Cost-effectiveness of providing patients with information on managing mild low-back symptoms in an occupational health setting

The original study design was a longitudinal cohort study with two embedded RCTs and a control group that underwent no intervention (natural course of LBP). The detailed study design and results of both RCTs have been published elsewhere [15, 23].

All employees (N = 2480) in a forestry company were invited to respond to a postal survey on LBP and related PHI during September 2001. On the basis of their responses (N = 1754, response rate 71 %), employees were allocated into three main categories: “no” low back (LB) symptoms, “some” LB symptoms i.e. mild LBP (RCT1), and LB symptoms that “potentially hamper work” i.e. moderate LBP (RCT2). The main results of both RCTs have already been published. RCT2 showed that two multidisciplinary and active interventions reduced LBP, sickness absence and physical impairment among employees who were fit to work but reported moderate level LBP [23]. RCT1 showed that among employees who reported mild LBP (but were still able to work), booklet based patient information was as effective (in terms of clinical effectiveness) as additional face-to-face verbal advice with the booklet [15].

In this paper, we will focus on employees that reported some LBP symptoms, defined as mild LBP [24]. Alongside our previous RCT1 [15], we analysed the clinical effectiveness and cost-effectiveness of two patient information strategies in comparison with the natural course of LBP (no intervention), two years after the employee survey. Figure 1 shows the flow chart of the study participants at different stages of the intervention trial.

Throughout the study, information about the study procedures and the management of LBP were shared in collaboration with all stakeholders of the company (e.g. employer, employees, study personnel, trade union representatives and workers’ compensation board).

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

Employees were eligible for the study, if they reported (in their responses to the postal questionnaire) LBP intensity between 10 and 34 mm on a 100-mm Visual Analogue Scale (VAS) during the past week, were aged under 57, and fulfilled at least one of the following criteria:

Exclusion criteria were: retirement within the time span of the follow-up, pregnancy, presence of acute nerve root compression symptoms, suspicion of malignant tumours, recent spine fracture, severe osteoporosis or any other disease that prevented participation in the follow-up.

Sickness absence at the inclusion phase of the study was neither an inclusion nor an exclusion criterion for the study. Initially, employees were able to work regardless of their previous or present symptoms (whether LB related or other).

Eligible employees (n = 312) were defined as a mild LBP cohort. Since there is no gold standard for distinguishing mild LBP from moderate symptoms, the pain limits were chosen arbitrarily. Pain categorisation follows, for example, the recommendations of Boonstra et al. [24], who has categorised mild pain as VAS ≤34 mm, moderate pain as between 35 and 74 mm, and severe pain as ≥ 75 mm.

Prior to the invitation to participate in the intervention, a random sample of employees (83) was extracted from the cohort in order to form a natural course (NC) control group. The remaining employees (n = 229) were invited to visit an OH nurse, who was specially trained in LBP problems and the main study procedures. A total of 182 employees were randomised (47 declined to participate) into two intervention arms, Booklet (Back Book) and Combined (Back Book and Advice), but soon after randomisation, one person was excluded from the Combined group due to retirement. Therefore, finally 181 participants (Booklet, n = 92 and Combined, n = 89) were included in the intervention groups. Of the 83 employees that were eligible for NC, 51 responded to the postal questionnaire, resulting in missing data for 32 employees.

Table 1 shows the distribution of some basic characteristics (employee survey data) in the study groups. Throughout the study, all participants were able to use HC services as usual, either at OH services or other available HC units.

The NC group was extracted from the mild LBP cohort as a random sample by means of a computer program. Remaining employees were invited to the first study visit.

The randomisation scheme for the trial was prepared by an independent biostatistician before the study was started by using a computer-generated randomisation table with a block size of eight. A research assistant prepared the sealed envelopes that included the group assignments, referral either to the Booklet or the Combined group.

During the first study visit, OH nurse informed the employee about the study, collected the informed consent, performed a clinical examination and explained the findings to the employee. After that, she opened the sealed envelope that included the assignment for either of the two interventions, according to the randomisation scheme. More details of the original RCT are explained elsewhere [15]. In order to ensure blinded analysis, all study data were entered into the data file by persons who were not related to the study.

The company OH services unit operated as usual during the study period, and information about the study was provided regularly in the personnel magazine of the company and intranet.

At the end of the first study visit, the OH nurse gave the Back Book (booklet) to all participants and presented additional face-to-face oral review of the booklet into those participants who were randomised into the Combined arm.

The randomisation visit lasted about 60 min in the Booklet group, whereas in the Combined group, the additional review of the booklet, face-to-face with the employee required 20 min more.

Follow-up visit intervals were similar in both intervention groups, scheduled at 3, 6, 12, and 24 months from the first visit. During scheduled visits, OH nurse checked the returned questionnaire responses, performed a simple clinical balance test, and answered questions that may have arisen among the participants. Generally, the follow-up questionnaires had been filled in by the employees during the week prior to the visit date. Both intervention groups were comparable as regards the follow-up intervals, visit activity and the time spent (30 min) at the follow-up visits.

NC group received a postal questionnaire, similar to those addressed to the intervention groups and scheduled at 24 months from the start of the whole study. For this paper, we have analysed the 24-month questionnaire data, because it is comparable between the intervention groups and NC.

About 24 months after the employee survey (i.e. baseline), participants in all study groups were asked to report their HC utilisation during the previous 12 months, i.e. 13–24 months after the baseline. The questionnaire included the following items: the number of visits to a physician, nurse, physiotherapist or other HC professional; each representing a unit of HC utilisation. These items were included in the following HC categories: OH, public (primary) HC, private HC, hospital outpatient clinics, and hospital inpatient care and rehabilitation clinics. The number of radiological procedures during the previous 12 months and the visits that relate to alternative or complimentary HC (acupuncture, massage, chiropractor etc.) were also included. Our HC usage data shows separately the visits in the OH, public HC and private sector, but they all may be considered as primary care visits. Service-wise the contents of care are comparable. Because their unit costs vary, we have presented them separately. The unit costs were obtained from the national working paper of the Finnish Ministry of Social Affairs and Health [25], expressed in euros, and converted to the 2004 level (the final follow-up visit was in 2004). Additional table shows the use of HC resources over 12 months, scheduled at 13–24 months from the baseline (see Additional file 1).

We have focused on the direct HC costs and considered SA as the primary outcome of the CE analysis. Travelling costs and productivity losses (i.e. employee not working in his/her working time) were not included in the costs. All study participants were working at the same industrial complex area. Intervention cost was evaluated with the extra time consumed for the verbal patient information in the OH nurse visit (20€/person). In proportion to the 12-month HC costs, we also collected all cause SA days from 12 months into our cost-effectiveness analysis (CEA). Cost-effectiveness (CE) was evaluated from the health care perspective and the HC usage as well as the SA data in CEA covered the same time period (12 months, timeline of 13–24 months from baseline).

Baseline measurements originate from the employee survey. The follow-up measurements were performed 24 months after the baseline. Time interval is the same in all study groups. Clinical effectiveness was evaluated with the following variables: physical impairment (PHI), intensity of LBP and accumulated, all cause sickness absence (SA). PHI was measured with the Roland-Morris Disability Questionnaire 18-item scale [26, 27], and LBP by the Visual Analogue Scale (VAS; 0–100 mm) [28]. SA data were gathered at one-year intervals up to two years after the individual randomisation date of the participants in the intervention groups. As regards the NC group, SA count started from the postal date of the employee survey. SA data is comprehensive and highly reliable because it is based on the administrative payroll system of the employer. There were no missing values in the SA data, either. Since all SA episodes in the OH service’s administrative registration system included ICD-10 [29] classification numbers, we were able to distinguish LB specific SA from the other cause SA episodes. Although the total (=all cause) SA days were included in the final analyses, each SA episode that was not LB-specific and at the same time longer than 30 days was omitted from the analysis in order to prevent severe diseases (other than LB specific) and sequels of injuries (other than LB specific) from interfering with the analysis. Therefore, SA data includes all LB specific SA episodes, regardless of their length, and all other cause SA episodes that are less than 31 days. The cut-off limit of 30 days per SA period was chosen arbitrarily.

In the two-arm randomised trial the power calculations were based on the following assumptions: the estimated standard deviation for the PHI score in our population was 4 units. A difference of 2 units between treatment arms is detectable with 85 % power in two-tailed tests with the alpha of 0.05 for a sample of 73 employees in each group, and the standardised effect size is 0.50.

All statistical analyses were performed at employee level, according to the intention-to-treat principle. About 29 % of the data concerning HC costs and health outcomes were missing, mostly due to missed follow-up visits. However, all participants in this study had an equal opportunity to attend the follow-up visits or send their questionnaire information to the study personnel. Therefore, we have assumed that the missing data is missing at random. All missing data were imputed with the multiple imputation method [30] by using the SPSS Statistical Package version 22.0 for Windows ® (SPSS Inc., Chicago, IL, USA).

The following items were either used as determinants or were also imputed (if used with outcomes or characteristics) in the multiple imputation procedure: 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 SA 12 months prior to employee survey, all cause SA in the first follow-up year.

We have presented our results from the basis of two databases, the complete case (original) data and the imputed data (main analysis). Basic characteristics were compared using descriptive statistics. Intervention groups were pooled for the comparison.

The effectiveness of an intervention was primarily estimated by the group difference (Combined vs. NC; Booklet vs. NC) of the outcome variable (PHI and LBP) after two years. For the clinical effectiveness analysis, SA days were accumulated in two years from the baseline. Statistical analyses of the clinical outcomes were performed with the SPSS 22.0 for Windows.

For the cost-effectiveness analyses, SA data were gathered in 13–24 months from the baseline in order to comply with the timeframe of HC costs in an equal way in all study groups. Typically, SA distributions were skewed.

Some of the total HC usage data required manual corrections before transmission into the computer database. For instance, the costs of some radiological tests were accounted manually because of variable unit costs of the different tests.

In order to assess the uncertainty in CEA, we performed a one-way sensitivity analysis and a probabilistic sensitivity analyses (Monte Carlo method and Bayesian, non-parametric bootstrapping with 10 000 replicates) of the comparisons between the intervention groups and NC.

Incremental cost-effectiveness ratio (ICER) is the cost-difference of two interventions divided with the difference of their effects. Hence, ICER summarizes the cost-effectiveness of an HC intervention by representing the average incremental cost (€) which associates in one additional unit of effect (SA day).

One way sensitivity analysis shows how the change in one unit cost influences in the ICER (€ per sickness absence day), when other values remain at their base level.

The results of the CEA main analysis are presented as CE-planes, mean incremental costs (IC) and effects (IE) with corresponding 95 % confidence intervals (95%CI) and the ICER. The uncertainty of CE planes was evaluated with cost-effectiveness acceptability curves (CEAC), which are presented in the additional files. One-way sensitivity analyses for the ICERs are presented in tornado diagrams. The results of the imputed main analysis have also been compared to the complete case analysis (original data).

Within the first three months, four participants from the Combined and five from the Booklet group left the study due to personal reasons, but granted permission to use their data. At the end of the two-year follow-up, 18 participants from the Combined, and 20 participants from the Booklet failed to return their questionnaires, resulting in missing data. The reasons for not continuing with the study remained mostly unknown to the researchers. In both intervention groups, 67 participants continued to the end of the two-year follow-up (activity rates: Combined 73 % and Booklet 75 %). In the NC group, 32 of the eligible 83 persons did not return the postal questionnaire, meaning that the complete data is available for 51 (61 %) participants.

Complete case analysis includes only those participants who returned their HC utilisation questionnaires in the 24-month visit. In the main analysis, however, SA data (as also the multiply imputed questionnaire data) were analysed among all study participants.

The South Karelian Central Hospital Research Ethics Board approved the study on 13.09.2001. All participants received written information regarding the study, in accordance with the Declaration of Helsinki. Only the participants who gave their signed informed consent were included in the study. The documents above are stored with the other study material.