Design

The economic evaluation was conducted alongside a multicentre randomised controlled trial (2 × 2 factorial design) comparing once-weekly versus twice-weekly sessions of psychotherapy (CBT or IPT) for depression with a 24-month follow-up. Data presented in this article concern the cost-effectiveness at 12 months after the start of treatment, and were previously published as a chapter in a PhD thesis.^{Reference Bruijniks9} Patients were randomly assigned to four conditions, using an allocation scheme generated by an independent researcher: (a) CBT sessions twice a week (n = 49), (b) CBT sessions once a week (n = 49), (c) IPT sessions twice a week (n = 47) and (d) IPT sessions once a week (n = 55). Block randomisation was used stratified for depression severity (high: Beck Depression Inventory-II (BDI-II) score ≥29; low: BDI-II score ≤28) and treatment site. Blinding of patients and therapists was not possible because of the nature of the intervention. Further details about the study design can be found elsewhere.^{Reference Bruijniks, Bosmans, Peeters, Hollon, van Oppen and van den Boogaard10}

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human patients were approved by VU Medical Centre Amsterdam (approval number 2014.337). The study is registered with The Netherlands Trial Register (https://www.trialregister.nl; trial registration number NTR4856). Participant registration took place from October 2014 to April 2018. All adult participants provided written informed consent to participate in this study.

Participants

Patients were adult out-patients referred to one of nine Dutch specialised mental healthcare centres located across The Netherlands. Inclusion criteria were as follows: (a) a primary diagnosis of DSM-IV or DSM-5 major depressive disorder (including chronic depression) or DSM-5-based persistent depressive disorder as confirmed by the Structural Clinical Interview for DSM-IV Axis I disorders (SCID-I^{Reference First, Spitzer, Gibbon and Williams11}) or the Mini-International Neuropsychiatric Interview Plus (MINI-Plus^{Reference Lecrubier, Sheehan, Weiller, Amorim, Bonora and Harnett Sheehan12}), (b) aged 18 to <65 years, (c) sufficient knowledge of the Dutch language, (d) pre-treatment score of ≥20 on the BDI-II^{Reference Beck, Steer and Brown13} and (e) access to internet facilities (some assessments were online). Exclusion criteria were as follows: (a) starting antidepressants or dosage change <3 months before baseline; (b) acute risk of suicide; (c) DSM-IV or DSM-5 diagnosis of substance use disorders; (d) presence of a DSM-IV or DSM-5 diagnosis of a cluster A or B personality disorder, as evaluated by a clinician during the intake with or without a structured interview and (e) having received more than five sessions of adequate CBT or IPT in the previous year (clinician-evaluated at intake).

Interventions

The same treatment manuals were used for both CBT and IPT regardless of session frequency. CBT was based on the manual by Beck et al^{Reference Beck, Rush, Shaw and Emery14} and IPT was based on the manual by Klerman et al.^{Reference Klerman, Weissman, Rounsaville and Chevron15} Both CBT and IPT consisted of 12–20 face-to-face 45-min sessions, with the total number depending on patient progress. Participants randomised to the condition with twice-weekly sessions received 16 sessions during the first 8 weeks of treatment, and four sessions during the final 8 weeks (up to 20 sessions over a period of 16 weeks). Patients randomised to the condition with once-weekly sessions received 16 sessions during the first 16 weeks of treatment, and four sessions during the final 8 weeks (up to 20 sessions over a period of 24 weeks).

Clinical outcomes

Severity of depressive symptoms was measured with the BDI-II at baseline, and at 3, 6, 9 and 12 months after baseline.^{Reference Beck, Steer and Brown13} The BDI-II is a 21-item self-report instrument assessing depressive symptoms over the past 2 weeks, with higher scores indicating more severe depression.^{Reference Wang and Gorenstein16} Quality of life was measured at baseline, and 3, 6, 9 and 12 months of follow-up, using the five-level version of the EuroQol (EQ-5D-5L^{Reference Versteegh, Vermeulen K, Evers S, de Wit, Prenger and Stolk E17}) and the RAND 36-Item Health Survey (RAND-36^{Reference Van Der and Sanderman18}). EQ-5D-5L health states were converted to utility values with the Dutch EQ-5D-5L tariff^{Reference Versteegh, Vermeulen K, Evers S, de Wit, Prenger and Stolk E17}. The utility values were then used to calculate quality-adjusted life-years (QALY), using an area-under-the-curve approach. Specifically, we multiplied the average utility value of two measurements with the time in years that had passed between these two measurements. Next, the QALY estimates for each time period were summed to calculate the total number of QALYs over 12 months. Based on the RAND-36, scores for two subdomains were calculated: the physical functioning score (PFS) and the social functioning score (SFS). These scores were transformed to a 0–100 scale, with higher scores indicating better quality of life.

Cost outcomes

Costs (given in Euros, index year 2021) were measured from a societal perspective, using a specifically adapted version of the Trimbos/iMTA Questionnaire for Costs associated with Psychiatric Illness (TiC-P)^{Reference Bouwmans, De Jong, Timman, Zijlstra-Vlasveld, Van Der Feltz-Cornelis and Tan Swan19} at baseline, and 3, 6, 9 and 12 months of follow-up. Thus, discounting was not necessary. Costs included healthcare costs, informal care costs and lost productivity costs. Lost productivity costs included absenteeism from paid and unpaid work, and presenteeism. Presenteeism is defined as coming to work despite having health problems, resulting in less efficiency when working. Use of healthcare services was valued with Dutch standard costs when available.^{Reference Roijen and Tan20} If not, tariffs of professional organisations or professionals themselves were used. Medication was valued with prices from the Dutch National Healthcare Institute (www.medicijnkosten.nl). Absenteeism from paid work was valued with the friction cost approach. The friction cost approach assumes that sick employees are replaced after a certain time period (the friction period), after which productivity is restored to the old level. A friction period of 12 weeks was used, and absenteeism was valued using gender-specific average wage rates for the Dutch population.^{Reference Roijen and Tan20} Participants registered their level of efficiency when present at their work with health complaints (i.e. the efficiency score). Lost productivity was calculated as (1 − efficiency score) × number of days with health complaints × hours per day. This was then valued using gender-specific wage rates. Absenteeism from unpaid work was valued using a shadow price for a legally employed cleaner.

Sample size

Based on a meta regression analysis that indicated an increase from one to two sessions per week increased the effect size g = 0.45,^{Reference Cuijpers, Huibers, Daniel Ebert, Koole and Andersson21} we estimated the post-treatment effect size to be around 0.45. Taking a 20% drop-out rate into account, based on an alpha of 0.05 and power of 0.80, a sample size of 200 patients was needed.

Data analyses

All analyses compared twice-weekly sessions with once-weekly sessions of psychotherapy, and were conducted according to the intention-to-treat principle. Missing clinical outcomes and cost data were estimated by using multiple imputation by chained equations, with predictive mean matching to account for the skewed distribution of costs.^{Reference White, Royston and Wood22} An imputation model was created that included baseline characteristics differing between treatment groups, baseline characteristics differing between participants with and without complete follow-up, baseline characteristics related to missing outcomes and all variables included in the analysis models. The number of imputations was increased until the fraction of missing information was <5%. The imputed data-sets were analysed separately as described below, and results were subsequently pooled according to Rubin's rules.^{Reference Rubin23}

To account for the 2 × 2 factorial design, we adjusted the analysis models for type of psychotherapy (CBT or IPT). Differences in BDI-II scores over time were estimated with a fixed-effects longitudinal mixed model, with patient and time as hierarchical levels. Baseline BDI-II value was included as a covariate in the longitudinal mixed model. Differences in PFS, SFS, QALYs and costs after 12 months were estimated with a linear regression model. To account for the skewed distribution of costs, statistical uncertainty was estimated by using bias-corrected bootstrapping with 5000 replications.

Incremental cost-effectiveness ratios (ICERs) were calculated by dividing the pooled difference in costs by the pooled difference in effects. Bias-corrected bootstrapping (5000 replications) was used to estimate the statistical uncertainty surrounding the ICERs. The proportion of bootstrapped cost-effect pairs in each quadrant of the cost-effectiveness plane was estimated across all imputed data-sets to show uncertainty surrounding the ICER. In addition, cost-effectiveness acceptability curves were estimated that show the probability that twice-weekly sessions are cost-effective in comparison with once-weekly sessions for different ceiling ratios. The probability of cost-effectiveness was determined with the net monetary benefit framework, using pooled estimates of net monetary benefit and the s.e. The ceiling ratio is the amount of money that society is willing to invest to gain one unit of improvement in a specific effect outcome. For outcomes like the BDI-II, PFS and SFS, no ceiling ratios have been defined. For QALYs, the National Institute for Health and Care Excellence in the UK established ceiling ratios between £20 000 and £30 000 per QALY gained (between €23 000 and €34 000 per QALY gained). The utility scores for having moderate and severe depression have been estimated at 0.52 and 0.39, respectively, resulting in a disease burden of 0.48 and 0.61 (i.e. 1 – utility score), respectively.^{Reference Kolovos, Bosmans, van Dongen, van Esveld, Magai and van Straten24} In The Netherlands, ceiling ratios between €20 000 and €50 000 per QALY gained are used for health conditions with a disease burden between 0.41 and 0.7.

Sensitivity analyses

Four sensitivity analyses were conducted to assess the robustness of the results. The first sensitivity analysis concerns an analysis from the healthcare perspective, meaning that only healthcare costs were included. In the UK for example, reimbursement decisions are made from the healthcare perspective. In the second sensitivity analysis, lost productivity costs owing to absenteeism from paid work were valued with the human capital approach. The human capital approach assumes that lost productivity costs are generated during the full period of absenteeism and is internationally more commonly used than the friction cost approach. For the third sensitivity analysis, outliers for costs were recoded as missing and imputed in the multiple imputation procedure. Outliers were arbitrarily defined as participants generating total societal costs of €10 000 or more at any of the four time points at which cost questionnaires were administered. In the fourth sensitivity analysis, total societal costs over 12 months after the start of treatment were adjusted for total societal costs in the 3 months before start of treatment.