for consent for examination of their emergency department records. Where consent was given, information was gathered regarding the presence of cervical bony tenderness, limitation of neck movement, and neurological signs and symptoms using a standard form. From these data the severity of the participants’ initial injuries was graded according to the Quebec Task Force WAD classification system.⁴ After we had excluded patients with no neck pain (grade 0) and those with fracture or dislocation of the neck (grade IV), the participants were classified as either grade I (neck pain with no other symptoms), grade II (neck pain with limited range of movement and/or cervical bony tenderness), or grade III (neck pain with neurological signs or symptoms).

Follow up data collection

Participants were followed up by postal questionnaire 1, 3, and 12 months after their emergency department attendance. At each time point, participants were asked whether they had had any pain in the area identified in fig 1, lasting for one day or longer in the week prior to questionnaire completion.

Analysis

Participants who reported neck pain in the post-collision questionnaire and at each follow up point were classified as having persistent neck pain. Thus, analysis was carried out only on those individuals who returned all three follow up questionnaires. The influence of each factor on the risk of persistent neck pain was analysed using Poisson regression with robust estimates of standard error²³ and was adjusted for age and sex. Thus, results are expressed as risk ratios (RR) with 95% confidence intervals (95% CI).

All analyses were conducted in Stata 8.2 (StataCorp, College Station, TX). Firstly, collision factors, pre-collision health, and psychosocial variables were examined for their univariable association with persistent pain (age and sex adjusted) to determine their aetiological influence on neck pain persistence. Secondly, a multivariable model was constructed to determine the relative contributions of the factors previously identified as predictors of persistent neck pain. Thirdly, variables were assessed as to the strength of their overall contribution to the univariable prediction models using Wald tests. Those that predicted persistence with a significance of p⩽0.2 were entered into a forward stepwise Poisson regression model. Variables were included in the final model at p⩽0.10 and excluded at p⩾0.15. Lastly, the risk of persistent symptoms was examined with respect to clinical factors and those relating to initial injury. A second multivariable prediction model was constructed: injury/clinical factors were offered to the previously identified “aetiological” model applying the same inclusion and exclusion criteria. With the exception of age and sex, on inclusion of any injury/clinical factors, variables previously included were allowed to leave the model if they met the appropriate exclusion criterion.

External validity

We anticipated that many individuals attending emergency departments following a motor vehicle collision may be reluctant to participate in epidemiological studies, resulting in a low participation rate. Thus, to assess external validity, we obtained some information from a subgroup of initial non-responders and, where possible, compared these data to those of the study participants. Also, we conducted a weighted analysis, weighting the responders back to the age and sex distribution of the target population.

RESULTS

Prognosis after a whiplash injury

During the study period 1500 eligible patients attended the emergency departments and were invited to participate in the study. Of these, 1342 individuals (89%) agreed to participate although, subsequently, only 765 completed a full baseline questionnaire (adjusted response rate: 51%). The median age of the subjects was 34 years (interquartile range (IQR) 25–44 years), 56% were female, and questionnaires were completed after a median of eight days following the collision (IQR 5–14 days, thus 75% of participants completed their questionnaire within two weeks of their collision).

Follow up response rates were 82%, 77%, and 70% at 1, 3, and 12 months, respectively. In total 480 participants (63%) completed a questionnaire at each time point, of whom 128 (27%) reported persistent neck pain. The prevalence of persistent neck pain increased with increasing age quintile (χ²_trend: 6.1; p = 0.01), but did not differ by sex (male: 26%, female: 26%).

Aetiological predictors of persistent neck pain

Pre-collision health and psychosocial factors

In general, self-reported general health prior to the collision was not associated with an increase in the risk of persistent neck pain (table 1). Similarly, a history of neck pain prior to the collision was not associated with any noteworthy increase in the risk of persistent symptoms. However, those reporting widespread body pain prior to the collision were at increased risk of persistent symptoms (RR 1.8; 95% CI 0.9 to 3.6). There were no consistent patterns to suggest that work related psychosocial factors were associated with an increase in the risk of persistent symptoms.

Mechanical (collision specific) factors

Relatively few collision factors predicted persistent neck pain, and those that did were only associated with moderate increases in risk (table 2). Compared with those who reported a low severity collision (lowest tertile), individuals who rated their collision as of medium or high severity experienced a 60% increase in the risk of persistent neck pain (both tertiles 1.6; 1.1 to 2.3). In contrast, neither the speed, direction, or anticipation of the collision nor the subject’s position in the vehicle were strong or significant influences on outcome. Similarly, lack of headrest use was associated with no increase in the risk of a poor outcome. Being in vehicle other than a car was associated with a doubling in the risk of persistent symptoms (2.0; 1.2 to 3.3), although this only applied to small proportion of the sample (3% of all study participants). This group was also significantly less likely to wear a seatbelt (χ²: 29.2; p<0.001), although adjusting for this had no impact on the risk of persistent neck pain (1.9; 1.1 to 3.4).

Psychological response to the collision

Individuals reporting a high level of general psychological distress (as assessed by the GHQ) experienced a considerable increase in the risk of persistent neck pain (highest v lowest tertile 2.4; 1.5 to 3.7) (table 3). Similarly, individuals with high levels of somatic awareness (MSPQ) experienced a 60% increase in the risk of persistent symptoms (1.6; 1.2 to 2.2).

Multivariable model—aetiological model

Three factors emerged as statistically independent (see table 4, model 1: having a high level of psychological distress following the collision, reporting widespread pain in the month prior to the collision, and being in a vehicle other than a car at the time of the collision). Each of these factors was associated, independently, with approximately a doubling in the risk of persistent neck pain.

Initial injury/clinical factors

A high subjective assessment of initial injury severity (top tertile of 100 mm VAS) was significantly associated with an increase in the risk of persistent neck pain (2.3; 1.6 to 3.5). Likewise, reporting a high score on the Neck Disability Index and a high number of WAD symptoms were significant predictors or poor prognosis (table 5).

Of the participants who were successfully followed up at each time point, 413 individuals (96%) gave consent for us to access their emergency department notes. Initial injury severity, as assessed using the WAD classification, was modestly associated with persistent neck pain: individuals with grade III injuries experienced a 50% increase in the risk of persistent symptoms compared with those with grade I injuries (1.5; 0.7 to 3.4). In particular, cervical bony tenderness (1.5; 1.02 to 2.2) and the presence of neurological signs and symptoms (1.6; 0.8 to 3.1) were predictive of a poor prognosis.

Multivariable model—overall prediction model

The addition of two injury/clinical factors significantly improved the fit of the multivariable model. Although the inclusion of these variables resulted in a decrease in the risk ratios associated with the three factors already in the model, none of these factors met the model exclusion criterion. Thus, after adjusting for age and sex, five variables remained as independent predictors of persistent symptoms (see table 4, model 2). To assess the functionality of the final model, and to simplify it for clinical use, all non-binary variables in the model were dichotomised at the median value, resulting in five binary variables:

1. Pre-collision widespread pain: yes or no

2. Vehicle type: car or non-car

3. Number of WAD symptoms: 0–4 or ⩾5

4. Neck Disability Index: 1–18 or ⩾19

5. Psychological distress (GHQ): 0–5 or ⩾6 (NB: originally scored using the two point scoring system (0-0-1-1) for each item²⁵)

No-one in the study reported a high value of all five variables, but there was a significant trend such that the greater the number of these factors, the greater the risk of persistent symptoms (χ²_trend: 39.5; p<0.001) (fig 2). Prevalence of persistent neck pain was 14% in individuals who, at baseline, reported none of these factors. In contrast, of those reporting four factors, four in every five persons experienced persistent symptoms—equating to more than a fivefold increase in risk in this group.

External validity

Of the 1500 subjects invited to participate, 587 consented but subsequently failed to complete a baseline questionnaire. Non-participants were significantly more likely than participants to be male (58% v 44%, respectively; χ²: 30.9, p<0.001) and were slightly younger (31 years v 33years; Z_{Mann–Whitney}: 4.11, p<0.001). A subgroup of non-participants (n = 329, 56%) was able to be contacted by telephone and provided some data on a limited number of exposures. These individuals were less likely to be involved in a rear shunt collision (2% v 7%; χ²: 16.2; p = 0.01) but, with respect to the other information gathered, had a similar distribution of potential risk exposures as study participants.

Of these initial non-participants, 142 (43%) were successfully re-contacted at each follow up time point and they provided information about neck pain (but were not able to be included in the main analysis because of missing data on a number of key baseline variables). The prevalence of persistent neck pain was lower in this group (18%) than in the participants (26%) (χ²: 3.7; p = 0.55) suggesting that we may have overestimated the prevalence of persistent neck pain in the group as a whole. Assuming that the prevalence of persistent neck pain was 18% in all non-participants, this would only reduce our estimate of overall neck pain persistence from 27% to 21%. However, weighting the analysis back to the age and sex distribution of the original population did not greatly affect the results (see table 4, model 3) and did not alter the study conclusions.

DISCUSSION

To our knowledge, this is the first large scale epidemiological study to examine prospectively the aetiological influence of collision specific individual and psychological factors in the development of persistent neck pain following a motor vehicle collision. We have demonstrated that among people consulting hospital emergency departments with neck pain within 24 hours of a motor vehicle collision, persistent neck pain is common: our results demonstrate that over a quarter still report symptoms 12 months subsequently. In general, few collision specific factors were significantly associated with neck pain prognosis, and the few exceptions (for example self-perceived collision severity) did not remain independently predictive of outcome on multivariable analysis. Rather, a high level of general psychological distress, reporting widespread pain in the month prior to the collision, and the type of vehicle the individual was travelling in at the time of the collision significantly predicted the persistence of symptoms. And finally, we present a model including these three factors, plus post-collision WAD symptoms and disability, that can, in combination, discriminate between individuals with a 14% and an 80% risk of persistent neck pain.

Methodological issues

A number of methodological issues need to be discussed.

1. At baseline, participation was low. Only 51% of eligible invited participants returned a full length baseline questionnaire. Non-participants were less likely to experience persistent neck pain, and were more likely to be young and male. However, a subsequent analysis, weighting the population back to the original distribution of these factors, revealed results similar to the main findings. Other than age and sex, there were few differences between participants and non-participants with respect to other known variables. Non-participants were significantly less likely to be involved in a rear shunt collision, although this was not a significant predictor of persistent neck pain and is not, therefore, likely to have affected our results.

2. Follow up among the 765 baseline participants was high at each time point, varying between 70% and 82%. Overall, 63% of individuals completed a follow up questionnaire at each of the three follow up time points. Those who were lost to follow up did not differ from participants in terms of age, sex, or the prevalence of the predictive factors identified. It is unlikely, therefore, that this loss to follow up introduced any major bias to our findings.

3. Despite five factors remaining in the multivariable model, each associated with broadly similar relative risks, the number of people exposed to each of these factors differed greatly. While psychological distress, post-collision symptoms, and the Neck Disability Index were divided into tertiles for analysis, the number of participants reporting pre-collision widespread pain, or who were not in a car at the time of the collision, was small. Consequently, the population attributable risks associated with these variables are low.

4. A major strength of the current study is its prospective design and, thus, the absence of recall bias—that is, that individuals with and without the outcome of interest (persistent neck pain) differentially recall prior exposures, thus augmenting any associations between exposure and outcome. However, the majority of exposures in the study were collected using a self-completion questionnaire and, therefore, did rely upon subject recall. Estimation of certain factors, such as the speed of the other vehicle, is difficult. It is possible that misclassification of these exposures may have introduced a bias towards the null—a concern when, in the current study, these factors demonstrate little effect. However, these results are consistent with other recent studies that demonstrate little or no prognostic value of collision specific factors. Also, many of the self-report measures used in the current study were straightforward and unlikely to be influenced by subjectivity (for example, vehicle type, direction of collision, position in vehicle) and many of these were only weakly associated with persistent neck pain. Even self-reported factors which one might hypothesise would be strongly influenced by post-collision symptoms and thus biased away from the null (for example, severity of collision) were not, on multivariable analysis, independent predictors of outcome.

5. By focusing on the predictors of persistent neck pain, one area of major concern is that we may have failed to identify predictors of poor outcome in the shorter term. All participants, by definition, reported neck pain at baseline, the majority of whom still reported symptoms at one and three months (70% and 59%, respectively), of which only a subgroup reported persistent symptoms. Additional analysis (data not shown) revealed that although factors relating to initial injury severity were important predictors of symptoms at one month, the greatest predictors of neck pain at three months were similar to the main analysis—that is, psychological and psychosocial rather than collision factors.

Post-collision neck pain

One of the major criticisms of much of the literature on neck pain following a motor vehicle collision is the failure to adequately control for compensation—the thesis being that those claiming compensation may be more likely to report persistent neck pain than other individuals. In the current study, the majority of participants (89%) had sought compensation at some point over the 12 months since their motor vehicle collision. Those reporting persistent symptoms were significantly more likely to have claimed compensation than other individuals (95% v 87%; χ²: 7.0; p = 0.008) and, among claimants, those with persistent symptoms were less likely to report the resolution of this claim (29% v 57%; χ²: 25.8, p<0.001).

In the current study, it is not possible to distinguish whether claiming compensation is antecedent to, or a consequence of, persistent symptoms. Of major concern, if the former, is that some of the risk factors we have identified (in particular, psychological distress) are not risk factors for persistent symptoms per se; rather, they are risk factors for claiming compensation. However, additional analysis (not shown) revealed that a high level of general psychological distress was associated with a significant increase in the risk of persistent symptoms even among non-claimants.

Previous studies have shown that a third of patients report disabling neck pain one year after the collision,^26,27 and others have reported that as many as 79% of patients still report symptoms at 12 months.¹⁴ In comparison with these studies, our estimate of the occurrence of persistent neck pain is low: 21%, after adjusting for non-participation. However, whereas previous authors have examined neck pain at 12 months we defined our outcome as neck pain at each of four time points: immediately after the collision, and at 1, 3, and 12 months after hospital attendance. Examining the 12 month data alone, our results show that 192 individuals (36%) reported neck pain at this time point, although not all (n = 64) had reported neck pain at each time point since their motor vehicle collision and, thus, are not classed as experiencing persistent symptoms. Additional analysis revealed that the risk factors for neck pain at 12 months (all subjects) were similar to those for persistent pain (data not shown). Further, a comparison of neck pain severity at presentation to the emergency department suggests that the current study population was less severely injured than in previous studies. Approximately three quarters of the current participants were classified as having WAD grade I injuries, whereas others have reported a predominance of grade II and III injuries (58–95%)^14,26,27; this may also help explain the higher occurrence of persistent symptoms in these studies.

Much of the previous literature in this area has focused on collision factors. Ryan et al demonstrated that awareness of the impending collision was the most important predictor of symptom persistence.¹³ In contrast, the current study shows no association between anticipation of the collision and persistent neck pain. These results are similar to those of Hendriks et al who demonstrate no relationship with long term prognosis,¹² although these authors do show a strong association between collision (un)awareness and outcome at three months. Other studies have reported a number of moderately sized associations across a range of collision specific factors, with no one factor emerging as particularly important.^26–29 Harder et al found that collision and individual factors were independently but weakly associated with a longer time to claim closure among a group of whiplash compensation claimants.²⁸ In particular, these authors demonstrated that being in a vehicle other than a car was associated with a poor prognosis and suggested that the lack of safety features in other non-car vehicles (or lack of use of these safety features) might account for this association. In the current study, people in vehicles other than cars were less likely to wear a seatbelt than car occupants (80% and 98%. respectively; χ²: 29.2, p<0.001). However, adjusting for seatbelt use did not explain the increase in risk associated with vehicle type.

Prognostic ability of the WAD classification system

A number of studies have examined the prognostic ability of the WAD classification system.^8,9,26 In the current study population, the WAD grade was only moderately predictive of persistent neck pain in univariable analysis. Analysis of the individual symptoms on which the classification system is based found cervical bony tenderness to be more important than range of movement in predicting outcome. This is in contrast to the findings of Hartling et al.⁹ These authors suggested that predictive ability of the classification system was improved after the creation of the grade II subgroup, which presented with a limited range of movement in the neck. Again, however, this may be a reflection of the fact that our subjects were less severely injured than the participants of previous studies.

Role of psychological factors

Few authors have examined the role of psychological factors in the prognosis of whiplash injury. In a small sample of collision victims presenting with spinal pain or stiffness (n = 32), Richter et al demonstrated that psychological factors, in particular social functioning and capability, were more important than collision severity in predicting symptom duration and severity.²⁹ Similarly, Gargan et al found that a high level of psychological distress three months, but not immediately, after the collision was predictive of intrusive or disabling symptoms at two years³⁰ and Olsson et al demonstrated that an active coping style, measured within one month of collision, predicted recovery from neck pain at one year.¹⁴

SUMMARY

In summary, a number of authors have highlighted the importance of psychological factors in the prognosis or whiplash injury.^12,15,29 The current study confirms these findings and demonstrates that, as opposed to factors relating specifically to the collision, among individuals attending hospital emergency departments with neck pain within 24 hours of a motor vehicle collision, the psychological response to the collision, the occurrence of pre-collision widespread pain, the type of vehicle and a number of factors relating to initial injury best predict the persistence of symptoms. These results lend further support to the bio-psychosocial model of whiplash disorder, as proposed by Ferrari.³¹ Further, we present a model for use in clinical practice, of five factors easily recorded in the emergency department, that in combination are able to discriminate between patients with a 14% and 80% risk of persistent symptoms. It is hoped that by being able to identify such subgroups of patients immediately after a collision it will be possible to improve the management of the condition through the targeting of treatments to those who experience the highest risk of a poor outcome.