Patients and assessments

In a multicentre trial in The Netherlands, 86 patients from four study locations were randomly allocated to either SFT (n=44) or TFP (n=42). A pre-randomisation assessment was performed; subsequently, 3-monthly assessments were made for 3 years, with a final follow-up assessment 4 years after the baseline interviews. Both interventions were individual therapies consisting of 50 min sessions twice a week for 3 years. Central to SFT is the assumption of four schema modes specific to borderline personality disorder; recovery is achieved when dysfunctional schemas no longer control or rule the patient's life. Central to TFP is a negotiated treatment contract between patient and therapist, being the treatment frame; recovery is reached when good and bad representations of self (and others) are integrated and when fixed, primitive internalised object relations are resolved. Twelve participants in the SFT group and 22 in the TFP group left the study early. A further 6 (14%) participants in the SFT group and 2 (5%) in the TFP group successfully (according to therapist and patient) terminated treatment within 3 years. For details, see Giesen-Bloo et al. ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen, van Asselt, Kremers, Nadort and Arntz3}

Resources used

At every assessment a structured cost interview was administered by an independent research assistant. Formal registries such as hospital information systems or insurer's databases are considered to be incomplete, since a considerable amount of resource use is situated outside (mental) healthcare institutions. ^{Reference Van Asselt, Dirksen, Arntz and Severens5} Besides, individual patient data cannot be traced from registries. Therefore, patient-reported prospective cost diaries, ^{Reference Goossens, Rutten-van Molken, Vlaeyen and van der Linden6} or retrospective cost interviews, ^{Reference Van den Brink, van den Hout, Stiggelbout, Putter, van de Velde and Kievit7} are the preferred instruments covering all relevant events. We chose a 3-month recall interview, ^{Reference Severens, Mulder, Laheij and Verbeek8} since a prospective cost diary was expected to lead to more missing items, given the patient characteristics. From a societal perspective, the cost interview covered work status and absence, sources of income, domestic activities, informal care, medication use, alcohol and drugs, out-of-pocket expenses, and consumption of healthcare and societal resources (including visits to general practitioners, hospitals, psychiatrists and psychologists, crisis centres, alternative healers, social work and drug rehabilitation centres). Informal care is an important item in mental healthcare, since people with mental illness can impose a heavy burden on their environment: those close to the patient (family, friends or neighbours) take care of the patient and take over domestic tasks. A key characteristic of informal care is that caregivers would not want to care for someone outside their social environment for a similar wage. ^{Reference Van den Berg, Brouwer and Koopmanschap9} Out-of-pocket costs are actual expenses reported by the individual and are highly relevant in people with borderline personality disorder. These costs relate to (for instance) excessive smoking and shopping, binge eating and extremely high telephone bills. ^{Reference Van Asselt, Dirksen, Arntz and Severens5}

The cost interview (based on the cost diary described by Goossens et al) ^{Reference Goossens, Rutten-van Molken, Vlaeyen and van der Linden6} was judged by various experts and pilot tested by research assistants in patients, after which both face validity and content validity of the interview were improved. The questions constituting the cost interview were read to the participants, who indicated whether or not they had been absent from work, used a particular healthcare facility, etc. Only costs related to the person's personality disorder were considered relevant, because unrelated costs were not expected to differ between the treatment groups. Patients were explicitly asked whether they thought resource use was linked to problems arising from their personality disorder. In the rare cases in which resource use was probably disorder-related but the patient indicated that it was not, patients' judgements were overruled. For instance, if patients had regular contact with social services because of problems with raising their children and they rated this as not disorder-related, the contacts were still included in the analysis. Since for alcohol and drugs it is difficult to decide which part of the cost is related to the personality disorder, all alcohol and drug expenses were considered to be disorder-related.

The number of therapy sessions was recorded by the therapist. The number, duration and time (inside or outside office hours) of telephone contacts with the therapist were extracted from the standardised session form therapists filled out after each session. Training and supervision costs were not taken into account, because they are training costs and as such not part of the therapy.

Unit prices

Standardised Dutch unit prices were used. ^{Reference Oostenbrink, Bouwmans, Koopmanschap and Rutten10} When a standardised unit price was not available, prices were based on tariffs. Medication costs were obtained from the Dutch Pharmacotherapeutic Compass. ¹¹ Productivity costs, incurred when patients are unable to perform paid work, were calculated according to the human capital approach, ^{Reference Rice and Cooper12} in which the number of hours of absence from work was multiplied by the actual gross wage per hour. For unpaid work and study, the number of hours absent was multiplied by a shadow price. The shadow price was also applied to the informal care. All unit prices were expressed in euros for the year 2000.

Outcome measures

Effectiveness was expressed as the proportion of recovered patients according to the Borderline Personality Disorder Severity Index (BPDSI) version IV, ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen, van Asselt, Kremers, Nadort and Arntz3,Reference Arntz, van den Hoorn, Cornelis, Verheul, van den Bosch and de Bie13} a 70-item index with nine dimensions, representing the DSM–IV borderline personality disorder criteria. Total scores range from 0 to 90. When the BPDSI score was below 15 at follow-up, patients were considered to have recovered. Additionally, effectiveness was expressed as quality-adjusted life-years (QALYs). A QALY is a measure of life expectancy, weighted by the health-related quality of life (HRQoL) represented by utility scores. Health-related quality of life was measured with the EQ–5D, which contains five dimensions: mobility, self-care, daily activities, pain/discomfort, and depression/anxiety. ^{Reference Brooks14} Each dimension is rated at three levels: no problem, some problems and major problems. Based on preferences elicited from a general UK population, EQ–5D health states can be converted into utility scores. ^{Reference Dolan15} The maximum possible number of QALYs within 4 years is 4 (4 years multiplied by 1, the optimal health state).

Cost-effectiveness

Two incremental cost-effectiveness ratios (ICERs) were determined. These ratios are calculated by dividing the difference in costs between the treatments by the difference in effectiveness between the treatments, and represents the extra amount of money that has to be invested or will be saved to gain or lose one extra unit of effect. The first ICER was the cost per recovered patient and the second ICER was the cost per QALY gained.

Data analysis

Costs and effects were discounted at 4%. ^{Reference Oostenbrink, Bouwmans, Koopmanschap and Rutten10} Missing items in otherwise completed assessments were imputed with SPSS (version 12 for Windows) Missing Value Analysis, option regression. Completely missing assessments because of patients who left the study or completed the treatment before the 3-year limit were analysed by carrying the last observation forward, in accordance with Giesen-Bloo et al. ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen, van Asselt, Kremers, Nadort and Arntz3}

For BPDSI recovery status results after 4 years, a logistic regression was performed with treatment group and baseline BPDSI score as covariates, again in accordance with Giesen-Bloo (further details available from the authors). Since cost data are generally highly skewed, and not distributed normally, bootstrap simulations with 1000 replications were performed to estimate uncertainty intervals around the mean costs. ^{Reference Briggs, Wonderling and Mooney16,Reference Efron and Tibshirani17} For the same reasons, QALY scores and both ICERs were bootstrapped with 1000 replications. The uncertainty interval is represented by the 2.5th and 97.5th percentiles. Results of ICER bootstraps are presented in cost-effectiveness planes and cost-effectiveness acceptability curves (CEACs). Cost-effectiveness planes show differences in costs on the vertical axis and differences in effect on the horizontal axis. Bootstrapped cost-effectiveness pairs located in the northwest quadrant indicate SFT to be inferior to TFP (more costly and less effective than TFP). Cost-effectiveness pairs located in the south-east quadrant show SFT to be dominant over TFP (more effective and less costly than TFP). With respect to the other two quadrants (higher costs for better effectiveness and lower costs for lower effectiveness) the choice of an intervention depends on the threshold value, i.e. what society is prepared to pay for an effectiveness gain, or willing to accept as savings for effectiveness loss. The CEAC represents the probability that, given a certain threshold for the willingness to pay for a QALY or for recovering a patient, the intervention is cost-effective. ^{Reference Van Hout, Al, Gordon and Rutten18}

Secondary analyses

Secondary analyses were performed to assess robustness of the results, initially using the last observation carried forward method for patients who completed the therapy early with consent of the therapist but did not complete later assessments, and imputing baseline values for true ‘drop-outs’, i.e. patients who stopped therapy without consent of the therapist. We then conducted a ‘completers’ analysis, analysing data for only those patients who completed treatment. Patients who completed early with consent of their therapist were included in this analysis, those who dropped out were not.

As costs at baseline were not comparable between both groups, a correction was performed for the following measurements by means of a regression analysis as recommended by Manca et al. ^{Reference Manca, Hawkins and Sculpher19} A correction was also performed for utility and BPDSI scores. An analysis per recovered patient with a healthcare perspective, meaning that only direct healthcare costs (including costs of therapy) were considered, was followed by an investigation of the impact of the unit price of a therapy session, the unit price of informal care and out-of-pocket costs on total costs. We separately doubled and halved the costs of these items. The results of the first four secondary analyses are again represented in a CEAC, combined with the CEAC of the main analysis.

Costs

After start of treatment

Bootstrapped societal costs over 4 years were €46 795 for TFP and €37 826 for SFT (non-significant difference; Table 2). The direct non-healthcare costs were significantly different between the two intervention therapies, mainly owing to informal care costs. With respect to direct healthcare costs, treatment costs (including costs of telephone contacts with the therapist for SFT) were about €2200 higher for SFT, whereas the costs of visits to mental healthcare centres and other healthcare facilities were about €3500 higher for TFP. Table 3 presents mean societal costs, EQ–5D utility scores, BPDSI scores and percentage recovered patients per measurement. Costs in both groups decreased with time. In addition, aside from the second assessment when costs were comparable between the two groups, the TFP costs were continuously higher.

Table 2 Total costs per cost item and bootstrapped costs for subtotals per patient for the schema-focused therapy group (n=44) and the transference-focused psychotherapy group (n=42)

Cost category	TFP costs, €	SFT costs, €	Incremental costs, €	2.5–97.5 percentile
Direct healthcare mean costs
Psychotherapy	10 876	12 946
Contact with therapist by telephone	2	132
Other care at treatment centre	875	565
General practitioner	191	112
Healthcare, other a	6883	3851
Medication	1494	1273
Alternative healers	190	91
Subtotal	20 512	18 969
Bootstrapped subtotal, mean (s.d.)	20 589 (2754)	19 020 (2097)	–1269	–8514 to 4973
Direct non-healthcare mean costs
Paid help	255	170
Societal services b	353	238
Informal care	8223	3053
Subtotal	8831	3460
Bootstrapped subtotal, mean (s.d.)	8915 (2189)	3482 (785)	–5433	–10 385 to 1469
Lost production mean costs
Paid work	890	2016
Unpaid work	3612	536
Domestic tasks	7107	8255
Subtotal	11 609	10 807
Bootstrapped subtotal, mean (s.d.)	11 495 (2886)	10 704 (2394)	–791	–8204 to 6443
Out-of-pocket mean costs
Reported by patient c	2821	1953
Alcohol and drugs	3238	2804
Subtotal	6059	4757
Bootstrapped subtotal, mean (s.d.)	6073 (897)	4752 (734)	–1321	–3578 to 991
Total societal mean costs	47 010	37 994
Bootstrapped total costs, mean (s.d.)	46 795 (5630)	37 826 (3411)	–8969	–21 775 to 3546

SFT, schema-focused therapy; TFP, transference-focused psychotherapy

Table 3 Costs, EQ–5D utility scores, disorder severity scores and patient recovery over the course of the study

	TFP group (n=42)							SFT group (n=44)
Assessment	Cost, € Mean (s.d.)	Utility score Mean (s.d.)	BPDSI score Mean (s.d.)	Patients recovered %	Cost, € Mean (s.d.)	Utility score a Mean (s.d.)	BPDSI score Mean (s.d.)	Patients recovered %
1	4238 (3319)	0.49 (0.26)	34.84 (8.76)	0	3339 (3222)	0.46 (0.31)	33.14 (7.13)	0
2	3712 (2702)	049 (0.32)	29.28 (7.91)	2	3785 (2461)	0.49 (0.30)	27.43 (9.53)	14
3	4049 (4027)	0.66 (0.26)	28.06 (8.86)	5	3549 (2057)	0.50 (0.31)	24.34 (11.33)	18
4	3711 (2905)	0.60 (0.31)	27.16 (8.53)	7	3113 (1806)	0.46 (0.35)	25.31 (9.58)	11
5	3413 (2502)	0.63 (0.26)	24.33 (10.56)	19	2804 (1621)	0.54 (0.32)	22.41 (10.40)	25
6	3160 (2589)	0.60 (0.28)	25.53 (9.66)	14	2536 (1479)	0.50 (0.35)	21.28 (11.14)	34
7	2878 (2273)	0.63 (0.29)	24.96 (10.81)	17	2370 (1511)	0.55 (0.32)	21.90 (11.02)	27
8	2798 (2325)	0.62 (0.30)	24.01 (11.03)	19	2500 (1807)	0.59 (0.32)	19.57 (11.99)	36
9	2788 (2325)	0.61 (0.30)	22.91 (11.40)	31	2348 (1653)	0.63 (0.31)	19.04 (10.64)	36
10	2552 (2273)	0.60 (0.32)	23.47 (10.42)	24	1961 (1402)	0.60 (0.34)	19.90 (11.03)	36
11	2554 (2304)	0.60 (0.31)	23.01 (11.79)	29	1947 (1348)	0.61 (0.32)	18.01 (11.52)	48
12	2546 (2343)	0.58 (0.33)	24.04 (11.80)	26	1963 (1392)	0.59 (0.32)	18.52 (11.64)	48
13	2652 (2599)	0.60 (0.32)	22.71 (10.61)	24	1946 (1617)	0.67 (0.29)	17.76 (11.32)	45
14 b	10 198 (10 878)	0.61 (0.32)	21.98 (11.00)	29	7172 (9912)	0.63 (0.34)	16.94 (11.34)	52

BPDSI, Borderline Personality Disorder Severity Index; SFT, schema-focused therapy; TFP, transference-focused psychotherapy

a. For utility scores in SFT, data for 1 patient were missing for all measurements, therefore n=43 here

b. Assessment 14 covers a whole year, therefore costs are about four times as high

a. Crisis help, first aid, psychotherapy other than the intervention under study, community healthcare, psychiatric hospital, general hospital, physiotherapy

b. Social work, addiction clinic and counselling, sex counselling, abortion clinic

c. Mostly costs of binge eating, excessive smoking, excessive use of mobile telephone, impulsive shopping

Effectiveness

Cost-effectiveness

Societal costs per recovered patient

As the societal costs in the SFT group were lower and recovery rate was higher compared with the TFP group, SFT was dominant over TFP (Table 4). The bootstrapped results (Fig. 1) show that 91% of the cost-effectiveness pairs are in the south-east quadrant, where SFT is dominant. Another 8% lie in the north-east (i.e. more costly, more effective) quadrant, and 1% is located in both west quadrants, where SFT is less effective than TFP. The CEAC (Fig. 2, main analysis) indicates that, regardless of the threshold value, the chance that SFT is cost-effective is over 90%.

Fig. 1 Bootstrapped costs and effects for cost per recovered patient, showing 1000 bootstrap replications for incremental cost-effectiveness of schema-focused therapy compared with transference-focused psychotherapy. Costs are plotted on the y axis and effects on the x axis, so a bootstrap replication in the south-east quadrant means that schema-focused therapy is less costly and more effective than transference-focused psychotherapy for that replication.

Fig. 2 Cost-effectiveness acceptability curves for secondary analyses of cost per recovered patient. BPDSI, Borderline Personality Disorder Severity Index; ICER, incremental cost-effectiveness ratio. The curves indicate the probability (y axis) of schema-focused therapy being cost-effective compared with transference-focused psychotherapy, given the threshold value (x axis) for recovering a patient.

Table 4 Results of main and secondary analyses

	TFP group (n=42)					SFT group (n=44)					ICER
	Costs, €	Proportion of patients recovered	QALYs	Costs, €	Proportion of patients recovered	QALYs (n=43)	Costs per recovered patient (n=86)	Costs per QALY, € (n=85)
Main analysis	47 010	0.29	2.27	37 994 a	0.52	2.15	Dominance	90 457 saving/QALY lost
Completers analysis (n=53)	42 267	0.38	2.47	35 567	0.59	2.34	Dominance	51 541 saving/QALY lost
	(n=21)	(n=21)	(n=21)	(n=32)	(n=32)	(n=32)
Baseline values for drop-outs	49 733	0.19	2.05	39 821 b	0.43	2.17	Dominance	Dominance
Regression corrected costs only	28 227	0.29	2.27	23 195	0.52	2.15	Dominance	41 933 saving/QALY lost
Regression corrected costs and QALYs	28 227	NA	1.97	22 115	NA	1.87	NA	64 008 saving/QALY lost
				(n=43)
Regression corrected costs and BPDSI score	28 227	0.67	NA	23 195	0.82	NA	Dominance	NA
Healthcare perspective	20 512	0.29	NA	18 969	0.52	NA	Dominance	NA
Half unit-price session	41 572			31 521
Double unit-price session	57 886			50 939
Halt unit-price informal care	42 899			36 467
Double unit-price informal care	55 232			41 047
Half out-of-pocket costs	45 599			37 017
Double out-of-pocket costs	49 831			39 947

BPDSI, Borderline Personality Disorder Severity Index; ICER, incremental cost-effectiveness ratio; NA, not applicable; QALY, quality-adjusted life-year; SFT, schema-focused therapy; TFP, transference-focused psychotherapy

a. Costs for n=43 are €36 630 (used for calculation costs/QALY)

b. Costs for n=43 are €39 538 (used for calculation costs/QALY)

Societal cost per QALY

The incremental cost-effectiveness ratio based on the QALY shows that €90 457 is saved when one QALY is sacrificed, as SFT was less costly and less effective than TFP (Table 4). The bootstrapped results (Fig. 3) show that 68% of the cost-effectiveness pairs are in the south-west quadrant, where SFT is less costly and less effective. The south-east quadrant (i.e. the dominance quadrant) contains 28% of the replications, and the final 4% are located in the inferiority quadrant. The cost-effectiveness acceptability curve (Fig. 4, main analysis) indicates that the probability of SFT being cost-effective decreases with an increasing threshold value. Assuming that society's maximum willingness to pay is €20 000 for a QALY gain (or accepts a minimum compensation of €20 000 for a QALY loss), the probability that SFT is cost-effective is 84%.

Fig. 3 Bootstrapped costs and effects for cost per quality-adjusted life-year, showing 1000 bootstrap replications for incremental cost-effectiveness of schema-focused therapy compared with transference-focused therapy. Costs are plotted on the y axis and effects on the x axis, so a replication in the south-west quadrant means that schema-focused therapy is less costly but also less effective than transference-focused psychotherapy for that replication.

Fig. 4 Cost-effectiveness acceptability curves for secondary analyses of cost per quality-adjusted life-year. ICER, Incremental cost-effectiveness ratio. The curves indicate the probability (y axis) of schema-focused therapy being cost-effective compared with transference-focused psychotherapy, given the threshold value (x axis) for a quality-adjusted life-year.

Secondary analyses

QALYs

Utility scores increased from 0.49 to 0.61 in the TFP group and from 0.46 to 0.63 in the SFT group, improvements of 0.12 and 0.17 respectively. The minimally important difference for EQ–5D utility scores is considered to be 0.03 points; ^{Reference Marra, Woolcott, Kopec, Shojania, Offer, Brazier, Esdaile and Anis20} both therapies thus had a clinically relevant effect on quality of life. Therefore, in this sample, the EQ–5D can be considered sensitive to change. However, the resulting QALY was unable to discriminate significantly between the two intervention therapies, although the proportion of recovered patients was higher in the SFT group. Quality-adjusted life-years based on the 4-year period were slightly lower for the SFT group, resulting in an ICER indicating a saving of over €90 000 for a QALY loss.

To date there is no consensus about a reasonable threshold value for cost-effectiveness. Most studies with the explicit goal of determining a threshold found values ranging from €10 000 to €27 000. ^{Reference Gyrd-Hansen21–Reference Laupacis, Feeny, Detsky and Tugwell23} However, in practice, thresholds used for appraisal of new interventions may be higher, around €40 000 for the UK and €80 000 for The Netherlands. ^{Reference Devlin and Parkin24,25} The CEACs show a range of thresholds, varying from €0 to €40 000. In this study, when applying a €20 000 per QALY threshold, SFT in the main analysis would have an 84% probability of being cost-effective. In the secondary analyses, the worst-case scenario had a probability of 68% and the best-case scenario had a probability of 94%. Given this range, the probability that SFT is more cost-effective than TFP in terms of QALYs can be considered moderate to high, in addition to the fact that this probability was already high for the analysis per recovered patient.

BPDSI v. EQ–5D

It is difficult to explain what the reason for the divergence between BPDSI and QALY results might be. When a patient recovers from borderline personality disorder, one would expect the HRQoL to improve correspondingly to at least some extent. In the long run, this was indeed the case: the slopes in increase of HRQoL were significantly steeper in the SFT group than in the TFP group. ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen, van Asselt, Kremers, Nadort and Arntz3} However, during the first half-year of treatment, QALY and BPDSI scores diverged markedly, and as a consequence, during the first 2 years of treatment, the HRQoL in the SFT patients was lower than that in TFP patients. This difference is difficult to understand as it was not paralleled in any of the clinical outcome measures. At follow-up at 4 years, the HRQoL in the SFT group was slightly higher than the HRQoL in the TFP group, but not enough to offset the total QALY difference. Maybe SFT is initially experienced as more burdensome by patients, and it may take time for clinical improvement to be translated into an increase in HRQoL. Another reason for the discrepancy might be a difference in responsiveness between the EQ–5D, which has 5 items each with three-point scale responses, and the BPDSI, which has 70 items with mostly ten-point scale responses. There is little published information concerning the use of the EQ–5D in a population with borderline personality disorder. For depression, the EQ–5D seems useful and sensitive to change; ^{Reference Hayhurst, Palmer, Abbott, Johnson and Scott26,Reference Sapin, Fantino, Nowicki and Kind27} in schizophrenia, results are contradictory. ^{Reference Van de Willige, Wiersma, Nienhuis and Jenner28,Reference Prieto, Sacristan, Hormaechea, Casado, Badia and Gomez29}

Limitations

A number of limitations should be addressed. First, we used the last observation carried forward method to impute missing assessments. Although this is in accordance with the clinical study, this method is regarded as naïve. ^{Reference Oostenbrink and Al30} We therefore also performed a multiple imputation. ^{Reference Rubin and Schenker31} However, probably because of the high cumulative drop-out at the last three measurements, point estimates resulting from the multiple imputation tended heavily towards extreme (and even impossible) values. Because this tendency could not be justified from the trends in the observed data, we have not included the multiple imputation results in this paper. Second, the fact that we only included disorder-related costs might have led to bias in the estimate of the total costs, since patient judgement of what is and what is not disorder-related may be flawed. However, as we do not expect patients' judgements in the SFT group to differ systematically from those in the TFP group, this presumably did not affect the incremental costs. Third, the time horizon of the analysis was 4 years; more long-term information should be collected to confirm the results now presented. Fourth, the economic evaluation was limited to the comparison of only two possible treatments. Other alternatives, such as natural course and care as usual, have not been considered. The use of a single randomised controlled trial as a vehicle for economic evaluation is often an insufficient basis for decision-making, because decision-makers need to be informed about costs and effects for the full range of alternative interventions. ^{Reference Sculpher, Claxton, Drummond and McCabe32} In that respect, our study has contributed evidence to the broader case of cost-effectiveness of treatment for borderline personality disorder.

Implications of the study

As this is the first cost-effectiveness analysis of psychotherapy in the field of borderline personality disorder from a comprehensive societal perspective, it is difficult to place the results in a broader perspective. Even for the much better-documented case of treatment for depression, comparisons between studies are complicated because of vastly different interventions, outcome measures and cost perspectives. ^{Reference Barrett, Byford and Knapp33} The same applies to personality disorders in general and borderline personality disorder in particular. However, our study shows that long-term psychotherapy for borderline personality disorder improves HRQoL and decreases societal costs. In terms of recovered patients, there is a high probability that SFT is more cost-effective than TFP. In terms of QALYs, the probability that SFT is cost-effective is moderate to high.

Future research should further investigate the reason for the discrepancy between the BPDSI and QALY scores, and the relation between the EQ–5D and other measures of quality of life among people with borderline personality disorder. Furthermore, we agree with Giesen-Bloo et al that different comparators such as care as usual and natural course should be studied, ^{Reference Giesen-Bloo, van Dyck, Spinhoven, van Tilburg, Dirksen, van Asselt, Kremers, Nadort and Arntz3} including economic evaluations with a longer period of follow-up.