j ourna l homepage: www.e lsev ie r.com/ locate / j ad
Research report
Cost-effectiveness of a minimal intervention for stress-related sick leave ingeneral practice: Results of an economic evaluation alongside a pragmaticrandomised control trial
Kimi Uegaki a,b, Ingrid Bakker c,d, Martine de Bruijne b,⁎, Allard van der Beek b,e, Berend Terluin c,Harm van Marwijk c, Martijn Heymans f,g, Wim Stalman c, Willem van Mechelen b,e
a Health Technology Assessment Unit, EMGO Institute-VU University Medical Center, Amsterdam, The Netherlandsb Department of Public and Occupational Health, EMGO Institute-VU University Medical Center, Amsterdam, The Netherlandsc Department of General Practice, EMGO Institute-VU University Medical Center, Amsterdam, The Netherlandsd Wageningen University, Public Health and Society, Wageningen, The Netherlandse Body @Work, Research Center Physical Activity, Work and Health, TNO-VUmc, Amsterdam, The Netherlandsf Department of Methodology & Applied Biostatistics, Institute of Health Sciences, Faculty of Earth & Life Sciences, VU University, Amsterdam, The Netherlandsg EMGO Institute-VU University Medical Center, Amsterdam, The Netherlands
Article history:Received 9 December 2008Received in revised form 15 April 2009Accepted 15 April 2009Available online 12 May 2009
Background: Stress-related mental health problems negatively impact quality of life andproductivity. Worldwide, treatment is often sought in primary care. Our objective was todetermine whether a general practitioner-based minimal intervention for workers with stress-related sick leave (MISS) was cost-effective compared to usual care (UC).Methods:We conducted an economic evaluation from a societal perspective. Quality-adjusted lifeyears (QALYs) and resource use were measured by the EuroQol and cost diaries, respectively.Uncertaintywasestimatedby95% confidence intervals, cost-effectiveness planes andacceptabilitycurves. Sensitivity analyses and ancillary analyses based on preplanned subgroups wereperformed.Results: No statistically significant differences in costs or QALYs were observed. The meanincremental cost per QALY was −€7356 and located in the southeast quadrant of the cost-effectiveness plane, whereby the intervention was slightly more effective and less costly. Forwillingness-to-pay (λ) thresholds from€0 to€100,000, theprobability ofMISSbeing cost-effectivewas 0.58–0.90. For the preplanned subgroup of patients diagnosed with stress-related mentaldisorders, the incremental ratio was −€28,278, again in the southeast quadrant. Correspondingprobabilities were 0.92 or greater.Limitations: Non-significant findings may be related to poor implementation of the MISSintervention and low power. Also, work-presenteeism and unpaid labor were not measured.Conclusions: The minimal intervention was not cost-effective compared to usual care for aheterogeneouspatientpopulation. Therefore,wedonot recommendwidespread implementation.However, the interventionmay be cost-effective for the subgroup stress-relatedmental disorders.This finding should be confirmed before implementation for this subgroup is considered.
Poor mental health in the working population is asignificant problem. The prevalence of mental health pro-blems in persons of working age has been reported to rangebetween 10 and 18% (Dewa et al., 2004; Kessler and Frank,
178 K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
1997; Wang et al., 2006), and it has been estimated that by2020, the global burdenwill be second only to ischemic heartdisease (Murray and Lopez, 1996). In up to 90% of all workerswith mental health problems, ‘stress’ has been identified asthe underlying cause (Nystuen et al., 2001). The word ‘stress’refers to “a particular relationship between the person andthe environment that is appraised by the person as taxing orexceeding his or her resources and endangering his or herwell-being” (Lazarus and Folkman, 1984). Stress is a complexprocess in which ‘stressors’ — such as (job and other)demands, psychosocial difficulties and life events that areperceived as threatening a person's well-being and function-ing — cause psychological distress. At the same time, thedistressed person tries to reduce the distress by ‘coping’ withthe ‘stressors’. Sick leave is often associated with a failure tocope, resulting in social dysfunctioning (Terluin et al., 2004).
Mental health problems can have a negative effect onquality of life, and lead to suboptimal productivity withconsiderable socioeconomic consequences (Lim et al., 2000).An upward trend of stress-related sick leave and earlyretirement has been reported across the European Union(McDaid et al., 2005). In The Netherlands, one-third of alllong-term sick leave or work disability cases are attributableto stress-related mental health problems (Anema et al., 2006;Nieuwenhuijsen et al., 2003). The combination of sick leaveand disability benefit payments along with prematuredeparture from active employment due to mental healthproblems strain the functioning and sustainability of socialwelfare systems from two sides: resources consumed aspayments are no longer available for other purposes, andpremature attrition in eligible working population decreasesa system's income base to meet the demands of a growingageing population (McDaid et al., 2005).
Worldwide, 60–95% of those with mental health problemsfirst seek assistance in primary care (Üstün and Von Korff,1995). In The Netherlands, themajority traditionally seek carefrom their general practitioner (GP). While stress-relatedmental health problems can significantly affect an individual'sability to work and result in (prolonged) sick leave, thisfunctional aspect is often not addressed by GPs (Buijs et al.,1999). Consequently, work disability may be unnecessarilyprolonged. Given the increasing prevalence of this healthproblem, the negative consequences, the reality of limitedresources and that care is often sought by the GP, there is aneed to identify cost-effective GP-based interventions. There-fore, the aim of our study was to determinewhether a generalpractitioner-based minimal intervention strategy was cost-effective compared to usual GP care for workers with stress-related sick leave.
2. Methods
2.1. Study design
We conducted an economic evaluation from a societalperspective alongside a pragmatic, cluster-randomised con-trolled trial in the Dutch primary care setting. The follow-upperiod was 12-months. We selected a societal perspective asthe impact of stress-relatedmental health problems is society-wide. To facilitate comparison with other full economicevaluations, we conducted a cost-utility analysis (CUA). The
protocol was approved by the medical ethical committee ofthe VU University Medical Center (Amsterdam, The Nether-lands). A brief overview is presented below.
2.2. Randomisation
Randomisation was conducted at the level of the GPs inorder to prevent contamination. GPs were recruited at fourdifferent time points, and at each moment, those consentingto participate were randomised to either usual care (UC;N=22) or MISS (N=24). GPs who were randomised to theMISS group followed an 11-hour course consisting of two 3.5-hour training and two 2-hour follow-up sessions to learn thecontent and how to apply theMISS. The trainingwas providedby the developer of the intervention and an experiencedoccupational physician. During the trial, in both groups, thecontent of the actual patient treatment was at the discretionof the respective GPs. GPs were not involved in the patientrecruitment procedure. GPs in both groups were informed ofa patient's participation in the study twomonths after a givenpatient's enrollment (Bakker et al., 2006).
2.3. Interventions
The Minimal Intervention for Stress-related mental dis-orders with Sick leave (MISS) is a for-GPs customized versionof an activating approach by occupational physicians for thetreatment of workers with adjustment disorder, which wasfound effective in the occupational health care setting (vander Klink et al., 2003). Customization to the general practicetook the practicing physicians' time constraints and nature asgeneralists into consideration. Unique aspects of the MISSwere its basis on time contingent and activating principles aswell as coping and focus on functional recovery (i.e. patientsare able to return to work) as opposed to solely symptomreduction. The MISS was developed on the basis of threeconsultations over a time span of four weeks, and encom-passed the following five key tasks: 1. diagnosing stress-related mental disorders; 2. providing education about theproblem and importance of taking an active role in one'sfunctional recovery; 3. advising patients on how to reflect,cope and problem-solve; 4. monitoring progress; and 5.referring to specialists (Bakker et al., 2006). Note that in theactual study, the number of consultations could vary.
Usual care provided during the study was comparable tousual care in real life. The GPs randomised to the usual caregroup did not receive any information or advice regardingstandardized management of stress-related mental healthdisorders nor about theMISS (Bakker et al., 2006). At the timeof the study, while the problem of stress-related mentalhealth disorders is recognized in Dutch general practice,specific guidelines for standardized GP management of thesepatients did not exist. As such, usual care provides ameaningful comparison to determine the added value of theMISS and potential inclusion in future guidelines.
2.4. Patient recruitment
Based on a sample size calculation on an expecteddifference of 15% in the number of patients back at full returntowork after a period of 3 months, a power of 80% at a level of
Table 1Cost prices used for valuation of resource use in the economic evaluation(Year 2004).
Units [units of measurement] Cost price
Health care sectorPrimary CareGeneral practitioner
Office consultation [No.] € 20.44a
Telephone consultation /renewal of prescription [No.] € 10.22a
House call [No.] € 40.88a
After-hours telephone consultation [No.] € 24.30b
Diagnostic testsBlood [No.] € 21.66a
Urine [No.] € 14.42a
X-rays [No.] € 49.80c
Other diagnostic tests [No.] Variableb
Psychologist in private practice [No. of sessions] € 76.90a
Social worker [No. of sessions] € 48.43b
Physical therapist [No. of sessions] € 23.02a
Other paramedical professionals [No. of sessions] Variablec
Medications [per medication] Variabled
Secondary careRegional Institute for Community Mental Health Care(including Drug & Alcohol Abuse Centre) [No. of sessions]
€ 125.47a
Hospital-based psychiatrist− [No. of sessions] € 64.18a
Part-time psychiatric day programs−General hospital[No. of sessions]
€ 90.58a
Medical specialist−Outpatient [No. of consultations] € 56.66a
Medical specialist−ER [No. of consultations] € 140.64a
Hospitalization−General hospital [No. of days] € 340.99a
Professional home health care [per hour, all services] € 31.06a
Professional family home assistance [per hour] € 27.10a
Alpha help [per hour] € 12.85a
Other sectorOccupational physician [No. of consultations] € 21.50c
Patient/familyAlternative care [No. of sessions] Variablec
Support group/self-help courses [No. of sessions] Variablec
Informal help [per hour] € 8.40a
Productivity lossesSick leave [per hour] € 35.39e
Intervention costsTraining costs for MISS [per MISS subject] € 120.46f
Cost price sources: a Dutch guidelines for costing studies; b Dutch CentralOrganization for Health Care Charges (CTG); c Respective providers orprofessional organizations; d Royal Dutch Society for Pharmacy; e Nationalaverage from the Dutch guidelines for costing studies; f Determined via abottom-up calculation; total training costs were € 27,344.
179K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
0.05, a 30% dropout rate (approximately 10% at each follow-up moment), and consideration for potential clustering ofeffects within practices, 415 workers were needed for thestudy (Bakker et al., 2006).
Patients were recruited by the researchers from a sourcepopulation of GP attendees accessed via participating GPs'computerized medical record systems. Patients who had asevere psychiatric disorder (i.e. mania or psychosis), aterminal illness, or poor command of the Dutch languagewere excluded by the GPs. Remaining patients who had aconsultation during the previous one to two weeks and wereaged between 20 and 60 years, were mailed an informationpacket about the study along with a screening questionnaireto determine inclusion. The inclusion criteria were: 1.presence of at least moderate distress, 2. gainful employmentand 3. (partial) sick leave for no longer than three months.Distress was measured with three questions of the Four-Dimensional Symptom Questionnaire (4DSQ) distress scale(Terluin et al., 2006). Patients were informed that the studyrevolved around stress and sick leave. However, patients wereblinded to the fact that two different interventions were beingevaluated. Patients were invited to fill in the screeningquestions and return the questionnaire only when theyfulfilled the inclusion criteria. However, in order to be ableto determine the response rate, a random sample of 336patients were asked to return the questionnaire irrespectiveof the screening outcome.
Patients who returned the screening questionnaires andscored positively for the three screening questions werecontacted by telephone. Enrollment in the study was finalizedduring the telephone contact, following verification of theinclusion criteria and receipt of oral informed consent.Writteninformed consent for participation and access to the subjects'medical records were obtained separately. This method ofpatient recruitment was used instead of screening by GPs inorder to minimize selection bias. During the study, patientbehavior in terms of seeking care from their GP or other healthcare providers was not dictated by any protocol (Bakker et al.,2006).
2.5. Clinical outcome measure
The primary clinical outcome measure for the economicevaluation was quality-adjusted life years (QALYs). A combi-nation of telephone interviews and questionnaires were usedto collect data on at baseline and 2-, 6- and 12-months follow-up. Health-related quality of life was measured by theEuroQol-5D (EuroQol Group, 1990), and the utilities weredetermined by Dutch tariffs (Lamers et al., 2005). QALYs forthe 12-month follow-up were computed by multiplying theutilities by the time spent in the given health state, and thenlinearly interpolating the transitions between the fourmeasured health states (Dolan, 1997).
2.6. Resource use measurement and valuation
We extracted the number of GP consultations, medicationsand laboratory tests from the computerized medical recordsfor a 12-month period starting from the first day of sick leave.Data on additional resource use over 4-week periods werecollected via telephone interviews and questionnaires at
baseline and 2-, 6- and 12-months follow-up. These datawereextrapolated over 12-months by linear interpolation. Wequantified productivity loss from paid work in terms of totalsick leave hours. Sick leave hours were derived from the totalduration of complete and partial sick leave periods from thefirst day of sick leave to 1-year thereafter. In cases of partialsick leave, we assumed that subjects were 100% productiveduring the hours of partial work resumption. As sick leaveperiods were measured in calendar days, we converted thisamount into work-hour equivalents based on a Dutch averageof 1540 work hours per year (Oostenbrink et al., 2004).
The cost prices used for valuing units of resource use arepresented in Table 1. The index year for the study was 2004.Wherever possible, we applied standard cost prices according tothe Dutch Manual for Costing (Oostenbrink et al., 2004). Whenstandard cost prices were not available, we used tariffs or anaverageprice according toprovidersorprofessionalorganization.We calculated medication costs using unit prices published by
180 K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
the Royal Dutch Society for Pharmacy (Z-index, 2006). Thetraining costs for theMISS groupwere determined via a bottom-up calculation that included trainer costs, GP opportunity costsassociated with attending the training, GP travel costs, room andequipment rental, refreshment and training material costs. Weestimated productivity loss costs by the Friction Cost Method(FCM),which assumes that costs are limited to the frictionperiod(i.e. the time it takes tofinda replacement), and that thedecreasein productivity is less than 100% of the time lost at work (i.e.elasticity) (Koopmanschap and van Ineveld, 1992; Oostenbrinket al., 2004). In this study, a friction period of 154 calendar daysand an elasticity of 0.8 were used. Because the follow-up waslimited to one year, discounting was not applied to the costs oroutcomes (Oostenbrink et al., 2004).
2.7. Data analysis
We analysed data according to the intention-to-treat princi-ple. Baseline characteristics were comparable between thegroups, therefore, we did not adjust our analyses for potentialconfounding. Descriptive analysis of resource use units werebased on complete cases.Mean utilities at baseline and the 2-, 6-and 12-month follow-up periods were based on available cases.Descriptive analyses were conducted in SPSS version 12.0.2.
The main cost, effect and joint cost-effect analyses werebased on imputed data. We conducted a cost-utility analysis(CUA) in which we calculated the incremental cost-effective-ness ratio (ICER) as the difference in total costs (i.e. TCFCM
where total costs equals the sum of the health care sector,other sector, patient/family and MISS training costs, andproductivity loss costs estimated by the FCM) divided by thedifference in QALYs. We imputed missing data using amultiple imputation (MI) procedure based on MultivariateImputation by Chained Equations (MICE) (Van Buuren andOudshoorn, 2000). In the MI procedure, five imputed datasets were generated, each of which were analysed separately.Mean QALY differences were analysed with parametric testsand the uncertainty was estimated by 95% confidenceintervals. The 95% confidence intervals around the meancost differences and ICER differences were obtained by a biascorrected and accelerated (Bca) bootstrapping procedurewith 1000 replications (Efron and Tibshirani, 1993). Weapplied Rubin's rules to obtain the pooled estimates of meancosts and QALYs, mean cost and QALY differences and 95%confidence intervals (Rubin, 1987). To gain insight into theuncertainty around the pooled mean ICER from the cost-utility analysis (CUA), we plotted the pooled bootstrappedcost-effect pairs in a cost-effectiveness plane (Black, 1990)and generated an acceptability curve (van Hout et al., 1994).In the CE-plane, we included a maximum willingness-to-pay(WTP) level that we determined using a formula proposed bythe Dutch Council for Public Health and Health Care:€80,000⁎(1−mean quality of life) (Council for Public Healthand Health Care, 2006). The multiple imputation, cost, effectand cost-utility analyseswere conducted in R (R DevelopmentCore Team, 2007).
2.8. Sensitivity analysis
We performed four sensitivity analyses to test therobustness of our main CUA findings. Given that productivity
loss costs from paid work may represent a significantproportion of the total costs (i.e. are a cost driver) and thisstudy involved a working population, we assessed the impactof three different estimations on the ICER. In the firstsensitivity analysis (SA1), we estimated the productivityloss costs using the Human Capital Approach (HCA) instead ofthe FCM. In the HCA, total sick leave time is not “capped” as inthe FCM nor is elasticity considered. In the second sensitivityanalysis (SA2), we quantified sick leave under the assumptionthat subjects who partially resume work are completelyunproductive (versus 100% productive) as they are not yetfully recovered and may only be working on a therapeuticbasis, and then estimated the corresponding productivity losscosts using the FCM. In the third sensitivity analysis, we usedthe same quantification of sick leave as in SA2 but used theHCA to estimate the corresponding productivity loss costs.Lastly, we repeated the main CUA using only the completecases.
2.9. Ancillary analysis
We conducted ancillary analyses based on three pre-planned subgroups. The subgroups were identified from thetotal study population according to the GPs' diagnosis and/orworking hypothesis extracted a posteriori from their electro-nic databases. The first subgroup, stress-related mentaldisorder (SMD), referred to those with elevated yet uncom-plicated distress or, in other words, the absence of adepressive or anxiety disorder, and resembling adjustmentdisorders, neurasthenia or nervous breakdown. The second,other mental health problems (Other MHP), referred to thosewho presented with mental health problems, includinganxiety or depressive disorders in addition to elevateddistress. The third, somatic problems (SP), included thosewho presented with physical complaints along with elevateddistress levels.
3. Results
3.1. Patient and data availability
An overview of the patient recruitment, reasons for notparticipating and inclusion procedure is presented in Fig. 1.Between September 2003 and October 2004, a screeningletter was sent to the source population of 22,740 GPattendees. The response percentage in the sub-sample of336 patients was 51.5%. The total number of responders whomet the inclusion criteria was 509 (MISS=266; UC=243).Ultimately, 433 patients enrolled in the study (MISS=227;UC=206). Baseline demographic and symptom severitycharacteristics are presented in Table 2.
After baseline, drop-out rates were 24/227 (10.6%) for theMISS group and 34/206 (16.5%) for the UC group. The overallnumber of drop-outs through to the end of the 1-year follow-up was 42/227 (18.5%) for the MISS group and 47/206(22.8%) for the UC group. Reasons for drop-out are presentedin Fig. 1. Statistically significant but small baseline differencesbetween drop-outs and completers were observed in age,depressive symptom scores and utility. There were nostatistically significant between-group differences in baselinecharacteristics among drop-outs, nor among completers.
Fig. 1. Patient flow chart and data availability between baseline and the 12-month follow-up.
181K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
An overview of available data is provided in Fig. 1. Therewere no statistically significant between-group differences inthe data availability nor the number of complete cases. Therewas a small, statistically significant difference in mean age
between complete cases and those with incomplete data.Among complete cases, there were no statistically significantdifferences in baseline characteristics between the MISS andUC groups.
Table 2Baseline demographics and symptom severity of enrolled patients.
Patients baseline measures MISS (N=227) Usual care (N=206)
Women N (%) 153 (67) 134 (65)Mean age (S.D.) 41.97 (8.8) 39.50 (9.6)Married or cohabiting N (%) 174 (77) 148 (72)Level of education N (%) Low 59 (27) 46 (22)
Intermediate 94 (42) 102 (50)High 70 (31) 57 (28)
Mean (S.D.) number of visits to the PCP, counted from the first dayof sick leave+3 months
2.55 (2.12) 2.50 (2.23)
Mean utility (S.D.; % available cases) 0.68 (0.22; 79%) 0.68 (0.26; 78%)4DSQ scores available N=180 (80%) N=161 (78%)Distress above threshold N (%)a 140 (78.2) 131 (81.9)Depression above threshold N (%)b 86 (48.0) 72 (45.0)Anxiety above threshold N (%)c 54 (30.2) 48 (30.0)Somatization above threshold N (%)d 103 (57.5) 86 (53.7)
aDistress scores range from 0 to 32. Threshold for an elevated distress level is a score N10. BaselineMISSmean score=19.21 (S.D.=8.5); UCmean score=18.79 (S.D.=8.1).bDepression scores range from 0 to 12. Threshold for an elevated depression level is a score N2. Baseline MISSmean score=3.46 (S.D.=3.7); UCmean score=3.38(S.D.=3.6).cAnxiety scores range from 0 to 24. Threshold for an elevated anxiety level is a score N7. BaselineMISSmean score=5.51 (S.D.=5.5); UCmean score=5.41 (S.D.=5.5).dSomatization scores range from 0 to 32. Threshold for an elevated somatization level is a score N10. Baseline MISS score=12.88 (S.D.=6.9); UC meanscore=12.35 (S.D.=6.8).
182 K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
3.2. Clinical outcome
For the MISS group, the mean utilities (standard devia-tions) from baseline and the 2-, 6- and 12-month follow-ups
Table 3Means and standard deviations (S.D.) of resource use per patient and utilization ra
Psychologist in private practice [No. of sessions] 4.1Social worker [No. of sessions] 1.6Physical therapist [No. of sessions] 5.6Other paramedical professionals [No. of sessions] 0.3Medications [per medication] –
Secondary careRegional Institute for Community Mental Health Care(including Drug & Alcohol Abuse Centre) [No. of sessions]
1.0
Hospital-based psychiatrist [No. of sessions] 0.4Medical specialist [No. of consultations] 1.8Part-time psychiatric day program [No. of sessions] 0.1Hospitalization [No. of days] 0.2Professional home health care [per hour, all services] 0.5Professional family home assistance [per hour] 0.0Alpha help [per hour] 0.0
Other sectorOccupational physician [No. of consultations] 4.5
Patient/familyAlternative care [No. of sessions] 1.5 b
Support group/self-help courses [No. of sessions] 0.4Informal help [per hour] 40.2
Productivity lossesSick leave from paid work [hours] 411.2
a Medication utilization rate in MISS group significantly higher than UC (X2-test;b Mean number of visits to an alternative care provider in MISS group significant
were 0.68 (S.D.=0.22), 0.75 (S.D.=0.24), 0.79 (S.D.=0.21)and 0.83 (S.D.=0.19), respectively. For the UC group, therespective utilities were 0.68 (S.D.=0.26), 0.73 (S.D.=0.22),0.76 (S.D.=0.24) and 0.81 (S.D.=0.21). There were no
te (%) per group during the 12-month follow-up (based on complete cases).
Fig. 2. Cost-effectiveness planes representing the uncertainty around the mean incremental cost and mean incremental effectiveness for the total group andsubgroups, and distribution of cost-effect pairs above and below a maximum willingness-to-pay (WTP) level of €25,600.
183K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
statistically significant between-group differences in utilities(pN0.25) nor QALYs (MISS: mean=0.78 (S.D.=0.16); UC:mean=0.76 (S.D.=0.18); difference: 0.02; 95% CI: −0.01;0.06).
Table 4Mean cost differences (ΔC), mean effect differences (ΔE), and incremental cost-effsensitivity analyses (SA1, SA2, SA3 and SA4) for the total group, and the cost-utility
a In the main analysis, ΔE=mean difference in QALY, ΔC=mean difference in totΔC = mean difference in total costs in which the productivity loss costs are estimproductivity loss costs are estimated by the FCM but an assumption is made thatdifference in total costs in which the productivity loss costs are estimated by the HCproductive; SA4 is a repetition of the main analysis using only the complete cases; Suhealth problems; SP = Somatic problems.
b Refers to the northeast quadrant of the CE-plane, which indicates that MISS is mc Refers to the southeast quadrant of the CE-plane, which indicates that MISS is md Refers to the southwest quadrant of the CE-plane, which indicates that MISS ise Refers to the northwest quadrant of the CE-plane, which indicates that MISS is
3.3. Resource use
Resource use is presented in Table 3. There were nostatistically significant differences in the mean number of GP
ectiveness ratios from the main cost-utility analysis and corresponding fouranalysis for each subgroup.
al costs inwhich the productivity loss costs are estimated by the FCM; in SA1,ated by the HCA; in SA2, ΔC = mean difference in total costs in which thethose who partially return-to-work are not productive; in SA3, ΔC = meanA but an assumption is made that those who partially return-to-work are notbgroup SMD = Stress-related mental disorders; Other MHP = Other mental
ore effective and more costly than UC.ore effective and less costly than UC.less effective and less costly than UC.less effective and more costly than UC.
Fig. 3. a. Cost-effectiveness acceptability curves of the CUA based on imputed data of the total group and four corresponding sensitivity analyses: SA 1 where thetotal costs include productivity loss costs estimated via HCA; SA 2 where the total costs include productivity loss costs estimated via FCM and assumption thatpartial work resumption is not productive; SA 3 where the total costs include productivity loss costs estimated by HCA and assumption that partial workresumption is not productive; and SA 4where the CUA is based on complete cases. b. Cost-effectiveness acceptability curve of the CUA based on imputed data of thesubgroup, including the total group for reference.
184 K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
visits or other resources during the 12-month follow-up exceptfor the difference in mean number of visits to alternativepractitioners (MISS: mean=1.5 (S.D.=3.9); UC: mean=4.1(S.D.=11.0);pb0.05).All subjects, regardlessof groupallocation,received help from family members or other informal sourcesoutside the health care sector. GP's, medications and occupa-tional physicians were also among the most commonly usedresources across both groups (63% and higher). Psychosocially-geared services were least frequently used (0% to 39%). With theexception of medication use, there were no statistically sig-nificant between-group differences in utilization rates (pb0.05).The mean between-group difference in productivity loss due tototal hours of sick leave was not statistically significant (MISS:mean=451.7 (S.D.=363.2); UC: mean=480.0 (S.D.=403.4);difference: 28.4; 95% CI:− 46.1; 102.89).
3.4. Cost analysis
There were no statistically significant between-groupdifferences for any of the disaggregate or total costs. For
both groups, mean productivity loss costs (MISS=€10,402;S.D.=€6628; UC=€10,643; S.D.=€7169) represented morethan 80% of mean total costs (MISS=€12,538; S.D.=7566;UC=€12,722; S.D.=8173). Higher incremental health caresector costs (€32) and the MISS training costs (€120) for theMISS group were offset by the lower incremental costs relatedto other sectors (−€11), patient/family (−€84) and produc-tivity loss (−€240).
3.5. Cost-utility analysis
The cost-effectiveness plane from the CUA with a max-imum WTP-level of €25,600 is shown in Fig. 2. The MISSoption was less costly and slightly more effective than UC(mean ICER: −€7357; Table 4). For willingness-to-paythresholds from €0 to €100,000 per quality-adjusted lifeyear, the probability of MISS being cost-effectiveness rangedfrom 0.58 to 0.90. The probability of MISS being cost-effectiveat a WTP-level of €25,600 was 0.80 (Fig. 3a).
185K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
3.6. Sensitivity analysis
The acceptability curves corresponding to the threesensitivity analyses involving alternative methods of estimat-ing the productivity loss costs were in line with the mainanalyses (Fig. 3a). The acceptability curve from the completecase analysis followed the same pattern as the main analysisbut indicated lower probabilities of the MISS being cost-effective compared to UC.
3.7. Ancillary analysis
A statistically significant between-group mean differencein QALYs was observed for the SMD subgroup only. None ofthe mean cost differences were statistically significant(Table 4). The bootstrapped cost-effect pairs and meanICERs are presented in Fig. 2. The mean ICER for the SMDsubgroup of −€28,278 was located in southeast quadrant,indicating that the MISS was more effective and less costlythan UC. At the WTP-level of €25,600, the correspondingprobability was 0.98 (Fig. 3b).
4. Discussion
Theminimal intervention forworkerswith stress-related sickleave (MISS) was not associated with superior clinical oreconomic impact than usual GP care for a heterogeneouspopulation with stress-related mental health problems. Theresults of the sensitivity analyses indicated that themainfindingswere reasonably robust. The ancillary analyses suggested that theMISS was dominant compared to UC with respect to the SMDsubgroup, but not for Other MHP and SP. Also, the favourabletrend for MISS at the total group level is likely attributable to theSMD subgroup. However, given that the sample size calculationdid not take a subgroup analysis into consideration and thedetermination of subgroups was conducted a posteriori, weinterpret this finding as hypothesis generating. This findingshould be confirmed in a future clinical trial comparing MISS toUC, in which individuals with a stress-related mental disorder(SMD) are identified a priori. All in all, we do not recommendwidespread implementation of the MISS at this time.
We were unable to directly compare our results to those inthe literature because of a lack of similar studies that comparetwo ormoreGP-based interventions. This underscores the needfor further research to continue to build the evidence base forcost-effective interventions to inform GP clinical managementof workers with stress-related sick leave.
4.1. Strengths and limitations
The main strengths of our study are the pragmatic design,measures taken to prevent bias and the use of a societalperspective for the economic evaluation. First of all, thepragmatic study design involved evaluating the cost-effective-ness of the MISS to UC under ‘real life’ clinical practice versus‘ideal’ circumstances, thereby increasing the generalisability ofour findings. Secondly, to prevent selection bias, patients wererecruited by the researchers instead of the GPs. To preventcontamination, we conducted the randomisation at the level ofGPs instead of patients. This randomisation procedure had anadditional advantage in that it safeguarded GPs from possible
ethical dilemmas related to withholding a potentially beneficialtreatment. Furthermore, GPs were only informed of theirpatients' participation in the study 2-month after enrollment.Thus, during the first two months of treatment when the mosttreatment was provided, knowledge of enrollment in the studycould not have influenced aGP's behaviour. Lastly,we performedthe economic evaluation from the broad societal perspective,which also increased the generalisability of our findings(Drummond et al., 2005).
Methodological limitations relating to a conceptual failure,lack of contrast, poor implementation of the MISS, low powerfor economic outcomes, and estimating costs may explain thenon-significant findings. First, a conceptual failure in definingour study population is unlikely. In general practice, thepatient population is heterogeneous, typically having anadmixture of psychological and somatic symptoms. Currentprimary care diagnostic or dimensional systems to distinguishbetween patients with subacute psychopathology that isclearly related to stress (SMD) from those with an admixtureof symptoms fall short. We expected that patients withdepressive, anxiety or somatic symptoms would still benefitfrom the MISS as a result of more accurate diagnosis andfocused treatment for these problems in addition to attentionfor the stress-related component.
Second, lack of contrast due to usual care not being the sameas usual care in “real life” cannot be ruled out 100%. However,we believe that any difference would be minor. First of all, inaddition to the measures taken to prevent contamination, GPsrandomised to the usual care group did not receive any extrainformation or instruction on treatment, and actual treatmentwas left to their discretion. Furthermore,while it is possible thatthe usual care of GPs in the study may have been influenced byaltered patient behaviour due to patients making a connectionbetween stress and sick leave from the recruitment informationand screening questionnaire, this would have happened aftercontactwith theGPwasalready initiated.Also, becausepatientsin both groups were provided with the same limited informa-tion about the study, such an effect would have occurredsimilarly in both groups.
Third, poor implementation is possible. The training providedto the GPs in the MISS group may not have been sufficient formastery of the desired skills, resulting in inadequate implemen-tation of the intervention. MISS GPs did not significantly differfrom their UC counterparts in terms of advising tasks such asdiscussing the work situation (MISS=74.0%; UC=74.8%),providing advice on the content of work (MISS=41.7%;UC=36.9%), discussion contact with the occupational physician(MISS=48.4%; UC=46.4%), or referral patterns (MISS=43.4%;UC=37.1%). Also, data on the three preplanned subgroupsindicated that elements of theMISSwere implemented to only aminimal extent to the SMD group, and rarely applied to addressthe stress-related components in patients with Other MHP orsomatic problems (data not shown). Implementation may beimproved by more training as well as greater dialogue betweenGPs and occupational physicians in order to establish a strongerculture of collaboration in the care ofworkerswith stress-relatedsick leave.
Fourth, our sample size calculationwas based on detectinga statistically significant effect on the number of workers whoreturned to work full-time after a period of three months, andnot costs. As such, insufficient power to detect relevant
186 K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
differences due to skewed cost data is possible (Briggs, 2000).To avoid magnifying the problem of lack of power andinefficiency due to missing data, we handled missing datausing a multiple imputation procedure and based our mainanalysis on imputed data rather than complete cases (Briggset al., 2003; Oostenbrink and Al, 2005). While multipleimputation is recommended for handling missing data, weacknowledge that it is not a solution to prevent missing datain the first place.
Last, there are two issues related to estimating costs. The firstissue is that while data on GP consultations, medications, labtests and sick leave were collected continuously, the remainingresource use data were collected at four discrete moments.Consequently, we linearly interpolated from 4 months to12 months by assuming a linear time trend in resource use.This may have resulted in an overestimation or underestimationif resourceusebetween twoconsecutivemeasurementmomentswas not linear over time. However, because these costs were asmall proportion of the total costs and the over- or under-estimation may have occurred in both groups, we do not expectthat thiswould alter our overall findings. The second issue is thatwe did not includework-presenteeism or unpaid labor as part ofour assessment of worker productivity (Dewa et al., 2004; Limet al., 2000; Uegaki et al., 2007). Before and after periods of sickleave, work performance and unpaid labor of individuals withstress-relatedmental health problemsmaybe reduced (Brouweret al., 2002; Dewa and Lin, 2000). It is possible that theMISS andUC can have different effects on these components, which werenot captured. Thus, the picture of functional recovery anddifferent effect of the interventions may be incomplete. Futureresearch studies should include these aspects.
4.2. Further considerations
The follow-up period of our trial was 1 year. For clinicalproblems with a multi-year time horizon, a 1-year follow-upmay be insufficient to capture all downstream consequencesof the interventions in question. Extrapolation by means ofdecision models may offer a way to bridge the gap betweenthe observations from trials and what may be expected overthe long term.
QALYs are a standard outcome for economic evaluations froma societal perspective as they allow comparisons betweendifferent disorders, However, generic QALYs — measured forexample by EuroQol questionnaire — may be insensitive forassessing the effects of interventions involving persons withmental health problems (Chisholm et al., 1997). To mitigate thisproblem of generic QALY measures, inclusion of a disease-specific instrument is recommended.
Finally, two practical points should be considered. First,from a clinical standpoint, the observations from the mainand ancillary analyses underscore that (preclinical/mild)mental health problems seen in primary care are hetero-geneous and cannot be treated in the same way. Continuedefforts to improve GPs differential diagnostic skills andprovide focused treatment are warranted. Second, futureintervention trials should pay particular attention toproviding adequate training of professionals in order toensure sufficient mastery of the new treatment method orapproach.
5. Conclusion
The minimal intervention for workers with stress-relatedsick leave (MISS) was not associated with superior clinical oreconomic impact than usual GP care for a heterogeneouspopulation with stress-related mental health problems. Assuch, we do not recommend widespread implementation ofthe MISS at this time. However, the MISS may be a promisingintervention for the subgroup, stress-related mental disor-ders. Future research should aim to confirm this observation.
Role of funding sourceFunding for the randomized controlled trial was obtained from the
Health Research and Development Council (ZonMw) in The Netherlands(Project number 4200.0003). The funding agency did not have any role in thestudy design, data collection, data analysis, interpretation of results,preparation of the manuscript or decision to submit the manuscript forpublication.
Conflict of interestsAll authors declare that they do not have any competing interests and
declare independence from the funders.
References
Anema, J.R., Jettinghoff, K., Houtman, I., Schoemaker, C.G., Buijs, P.C., van den,B.R., 2006. Medical care of employees long-term sick listed due tomentalhealth problems: a cohort study to describe and compare the care of theoccupational physician and the general practitioner. J. Occup. Rehabil. 16,41–52.
Bakker, I.M., Terluin, B., van Marwijk, H.W., Gundy, C.M., Smit, J.H., vanMechelen, W., Stalman, W.A., 2006. Effectiveness of a Minimal Interven-tion for Stress-related mental disorders with Sick leave (MISS); studyprotocol of a cluster randomised controlled trial in general practice[ISRCTN43779641]. BMC Public Health 6, 124.
Black, W.C., 1990. The CE plane: a graphic representation of cost-effectiveness.Med. Decis. Making 10, 212–214.
Briggs, A., 2000. Economic evaluation and clinical trials: size matters. BMJ321, 1362–1363.
Briggs, A., Clark, T., Wolstenholme, J., Clarke, P., 2003. Missing… presumed atrandom: cost-analysis of incomplete data. Health Econ. 12, 377–392.
Brouwer, W.B., van Exel, N.J., Koopmanschap, M.A., Rutten, F.F., 2002.Productivity costs before and after absence from work: as important ascommon? Health Policy 61, 173–187.
Buijs, P., van Amstel, R., van Dijk, F., 1999. Dutch occupational physicians andgeneral practitioners wish to improve cooperation. Occup. Environ. Med.56, 709–713.
Chisholm, D., Healey, A., Knapp, M., 1997. QALYs and mental health care. Soc.Psychiatry Psychiatr. Epidemiol. 32, 68–75.
Council for Public Health and Health Care, 2006. The Hague: Council forPublic Health and Health Care (RVZ).
Dewa, C.S., Lin, E., 2000. Chronic physical illness, psychiatric disorder anddisability in the workplace. Soc. Sci. Med. 51, 41–50.
Dewa, C.S., Lesage, A., Goering, P., Caveen, M., 2004. Nature and prevalence ofmental illness in the workplace. Healthcare Papers 5, 12–25.
Dolan, P., 1997. Modeling valuations for EuroQol health states. Med. Care. 35,1095–1108.
Drummond, M., Manca, A., Sculpher, M., 2005. Increasing the generalizabilityof economic evaluations: recommendations for the design, analysis, andreporting of studies. Int. J. Technol. Assess. Health Care 21, 165–171.
Efron, B., Tibshirani, R.J., 1993. An introduction to the bootstrap. Chapman &Hall, New York.
EuroQol Group,1990. EuroQol—a new facility for the measurement of health-related quality of life. The EuroQol Group. Health Policy 16, 199–208.
Kessler, R.C., Frank, R.G., 1997. The impact of psychiatric disorders on workloss days. Psychol. Med. 27, 861–873.
Koopmanschap, M.A., van Ineveld, B.M., 1992. Towards a new approach forestimating indirect costs of disease. Soc. Sci. Med. 34, 1005–1010.
Lamers, L.M., Stalmeier, P.F., McDonnell, J., Krabbe, P.F., van Busschbach, J.J.,2005. Measuring the quality of life in economic evaluations: the DutchEQ-5D tariff. Ned. Tijdschr. Geneeskd. 149, 1574–1578.
Lazarus, R.S., Folkman, S., 1984. Stress, appraisal and coping. SpringerPublishing Company, New York.
187K. Uegaki et al. / Journal of Affective Disorders 120 (2010) 177–187
Lim, D., Sanderson, K., Andrews, G., 2000. Lost productivity among full-timeworkers with mental disorders. J. Ment. Health Policy Econ. 3, 139–146.
McDaid, D., Curran, C., Knapp, M., 2005. Promoting mental well-being in theworkplace: a European policy perspective. Int. Rev. Psychiatry 17, 365–373.
Murray, C.J.L., Lopez, A.D., 1996. The global burden of disease. A comprehensiveassessment of mortality and disability from diseases, injuries and riskfactors in 1990 and projected. Harvard School of Public Health, Cambridge.
Nieuwenhuijsen, K., Verbeek, J.H., Siemerink, J.C., Tummers-Nijsen, D., 2003.Quality of rehabilitation among workers with adjustment disordersaccording to practice guidelines; a retrospective cohort study. Occup.Environ. Med. 60 (Suppl 1), i21–i25.
Nystuen, P., Hagen, K.B., Herrin, J., 2001. Mental health problems as a cause oflong-term sick leave in the Norwegian workforce. Scand. J. Public Health29, 175–182.
Oostenbrink, J.B., Al, M.J., 2005. The analysis of incomplete cost data due todropout. Health Econ. 14, 763–776.
Oostenbrink, J.B., Bouwmans, C.A.M., Koopmanschap, M.A., et al., 2004. Dutchmanual for costing: methods and standard costs for economic evalua-tions in health care. Health Insurance Council (CVZ), Diemen.
R Development Core Team, 2007. R: A language and environment forstatistical computing. R Foundation for Statistical Computing, Vienna,Austria.
Rubin, D.B., 1987. Multiple imputation for nonresponse in surveys. John Riley& Sons, New York.
Terluin, B., vanMarwijk, H.W., Ader, H.J., de Vet, H.C., Penninx, B.W., Hermens,M.L., van Boeijen, C.A., van Balkom, A.J., van der Klink, J.J., Stalman, W.A.,
2006. The Four-Dimensional Symptom Questionnaire (4DSQ): a valida-tion study of a multidimensional self-report questionnaire to assessdistress, depression, anxiety and somatization. BMC Psychiatry 6, 34.
Terluin, B., van Rhenen, W., Schaufeli, W.B., de Haan, M., 2004. The Four-Dimensional Symptom Questionnaire (4DSQ): measuring distress andother mental health problems in a working population. Work Stress 18,187–207.
Uegaki, K., de Bruijne, M.C., Anema, J.R., van der Beek, A.J., van Tulder, M.W.,van Mechelen, W., 2007. Consensus-based findings and recommenda-tions for estimating health-related productivity loss from a company'sperspective. Scand. J. Work Environ. Health 33, 122–130.
Üstün, T.B., Von Korff, M., 1995. Primary mental health services: access andprovision of care. In: Üstün, T.B., Sartorius, N. (Eds.), Mental Illness inGeneral Health Care: An International Study. John Wiley & Sons,Chichester, pp. 347–360.
Van Buuren S., Oudshoorn C.G.M., 2000. Leiden.van der Klink, J.J., Blonk, R.W., Schene, A.H., van Dijk, F.J., 2003. Reducing long
term sickness absence by an activating intervention in adjustmentdisorders: a cluster randomised controlled design. Occup. Environ. Med.60, 429–437.
van Hout, B.A., Al, M.J., Gordon, G.S., Rutten, F.F., 1994. Costs, effects and C/E-ratios alongside a clinical trial. Health Econ. 3, 309–319.
Wang, J., Adair, C.E., Patten, S.B., 2006. Mental health and related disabilityamong workers: a population-based study. Am. J. Ind. Med. 49, 514–522.
Z-index, 2006. Z-index. G-Standard. The Netherlands, The Hague.
