In the present study, we observed that good perception of general health and low pain may predict a higher RTW 2 years after a rehabilitation program. These results confirm those of others for pain [
7,
27‐
29] and self rated health [
30,
31]. Moreover, in accordance with earlier results [
17,
18] changes in pain and perceived health, during hospitalisation, were also predictors of RTW. It means that, among patients with an identical initial pain, those who experience a larger decrease in pain during hospitalisation are more prone to return to work 2 years later. From these results, it can be hypothesized that an intervention on pain and perceived health may favour return to work but this issue remains to be investigated more thoroughly. A complementary variable to health perception is the perceived severity of the lesion. The present study suggests that patients who believe their lesions are very light to moderate (compared to severe or very severe) have increased likelihood to RTW. These results are compatible with those of others [
32].
Among biomedical variables, the demographic factor most commonly found to be associated with chronic disability is older age [
33,
34] or age below 41 years (and gender [
35]). For this reason, these variables were taken as confounders in the present study. The important fraction of male patients (i.e. 81 %) is expected because these patients are mostly blue collar workers victim of work or traffic accidents. It appears from the literature that pain and psychosocial factors are of primary importance to explain long term sick leave [
36‐
38]. Moreover, general health perception and perception of health change/improvement was strongly associated with a duration of sickness absence and with recurrence of new sick leave episodes for the same musculoskeletal complaints [
31,
39]. To our knowledge, this is the first study to identify sub scores of the IES-R questionnaire, an indicator of the post traumatic stress disorder (PTSD) (for a review see [
40]) as predictors of RTW in patients with traumatic injuries. The cut off value of 33, identifies the PTSD with a sensitivity of 0.91 and a specificity of 0.82 [
41]. Interestingly, our population had a high mean total score of the IES-R (i.e. 35 ± 28 [SD]). More precisely the IES avoidance (e.g. effortful avoidance of situations that are reminders of the accident) and hyperarousal (e.g. being irritable, having trouble falling asleep, watchful and on guard) was retained in our final model. Apparently, our results are not in accordance with those of Toien et al. [
16] who reported that IES was not selected in their model prediction of RTW in trauma patients 1 year after the first assessment. However, in the latter study, the total score of the non revised version of the IES was used, which may be less sensitive to identify intrusion and avoidance [
42]. Moreover, total IES-R score may not be appropriate in the prediction of the outcome because each sub-score (i.e. avoidance, hyperarousal and intrusion) may behave differently. For instance, from results of the final minimal model in the present study, patients with hyperarousal problems were apparently more prone to RTW (Fig.
2). These results are surprising because the opposite effect can be expected. However, the non-adjusted simple regression presents an odds ratio of 0.7 for the IES-R hyperarousal score, i.e. a lower likelihood of RTW. To explore this point, models were calculated, containing IES-R hyperarousal plus the confounders and all possible combinations of 1–4 of the remaining final-model predictors. The OR for IES-R hyperarousal switched from negative to positive only in the presence of the IES-R avoidance variable. Thus, IES-R avoidance seems to be an important confounder of IES-R hyperarousal. Interestingly, clinical practice confirms the possibility of patients with hyperarousal symptoms to develop strategies compatible with professional environment. This empirical experience of our medical staff, which is used to treat patients with PTSD, is compatible with the concept that some hyperarousal behaviour may not be deleterious for RTW. Moreover, a study performed on Oklahoma city bombing survivors [
43] has reported that patients with avoidance behaviour received more mental health treatment, had much more interference with activities and were more dissatisfied with work than those with an hyperarousal behaviour. However, we cannot fully explain our finding and this issue has to be investigated further. Finally, it must be kept in mind that a clinical diagnosis of PTSD cannot be formally determined from the response of the IES-R questionnaire. Consequently, further research should be performed in which trained clinicians may systematically screen this pathology in patients. A 1-year follow-up study performed on patients with back and/or neck pain, found strong associations between pain, expectancy, pain-related fear and a belief in an underlying and serious medical problem [
29]. These associations can also be extended to PTSD [
44] and are also compatible with the present results. These results suggest that patients’ pain care should also involve the treatment of fear avoidance [
45], PTSD and some other psychological aspects.
The main limitation of this study is the low response rate of the eligible patients 2 years after hospitalisation i.e. 34 %. In our study, the descriptive statistics of responder, predictors and confounders were relatively stable through time except for the proportion of local native language speakers. It has been reported that patients with local native language were more prone to RTW [
49] and also more likely to respond to questionnaires [
19]. This suggests that data were not missing at random and, therefore, loss to follow-up caused some degree of bias in our OR estimations. Furthermore, because the RTW was obviously only assessed from the responders, our results probably overestimate overall RTW proportion and means and proportions of predictor values. Another limitation of this protocol is that these results cannot be extended to all patients with musculoskeletal injuries. However, because only patients with persistent health problems after an accident are hospitalised in our clinic, the results can be useful for this kind of patients who are following treatment for long periods.
Unfortunately, for technical reasons the present study did not assess the working status continuously but only at 2 years by sending questionnaires for the following reasons: some patients did not know the exact date of return to work and continuous follow up was a too demanding protocol. Moreover, the files of our insurance company did not allow finding the date of RTW i.e. only rents and work ability were available. Our patients often exhibit trauma associated with psychological, social and occupational problems, Consequently, it is not surprising that an important fraction of those patients did not return to work after 2 years.
In conclusion, this study assessed predictors of RTW on patients with musculoskeletal injuries 2 years after a rehabilitation program. Patients who reported a higher general health perception and a low pain at the start of rehabilitation period as well as those who exhibit a large general health improvement and pain decrease during rehabilitation were more prone to return to work. Conversely, individuals who presented avoidance behaviour had a low probability to return to work. Our findings suggest that rehabilitation interventions should also depend on the patient general health perception as well as pain and fear-avoidance beliefs related to the accident.