Methods for additional analyses (e.g. subgroup analyses) {20b}

Baseline data will provide cross-sectional results on work ability, work status, pain intensity, quality of life, physical activity, sleep quality, kinesiophobia, healthcare expenditure and functioning for the complete PSPS-T2 group and comparisons between possible subgroups. Based on the distribution of the data, group differences at baseline will be assessed using a parametric test or its non-parametric alternative at alpha lower than 0.05. Furthermore, correlation analyses will be performed to unravel correlations between the different outcome measures in patients with PSPS-T2, eligible for SCS. Correlation analysis will be performed with Pearson correlation coefficients if the assumption of a linear relation between two variables is met; otherwise, Spearman correlation coefficients will be calculated and tested at alpha < 0.05. Additionally, the association between self-reporting of the work ability (WAI; primary outcome measure) and objective measurements of work capacity (functional capacity evaluation; secondary outcome measure) will be determined.

Correlation analyses will also be performed to evaluate whether changes in physical and mental ability are related to secondary outcome measures, considering appropriate measures for multiple testing. Furthermore, based on the baseline data, we will try to determine predictive factors and which subgroup of patients will benefit the most from the interventional rehabilitation programme. For this, data obtained at the 12-month assessments will be dichotomised and the baseline variables used as explanatory variables. The predictive value of the secondary outcome measures at baseline for the treatment response at 12 months will be examined.

First, a within-trial economic evaluation will be conducted. All costs of all patients will be considered, for the time horizon starting from IPG implantation until the end of the follow-up period. Intervention costs will be based on the study notes documenting the duration of each session per patient. The valuation of resource use is based on national tariffs. A societal perspective is adopted as indirect costs of productivity loss are a crucial part in the analyses. Health outcomes will be expressed in two ways. Effects are expressed in percentage physical and mental work ability, which is the primary outcome in this trial. Next, health outcomes will be considered expressed in utility using health state values from the general public. The comparator is usual care (control group). The overall result is expressed in an incremental cost-effectiveness ratio (ICER, i.e. incremental cost divided by the percentage increment in physical and mental work ability and incremental cost divided by the incremental QALY gained). Differences in cost between both groups will be analysed using generalised linear models. A modified Park test will be used to identify the appropriate link function. The point estimates of incremental costs and increment health benefits as described above (deterministic analyses) are subject to uncertainty which will be addressed in probabilistic analyses [68]. We will apply non-parametric bootstrapping to test for statistical differences in costs and health benefits to investigate the uncertainty around these outcomes and summarised in cost-effectiveness acceptability curves indicating the likelihood of the intervention to be cost-effective at any willingness-to-pay threshold. Reporting on the results of the health economic evaluation will be in line with the CHEERS II guidelines [69].

Besides the within-trial health economic evaluation, a model-based evaluation will be conducted in order to estimate the expected costs and health outcomes in both the control and intervention group beyond the follow-up period of the trial. A Markov model will be developed compliant to the commonly used guidelines [70]. We assume a cycle of 1 year in the model and apply a lifetime horizon. Lifetime incremental costs and QALYs will be the input for the ICER calculation. Discount rates of 3% for costs and 1.5% for utilities will be applied, which is in line with the Belgian guidelines [71]. The subsequent probabilistic analyses and reporting strategies are identical to those described above.