The methods have been previously reported in the study protocol [10], and the publication of clinical effectiveness [11]. We followed the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) Cost-Effectiveness Analysis Randomised Clinical Trial taskforce recommendations [14].

Study design

An economic evaluation was conducted alongside a three-arm, pragmatic RCT [10]. The trial included participants between April 2019 and October 2020. As well as the economic evaluation, the trial had a follow-up at six months. The trial was conducted in cooperation with the Norwegian Labour and Welfare Administration (NAV). The project was approved by the Norwegian Centre for Research Data (project nm. 861249), and the trial was conducted according to the Helsinki declaration and the General Data Protection Regulation.

Participants, randomisation, and stratification

Eligible participants were workers aged 18–67 years, employed either full or part-time, and on sick leave due to MSDs for at least 50% of their contracted work hours for more than seven consecutive weeks. All participants were diagnosed with MSDs listed in the 2nd edition of the International Classification of Primary Care (ICPC-2) [15]. Excluded were those with serious somatic or mental health disorders affecting their work ability and in need of specialised treatment (e.g., cancer, psychotic disorders), pregnant women, unemployed, freelancers and self-employed workers, and those lacking sufficient proficiency in either Norwegian or English to answer questionnaires or communicate by telephone.

Candidates who agreed to participate received an electronic link to written information about the trial, an electronic informed consent form and a baseline questionnaire.

The Örebro Musculoskeletal Pain Screening Questionnaire Short Form (ÖMPSQ-SF) [16], and the Keele STarT MSK Tool [17, 18] were used to stratify the participants into two risk groups of long-term sick leave. Participants with ≥ 9 on the Keele STarT MSK Tool and ≥ 60 on the ÖMPSQ-SF were stratified to a ‘high-risk group’, all others were stratified to a ‘medium/low-risk group’. After stratification, participants were randomised using a 1:1:1 ratio. Allocation was concealed for the recruitment staff. A blinded statistician prepared a computer-generated allocation sequence for each risk-group, only available for the person in charge of group allocation.

Interventions

A detailed description of the rationale, development and content of the intervention can be found elsewhere [10, 11]. A fidelity assessment of the MI intervention [19], and a process evaluation of the SVAI have been published previously [20]. All participants received UC for sick leave, consistent with Norway’ standard practice, which provides full wage replacement benefits for up to 12 months. The usual case management has the following timeline: within the first 4 weeks of sick leave, an RTW plan is made by the employer and employee; within 7 weeks, a dialogue meeting between the employee, employer, and other relevant stakeholders such as general practitioner (GP), is arranged by the employer. Within week 26 of the sick leave period, NAV arranges a second dialogue meeting between the employee, the employer and in some cases the GP who issued the sick leave.

In addition to UC, participants randomised to the UC + MI arm were offered two face-to-face sessions of MI from a NAV caseworker; the first as soon as possible after random allocation, and the second two weeks later. The NAV caseworkers were educated in MI [10].

The participants in the UC + SVAI arm were offered UC and vocational advice and case management from physiotherapists. In the UC + SVAI group, those stratified to the low/medium-risk group were offered 1–2 telephone sessions, while participants in the high-risk group were offered 3–4 sessions. The first session was conducted as soon as possible after inclusion, and the intervention ended when the participant reached six months of consecutive sick leave or had RTW for four consecutive weeks. Eight physiotherapists were trained over a five-day course to provide SVAI.

Effect measures

The primary effect measure was the number of sickness absence days over a six-month period, defined as lost workdays. To accurately represent time away from work, we accounted the participants’ contracted work hours and amount of sick leave. This was then summed up and converted to lost workdays, assuming a five-day working week. Data was obtained from national registries, including information on sick leave benefits, sick leave certificates, disability pensions, and contracted work hours. In Norway, people may work alongside part-time disability pensions, so any increase in disability pensions from baseline was counted as sick leave.

The secondary effect measure was health-related quality of life expressed in terms of quality adjusted life years (QALYs). First, the participants’ health states were measured by the EuroQol-5 Dimensions-5 Levels (EQ-5D-5L) [21]. Then, the UK tariff was used to convert these health states into utility scores, anchored at 0 “death” and 1 “perfect health”, with negative values representing health states worse than death. We used the UK tariff, as a Norwegian tariff is not available. QALYs were calculated using the “area under the curve” approach. The willingness-to-pay threshold for this outcome was based on the Norwegian governmental report No. 34 to the parliament with a value of NOK 275,000 (Euro (€) 27,500/USD 35,628) per QALY [22].

Cost measures

Since this study adopted a societal perspective, we included both direct and indirect costs. Direct costs included costs of the intervention, primary healthcare use (e.g., general practitioner, physiotherapist, manual therapist, or other therapists), and secondary healthcare use (e.g., hospitalisation or rehabilitation). To calculate intervention costs, we employed a micro-costing approach and included training and mentoring costs. Intervention costs were provided per hour by NAV. Information on other health care use and costs was retrieved from national registers: The Norwegian Health Economics Administration and the Norwegian Patient Registry. Indirect costs consisted of work absenteeism and productivity losses due to paid and unpaid work. We obtained absenteeism data from national registries and valued it using estimates from official statistics on average income stratified by gender. Productivity losses due to unpaid work were measured using the Institute for Medical Technology Assessment Productivity Cost Questionnaire (iPCQ) [23]. The iPCQ has been translated and culturally adapted to Norwegian and found to have good measurement properties when used among patients with long-term MSDs [24]. These costs were valued using a recommended Norwegian shadow price (€150). All costs were converted to 2021 Euros, the last year of data collection, using exchange rates from the European Central Bank. Since the follow-up period of the intervention was less than one year, there was no need to discount the costs and effects.

Statistical analyses

Analyses were performed in accordance with the published statistical analysis plan [10]. All analyses were performed according to the intention to treat principle. Unless stated otherwise, data were analysed using Stata (version 16, Stata Corp, College station, TX).

Missing data

We anticipated few missing values for the primary outcome and the work-related secondary outcomes, as information was obtained from the Norwegian national social security system registry. In this registry, all individuals who received any form of benefits are registered by their social security number. We assumed that missing data from the EQ-5D-5L were missing at random and imputed missing values with a multiple imputation model. Missing data was imputed using Multivariate Imputation by Chained Equations (MICE) with Predictive Mean Matching [25]. The imputation model included duration of sick leave at baseline, risk groups from the Keele STarT MSK and ÖMPSQ-SF, work satisfaction, and self-rated health. Ten complete datasets were imputed. Analyses were performed per imputed dataset separately, and the results were then pooled using Rubin’s rules [25]. MICE was performed using SPSS statistics 25 (IBM).

Cost-effectiveness analysis & cost-utility analysis

In the cost-effectiveness analyses, the outcome measure was sickness absence days, and productivity costs were excluded to prevent double counting. In the cost-utility analyses, productivity costs were included. We used linear regression models, both adjusted and unadjusted for confounders (sex, age, BMI, smoking, education level and physical activity) to analyse disaggregate cost differences. Differences in total costs and effects between treatment groups were obtained from a system of seemingly unrelated regressions that accounted for the potential correlation between costs and effects [26]. These total cost and effect differences were adjusted for baseline and confounders. In both analyses, the incremental cost-effectiveness ratio (ICER) was calculated by dividing the corrected differences in costs by those in effects. To assess uncertainty, we used a bootstrap method with 10,000 replicated datasets. To illustrate the statistical uncertainty surrounding the ICERs, bootstrapped cost and effect pairs were plotted on a cost-effectiveness plane (CE plane) with incremental costs on the y-axis and incremental effects on the x-axis, and on cost-effectiveness acceptability curves (CEACs).

Cost–benefit analysis

The cost–benefit analysis (CBA) was performed from NAV’s perspective. Costs were defined as intervention costs, and benefits as the difference in total monetized outcome measures between the intervention groups and control group. Positive benefits indicate reduced spending of the intervention groups compared with the control group. Two cost–benefit metrics were calculated: (1) net benefits (NBs), and (2) benefit cost ratio (BCR).

To quantify precision, 95% bootstrapped confidence intervals (CIs) were estimated, using 10,000 replications. Financial returns are positive if NB > 0 and BCR > 1.

Sensitivity analyses

The following sensitivity analyses were carried out: 1) Complete-case analysis (including participants with complete data only). 2) Uncertainty of the ICER (incremental cost-effectiveness ratio) will be tested by bootstrapping with 5,000 repetitions.