A detailed description of the design of the RCT is available elsewhere (18) but is briefly summarized below. Methods for the economic evaluation are based on the Blue book by Drummond et al. (20) and follow the International Society For Pharmacoeconomic and Outcomes Research (ISPOR) guidelines on economic evaluation alongside RCTs (21). Reporting of the economic evaluation follows the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement (22). Medical ethics approval for the RCT has been obtained (METiGG: NL34721.097.10) and the study was conducted in compliance with the Declaration of Helsinki (23).

The target population consisted of patients with a history of two or more major depressive episodes (MDEs), which is a group with an elevated risk for relapse and recurrence of depression. Patients were recruited between January 2012 and August 2014 in the Netherlands. All received written information about the study and were asked to sign the informed consent.

Patients were included if they (a) had at least two previous MDEs, (b) were currently in remission according to DSM-IV criteria for at least two months as assessed by the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) (24), (c) had no-to-mild depressive symptoms defined as a current score of <14 on the 17 item Hamilton Depression Rating Scale (25), (d) had received prior cognitive therapy, with a minimum of eight sessions, (e) are fluent in Dutch, (f) had no current or history of bipolar illness or any psychotic disorder, (g) had no current alcohol or drug misuse, (h) had no acute predominant anxiety disorder, and (i) provided informed consent.

Every participant was followed-up over a period of 15-months. The follow-up assessments were self-report questionnaires at 3-, 6-, 12- and 15-months post-randomization. The assessments at baseline and at the 15-month follow-up consisted of an interview and self-report measures. All interviews were done by trained assessors who were blinded to the allocation of the participants and who attended regular consensus meetings.

For the economic evaluation, we took the societal perspective as our base case perspective. Therefore, all costs related to the PCT and CAU interventions, other health care uptake, and productivity losses or gains were included. All costs (Table 1) are expressed in euros and were indexed to the reference year 2014 using an inflation and price level correction based on the Harmonized Index of Consumer Prices (HICP) (26).

Health care utilization and productivity losses were measured using the Trimbos Questionnaire for Costs associated with Psychiatric illness (TIC-P) (27). The TIC-P is a self-report measures that measures health care utilizations regarding mental health care, other health care and medication. It also measures productivity losses (absenteeism and presenteeism). Using the TIC-P, the number of contacts with health professionals and amount of informal care was collected for four weeks. This reported number of contacts was extrapolated to the period between the current and previous measurement wave.

To measure the cost of health care, we considered all health care received by the patient, including the costs of PCT/CAU. We used a societal perspective according to the Dutch guidelines for health economic research. Dutch standard cost prices were used to value resource utilization (28) by multiplying the number of contacts with the standard cost prices per contact. Medication costs were valued based on the reports of medication use by the trial participants, collected during the assessments. Frequency of use was assumed to be daily if these data were missing. Cost prices per dose of medication were extracted from the Netherlands Ministry of Health maximum cost prices of medication registry (29). Medication costs represent the combined costs for AD, other psychiatric medication, and medication for somatic illnesses.

Using the Short Form-Health and Labor Questionnaire (SF-HLQ), a subscale of the TIC-P, both absenteeism and presenteeism were measured. Together, these measures comprise our measure of productivity losses. Productivity losses were reported by the patients over the two weeks prior to the day of data collection, as per the TIC-P instructions. The reported hours/days of productivity losses over the two weeks were then extrapolated to the period between the current and previous measurement wave. Productivity losses in hours were multiplied by an estimate of labor costs of 37.90 euro for men and 31.60 euro for women, respectively (28). Productivity losses were valued using the friction cost method. A maximum friction costs period of 85 days and an elasticity factor of 0.8 were applied (30).

The outcome measure for the cost-utility analysis (primary analysis) was the number of quality-adjusted life years (QALY) gained or lost between randomization and the 15-month follow-up. The 5-dimensional, 3-level EuroQol (EQ-5D-3L) quality-of-life instrument was used. The EQ-5D-3L is a self-report instrument which was in this study used to calculate health utilities based on the Dutch tariff (31) were calculated by multiplying the utility of a health state by the time spent in this health state using linear interpolation between measurement time points.

The outcome measure for the cost-effectiveness analysis was the number of depression-free days until relapse/recurrence. The number of depression-free days was calculated by using the date of baseline measurement and the date of relapse/recurrence within the 15-month follow-up period. Relapse/recurrence was operationalized as meeting DSM-IV criteria for a MDE according to the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) over a follow-up period of 15 months (24). The SCID-I is considered to be the gold standard in semi-structured instruments for depression. For the Dutch version of the SCID-I the inter-rater reliability has been shown to be fair to excellent (32). The SCID-I was administered at baseline and at follow-up 15 months later.

We carried out all analyses following the intention-to-treat paradigm. Missing observations in costs and effects data were handled using multiple imputation in the base case scenario, meaning that all data in the tables are based on multiple imputed data unless otherwise indicated. The multiple imputation software package we used is Amelia II (33) for R (34). We have used this package as there are indications that Amelia II handles non-gaussian distributions relatively well (35). Using this software, we imputed the original dataset 10 times, which is sufficient given the low overall rate of missing data in the dataset used for the economic evaluation (6.5%, range 0%–13%) (36). Cost parameters were square root transformed before imputation and back-transformed afterwards to take in account the skewed distribution of these variables because Amelia assumes multivariate normal distribution of data. Analyses were performed on each of these 10 imputed datasets separately, and the outcomes were then combined using Rubin’s rules for combining means and standard errors from multiple imputed datasets (37, 38).

For all participants, we multiplied units of health care (e.g., sessions, contacts, and medication), time investments, and productivity losses by their associated costs. Differences in costs and effects between PCT and CAU were calculated as the difference in cumulative costs over the 15-month time horizon of this study. There was no need to apply baseline correction as randomization had resulted in sufficient comparability across conditions at baseline. All costs were standardized using Organization for Economic Co-operation and Development purchasing power parities for base year 2014. As the time horizon of the study was only a little over a year, no future costs / effects discounting has been applied. As for the costs perspective, the societal perspective was chosen in the base case scenario, while the health care sector perspective is presented the alternative scenario. In order to assess the sensitivity of our findings to misspecification of costs, one-way sensitivity analyses were performed to evaluate the impact on the ICERs of a − 20% to +20% misspecification in the cost categories.

We extracted 5,000 nonparametric bootstrapped samples (with several patients per trial arm equal to the number of patients in the original dataset) from the MI dataset (500 per imputed data set). For each of these 5,000 bootstrapped samples of the dataset, we calculated the incremental costs, incremental effects, and incremental cost effectiveness ratio (ICER). The ICER was calculated as: ICER = (CostsPCT – CostsCAU) / (EffectsPCT – EffectsCAU), where effects were either QALYs or depression-free days. The ICERs were used for further calculations and plotted on the cost-effectiveness plane. The reference intervention (CAU) is positioned in the origin of the cost-effectiveness plane. The horizontal axis indicates differences in health gains between PCT and CAU, while the vertical axis represents differences in costs. Along the horizontal and vertical axis, Figure 1 is divided into quadrants, each with a specific interpretation. ICERs that fall in the upper right (“North East”) quadrant indicate that PCT generated better health at additional costs; the lower left (“South West”) quadrant indicates fewer health gains from PCT than CAU at lower costs. In the upper left (“North West”) quadrant, PCT is dominated by CAU, as poorer health outcomes are then obtained at additional costs. In the lower right (“South East”) quadrant, PCT dominates CAU with better health outcomes against lower costs.

Based on the distribution of the ICERs over the cost-effectiveness plane, cost-effectiveness acceptability curves (CEACs) (39) have been drawn (Figure 1). CEACs show the probability that PCT is more cost effective than CAU as a function of the willingness to pay (WTP) for 1 additional unit of effect (1 QALY or 1 depression-free day). At a probability of 0.5 on the vertical axis, the indifference point is reached. Above this indifference point, PCT has a better likelihood to be preferred over CAU with regard to cost-effectiveness (with a likelihood equal to the probability on the vertical axis). As the WTP per unit of effect is generally an unknown quantity it is presented as a series of increments on the horizontal axis.

A total of 214 patients were included in the RCT. The majority (68%) were women, and the mean age was 43.4 years. The mean number of previous depressive episodes was 3.98 and 28.5% of patients used antidepressant medication (AD) at baseline. None of the baseline characteristics differed between the Preventive Cognitive Therapy (PCT) and care as usual (CAU) conditions, see Table 1.

Table 2 presents the costs for the different cost categories. There was no difference in total societal costs during the 15-month follow-up period, with a bootstrapped mean difference of 858 euro (95% CI −5 to 1753).

Table 3 presents the effect parameters of the economic evaluation in the two conditions. Each row in the table presents the data for each of five consecutive assessment waves. QALY results are cumulative over the current and all previous assessments. What can be observed from Table 3 is that the differential effect on Quality-adjusted life years (QALYs) between the two conditions (−0.016 – (−0.010) = 0.01) is not statistically significant (p = 0.69). The differential effect on depression-free days is statistically significant (p = 0.02) and indicates that the PCT group experienced more depression-free days then the CAU group.

In Table 4, the costs, depression-free days and QALYs gained or lost over the 15-month follow-up period are presented for PCT and CAU. Also, the difference (increment) in costs between the two groups and the incremental cost-effectiveness ratio (ICER) with QALYs, respectively, depression-free days as the effect parameter is shown. The two plots positioned at the top of Figure 1 present two cost-effectiveness planes, with QALYs (top plane) and depression-free days (bottom plane) as the effect parameters. Each dot in the plane represents a bootstrapped mean ICER. By dividing the incremental costs by the incremental effects, the bootstrapped mean ICER of PCT compared with CAU from the societal perspective is calculated 15 months post-randomization.

Figure 1 shows that, after calculating the distribution of the ICERs over 4 quadrants, we found that there was a 47% likelihood that PCT led to additional QALYs at higher costs, and there was a 26% likelihood that PCT led to fewer QALY’s at higher costs. There was a 19% likelihood that PCT led to more QALYs at lower societal costs and an 8% likelihood that PCT led to less QALY’s at lower societal costs. When depression-free days was used as an outcome we found that PCT had a 72% probability of leading to more depression-free days at higher societal costs relative to CAU and a 27% probability of leading to more depression-free days at lower societal costs relative to CAU. There was a 1% likelihood that PCT led to less depression-free days at higher societal costs. At a WTP of ≥175.000 euro per QALY the probability of PCT being cost-effective compared to CAU was ≥50%. At WTP of ≥20 euro per depression-free day the probability of PCT being cost-effective was ≥50%.

To assess the sensitivity of our results to misspecifications of the key cost drivers, we repeated the analyses after raising and lowering the estimated total costs of each of the key cost categories by 20%. The sensitivity analysis was performed on the whole sample and based on the incremental cost-effectiveness ratio over 15 months. We found mental health care costs to be the most influential costs in these analyses and productivity costs were the second most influential.