Effectiveness of three interventions for secondary prevention of low back pain in the occupational health setting - a randomised controlled trial with a natural course control

This study is a part of a larger longitudinal cohort study which was executed in the occupational health setting. All employees (N = 2480) in a forestry company were invited to respond to a postal survey that included questions on LBP history, LBP intensity, physical impairment (PHI) and pain related fear. Based on their responses, employees (N = 1754, response rate 71%) were assigned into three main categories: “no” symptoms, “some” LBP related symptoms and LBP symptoms that “potentially hamper work”. The cohort includes two embedded RCTs and two respective control groups, based on random sampling, representing the natural course of LBP. The main results of both RCTs have already been published elsewhere [21, 22].

In the present study, we assessed long-term effectiveness of three interventions in comparison to the natural course of LBP alongside with our previous RCT [21] on subjects with low back symptoms that potentially hamper work, later labelled as “moderate” LBP as described in Table 1.

The South Karelian Central Hospital Research Ethics Board approved the study and it was performed according to the Declaration of Helsinki.

The inclusion and exclusion criteria for the study are presented in Table 1.

Eligible employees (n = 193) were included in the sub-cohort “moderate LBP”. At first, we extracted a random sample of 50 employees from the sub-cohort “moderate LBP” and defined them as the control group (Control), representing the natural course of LBP (Fig. 1). The remaining employees (n = 143) were invited to participate in the RCT, of which 17 declined. After an informed consent, 126 participants were randomised into three intervention groups: physical medicine outpatient unit in a hospital (Rehab, n = 43), progressive back exercises in an outpatient clinic (Physio, n = 43), and self-care advice by the OH physician, (Advice, n = 40) (Fig. 1). Employees that were selected in the Control group were followed up for 2 years, similarly as the RCT participants.

There were no differences in the baseline characteristics between the pooled intervention groups and the Control group (Table 2).

The Control group, 50 employees, was extracted from the original, moderate LBP sub-cohort of 193 employees as a random sample by means of a computer program. The remaining 143 employees of the sub-cohort were then invited to participate in the baseline visit for the RCT.

The randomisation scheme for the RCT was prepared by an independent biostatistician using a computer-generated randomisation table. To prevent unequal randomisation of subjects by age and gender in the treatment arms, scripted four-digit identification codes were sorted by gender and age (≤45 years, > 45 years), resulting in four strata. Block randomisation (with blocks of 15) was applied to ensure equal group sizes within each stratum. Based on the randomisation scheme, before the start of the study, a research assistant prepared sealed envelopes containing either a referral to the Rehab group, the Physio group or the Advice group. During the baseline visit for the RCT, after signing the informed consent form, each employee opened a sealed envelope containing the group allocation. The research personnel were not able to identify the workers or their group assignments before randomisation.

Due to the nature of the interventions, the participants and OH professionals could not be blinded of their group assignments. Sickness absence and questionnaire data were gathered and entered into the computer by separate research assistants, ensuring blinded analysis of the data by the researchers.

All participants were free to use all health care services during the study interventions and follow-up, as usual.

The baseline visit, including the randomisation, lasted about 60 min in both active intervention groups, but in the Advice group, the first visit took an extra 20-min time because of the Back Book® information session.

After the randomisation visit, there were four scheduled follow-up visits in all three intervention groups at 3, 6, 12, and 24 months. Employees were instructed to fill out their follow-up questionnaires during the preceding week prior to their respective visit date. Intervention groups were comparable regarding the follow-up intervals, visit activity and the 30 min’ time spent at the follow-up visits. Detailed description of RCT study visits and interventions have been published elsewhere [21].

The company occupational health services unit operated as usual during the study period. Information about the study was provided regularly in the company magazine, bulletin boards, and intranet.

The questionnaire outcomes were evaluated at the 24-month time-point in comparison to the Control group. Baseline measurements originate from the employee survey.

All cause (=total) sickness absence data were obtained from the electronic medical records of the occupational health services. Based on the secondary prevention focus in this study, we accumulated all cause sickness absence days and periods over 48 months in order to evaluate long-term effectiveness and also to reduce the skewness that was related to sickness absence data.

The individual inclusion date serves as a starting point for the sickness absence follow-up until 4 years. In the intervention groups, inclusion date is the date of randomisation and in the Control group, postal date of the employee survey. Sickness absence data are comprehensive and highly reliable because they are based on the administrative payroll system of the employer. There were no missing values in the sickness absence data, either. However, typically sickness absence data is highly skewed, over-dispersed with zeros and includes some extremely high values. Long-term non-LBP-specific sickness absence episodes (over 30 days each), that originate e.g., from severe diseases and sequels of other than low-back specific injury, generally interfere with the analysis but, most likely, are not connected with the effectiveness of the interventions. Because each sickness absence episode in our data also holds a specific ICD-10 diagnosis code [24], we were able to identify and exclude all non-LBP-specific episodes that were longer than 30 days from the original data. In 4 years, excluded SA periods affected about 26–28% of the participants in all four study groups, with no difference in sickness absence periods or days between the groups. In summary, our cumulative sickness absence days and periods in 4 years include all LBP-specific sickness absence days and periods, regardless of their length and all other cause sickness absence that are less than 31 days. The 30-days’ cut-off limit was chosen arbitrarily before conducting any analyses.

Power calculations were performed at the design phase of the study, based on the group difference in LBP intensity (VAS). The standard deviation was expected to be 15 units (mm). Calculations showed that differences in LBP intensity of 10 mm between groups will be detectable with 80% power in two-tailed tests with a significance level of 0.05 for a sample of 40 employees in each group.

All statistical analyses were performed at the employee level, according to the intention-to-treat principle. Baseline characteristics were compared using descriptive statistics, in which the intervention groups were pooled for the comparisons to Control group.

About 29% of the questionnaire follow-up data were missing, but equally in all groups and mostly due to some totally missed follow-up visits. We are not aware of any systematic reasons or motives that would explain the missed visits or non-response in any of the study groups. All participants in this study had an equal opportunity to attend follow-up visits during their working hours and they were also offered several alternatives to change their visit schedule. Some participants sent their questionnaire data to the study personnel if they were not able to attend personally. Hence, based on our best knowledge, missing questionnaire data is missing at random. Missing values in the questionnaire-based outcome variables were imputed with the multiple imputation method [36] of the IBM SPSS Statistical Package version 24.0 for Windows ® (IBM Inc., Chicago, IL, USA). The following items were used as determinants when conducting the missing data imputation: age, gender, marital status, education, smoking, lifetime duration of LBP, self-assessed health status, working status, shift work, physical workload, mental workload, self-assessed work ability, job satisfaction, physical impairment, LBP, pain-related fear, all cause sickness absence 12 months prior to the employee survey and all cause sickness absence over the first year.

The effectiveness of an intervention was primarily estimated with the group difference of all outcome variables in comparison to Control (i.e. Rehab vs. Control; Physio vs. Control, and Advice vs. Control).

As regards the questionnaire variables, primary (PHI, LBP intensity, QoL) and secondary outcome variables (disability, FABQ, FABQ_w, FABQ_ph) were assessed at 24 months after the employee survey. The 95% confidence intervals (CI) for the mean differences were computed using the generalized linear model ANOVA where the respective baseline values were used as covariates (when appropriate). Effect sizes of all outcomes, in all study group comparisons, were estimated with Cohen’s d [37].

For each calendar year, sickness absence data was highly skewed, including both an excess number of zero values and some very high values. However, accumulation of the sickness absence data in 4 years resulted into a smaller proportion of zero values. We tested several linear and non-linear models and hierarchical structures. Previous year sickness absence data, i.e. one-year sickness absence days and periods before the study were used as covariates, respectively. Finally, the best model for the sickness absence distribution in the present study was achieved by the negative binomial model with a logarithmic link function.

By the end of the follow-up, one participant from the Advice group left the study due to personal reasons, but granted permission to use the data. At the end of the two-year follow-up, 6 participants of the Rehab group, 10 participants of the Physio group and 11 participants of the Advice group failed to return their questionnaires, resulting in missing data. One participant from the Physio group deceased 3 months before the final visit. Regarding the intervention groups, 99 participants (Rehab, n = 37; Physio, n = 33 and Advice, n = 29) continued to the final visit, resulting in participation rates 86%, 77% and 73%, respectively. In the Control group, data was available for 31 (62%) participants.

Figure 1 presents the flow of the participants in this trial.

The results are shown in Tables 3, 4 and 5.

No adverse events were reported during the interventions.

In comparison to natural course of LBP, both active interventions decreased physical impairment and increased the quality of life at 2 years. Pain intensity decreased in the Rehab group, but we found no effectiveness in total sickness absence days in any of the group comparisons. Patient information by an occupational physician was not effective in comparison to natural course of LBP. Our study shows that a simple questionnaire is feasible in selecting for whom, and the two active interventions were effective in determining how secondary prevention of disabling LBP may be administered.

We performed a controlled, quasi-experimental study for the secondary prevention of LBP among non-sick-listed employees. The main strengths of this study are at the pragmatic and secondary preventive study design, comprehensive sickness absence data and a real world occupational health setting with the presence of a natural course control group.

The study base (2480 employees) aptly represents the general distribution of Finnish work-force (age, gender, socio-economic class, physical and mental workload). Although all participants reported non-acute, yet moderate level and chronic LBP in the screening phase, they were all primarily able to work. The control group was selected as a random sample from the same cohort of eligible employees with the intervention arms, prior to the randomisation procedure. Based on the inclusion criteria, we suggest good generalisability of the results. The study setting also fits well to our secondary preventive purpose.

The quasi-experimental study design may be considered as a weakness. Although the Control group was a random sample, the intervention groups were randomised separately. One of our main goals was not to interfere the participants in the Control group with any study visits. Moreover, they were sent only one follow-up questionnaire after 2 years.

Both Rehab and Physio were well established in clinical practice for chronic or subacute LBP patients [21] at the time of designing this study. Therefore, we considered these interventions as suitable for the secondary prevention of low back symptoms as well.

All questionnaire-based outcome variables in this study have previously been used in several other intervention studies [18, 19].

Instead of choosing LBP specific sickness absence as an outcome measurement in this study, we chose all cause sickness absence because it is generally considered as a measure of health in the working population when health is understood as a mixture of social, psychological and physiological functioning [38, 39]. Recorded sickness absence data have several advantages: the quality of the data in terms of coverage, accuracy and consistency over time is superior to that achievable via self-reports. Our sickness absence data were skewed and included several outliers, which is typical in the analyses of sickness absence [39, 40].

Multiple imputation is a modern method to impute missing values in longitudinal intervention studies [41, 42]. If only original data were analysed, substantial parts of the data would have been left out from the analyses and the risk of losing essential information would rise. However, our study results (by using multiple imputation) were consistent with the results based on the original data (data not shown).

There are only few comparable studies that have randomised employees with non-acute LBP into active exercise interventions in the OH setting [9, 10]. A recent systematic review on secondary prevention of LBP found low-quality evidence for exercise alone and moderate-quality evidence for exercise with education to lower the risk of future LBP episodes among employees [11]. However, most of the earlier studies were not comparable to our study, because of the differences in patient recruitment strategy, gender, age or profession.

A previous Cochrane review has concluded that LBP specific physical exercise, alone or together with psychosocial intervention or pain management was effective in reducing both clinical symptoms and sickness absence in chronic LBP [19]. Also some recent studies [13, 14, 43, 44] have included patients that were initially more symptomatic than the subjects in our study. Recruitment in these prior studies was based on work absence records or back clinic consultations [8, 18, 45, 46]. These studies were also focused at increasing return-to-work from sickness absence. Because of higher symptom level, different recruitment strategy and a large variety of interventions, comparisons between these studies are complex.

In our previous RCT [21], both active interventions (Rehab and Physio) reduced short-term LBP and improved long-term quality of life (Physio) in comparison to low back specific information that was delivered by an occupational physician (Advice). Rehab also decreased the incidence of new sickness absence periods (all cause), but not earlier than after 3 to 4 years [21]. The results of the present study are consistent with our earlier findings.

Although quality of life did increase in both active intervention arms, results in several other studies have shown mixed effectiveness in QoL [17].

Hay et al. introduced an effective strategy for the early management of LBP in the UK. Participants were distributed into low-, medium- and high-risk LBP subgroups in the primary care [47]. Stratified interventions that were targeted in each subgroup accordingly, decreased short-term physical impairment in the medium- and high-risk groups more than non-stratified best practice. Patients were included in the study, if they consulted their primary care physician [48]. Despite of different recruitment strategy, there are many similarities with the design in comparison to our study. Importantly, subgrouping patients into pre-defined risk groups and delivering active interventions accordingly resulted in consistent, positive results that were comparative with our study results [47].

Although patient information that was delivered by the OH physician was not effective in moderate level low back symptoms, our previous studies have concluded that simple patient information delivered by OH nurse is effective [22, 49] and cost-effective [49] in mild level LBP.

In order to reduce recurrent, sub-acute and chronic LBP at personal, workplace or community level, current evidence already suggests targeted and stratified approach [48] but also the ability to adopt multiple management strategies. Especially, when we are dealing with heterogenic patient groups that are generally met in the primary or OH care, successful management strategy includes exercise interventions, holistic assessment [8], advice and patient information [50], return to work procedures [51, 52] and also ergonomic interventions when needed [53, 54].

Our pragmatic study was performed in the OH centre of a large forestry company in Lappeenranta, Finland. The OH centre was situated near the factory area, similarly as any other primary care unit that is serving its customers. Participants were men and women, between 24 and 56 years old, who reported various physical and mental demands in their work. The participants’ mean total pain level was 59 mm (SD 17 mm; VAS: 0 – 100 mm) and physical impairment was 8 units (SD 5 units; Roland-Morris: 0–18 units) at baseline. Such individuals are at risk of recurrent, progressing LBP [55], representing 11% of the total number of survey respondents. Although their working ability was already reduced, they were still working at the inclusion, suggesting that the target group was suitable for secondary prevention of LBP. Only 12% (n = 17) of the invited employees were excluded or refused to participate in the study. Therefore, we consider that the external validity is good.

Although Roland-Morris Disability Questionnaire is rather insensitive to change when symptom level is low [56], active interventions were still effective in reducing long-term physical impairment. Respective effect sizes were between medium and large (Table 5).

Our main results emphasise the use of simple, low back specific questionnaire as an early screening instrument and the effectiveness of active, biopsychosocial exercise interventions (Rehab and Physio) for employees that are still working but have reported moderate level LBP symptoms. The main effects were consistent, long-lasting and were also supported by the results of the secondary outcomes. Together with our previous RCT [21], we suggest that early management of LBP is feasible for this employee group in the OH setting.