Vocational Rehabilitation with or without a
Work Module: an Economic Evaluation
S.J. Welling*
Supervisors: Dr. J.O. Mierau and Prof. Dr. M. F. Reneman
Master’s thesis Economics
University of Groningen
11 January 2019
Abstract
Vocational rehabilitation is a widely used treatment to optimize work participation for workers on sick leave due to chronic musculoskeletal pain. The objective of this paper is to evaluate the cost-effectiveness of vocational rehabilitation with or without an additional work module. In this study of 1272 patients, questionnaires from Dutch rehabilitation centers are used to create health care cost variables and to measure the effectiveness of the work module. The work module as addition to vocational reha-bilitation is currently not reimbursed by insurers but the results show that employers funding the treatment can expect a return of over two times their investment directly after the treatment. From a societal point of view, the work module reduces direct and indirect health care costs and improves various effectiveness parameters. Therefore, it can be concluded that the work module is a cost-effective treatment.
Keywords: economic evaluation, vocational rehabilitation, work module, chronic
mus-culoskeletal pain
JEL-codes: I110, I120, G220
*s.j.welling@student.rug.nl
1
Introduction
In the Netherlands one of the most frequent causes of work inability is chronic pain with a prevalence of around 18 percent (Bekkering et al., 2011). Work inability places a large burden on both patients and on the Dutch economy through direct costs like medical expenses and indirect costs which arise from sick days leave and productivity losses (Picavet & Schouten, 2003). The economic burden of chronic musculoskeletal pain was estimated at AC3.5 billion in 2007 and is inspected to increase as the number of chronic musculoskeletal pain patients is expected to rise with 7 percent by 2025 (Lambeek et al., 2007). For employers, health insurance companies and the government it is worthwhile to seek for possibilities to reduce these costs. Economic evaluations must be performed to decide which possible treatment decreases costs the most.
The economic evaluation of health care treatments is relevant according to Drummond, Sculpher, Claxton, Stoddart, and Torrance (2015) because health care funding resources are scarce. Cost-effectiveness analysis weighs both outcomes to value health services and can be used as a framework to consider the quality of different treatments. Palmer et al. (2011) address the need to focus on studying the cost-effectiveness of interventions to improve vocational outcomes. These cost-effectiveness studies are most likely to come from employers who need evidence of a return on investment. Palmer et al. (2011) state that focusing on low-cost measures with workplace elements is unlikely to do harm and has a good chance of proving cost-effective.
Sub-health care costs, absenteeism and productivity losses. A positive return on investment pro-vided by the work module would increase the chance of employers funding this treatment. Therefore, the following research question is formulated:
Does the addition of a work module to vocational rehabilition have a positive return on investment?
The outcomes of this study could prove relevant to employers, health care insurers and society as it provides information on the cost-effectiveness of the work module and may influence policy advice.
2
Literature
Vocational rehabilitation is a multi-professional approach that is provided to individuals of working age with health-related impairments, limitations, or restrictions with work func-tioning and whose primary aim is to optimize work participation (Escorpizo et al., 2011). Vocational rehabilitation differs from usual care by general practitioners as the assessment will be performed by a multidisciplinary team including a psychologist, physiotherapist and occupational specialist. A vocational rehabilitation program in the Netherlands lasts four-teen weeks with approximately two sessions of four hours per week, which amounts to 100 hours in total. Sullivan (2000) demonstrated that vocational rehabilitation is an effective addition compared to usual treatments that were common years ago.
van Bennekom, Frings-Dresen, & Reneman, 2015).
Previous research on this topic by Norlund, Ropponen, and Alexanderson (2009) demon-strated that vocational rehabilitation for patients with chronic musculoskeletal pain is a cost-effective intervention, but is not yet validated in the Dutch situation. There is broad consensus on the effectiveness of vocational rehabilitation for patients with chronic mus-culoskeletal pain, which arises when acute pain to muscles and bones is still present after 12 weeks (Lambeek et al., 2007). However, the costs are higher as additional resources are utilized compared to usual treatments. In addition, there is no agreement on the cost-effectiveness of a work module supplementary to the multidisciplinary approach.
According to Waddell, Burton, and Kendall (2008), vocational rehabilitation is more effective if it is linked to the workplace. Hoefsmit, Houkes, and Nijhuis (2012) contribute to the idea that a work module is effective in returning to work earlier. If the workplace has a negative influence on the patient, alterations to this environment are likely to increase the chances of earlier return to work. Van Beurden, Vermeulen, Anema, and van der Beek (2012) provide evidence on the benefits of a participatory return-to-work program in the Netherlands. The program resulted in a double rate of return to work and greater social benefit of AC2073 per worker compared to usual care. On the other hand, van Vilsteren et al. (2015) dispute evidence of the effect workplace adaptations have on the time of return to work. However, they do note that productivity losses decrease if the source of pain is work-related.
3
Data
around fourteen weeks later at discharge (T2). This treatment period is equal for patients with vocational rehabilitation and for those with the additional work module. Furthermore, after the treatment there were follow-up measurements at six months after discharge (T5) and twelve months after discharge (T8). Most questions were asked with a maximum of four weeks recollection in order to reduce the risk of recall bias (Beemster, van Velzen, van Ben-nekom, Reneman, & Frings-Dresen, 2018). Each of the different questionnaires were filled in at every time point. See figure 1 for the questionnaire timeline.
Figure 1: Timeline of questionnaires
Week: Measurements: −4 0 T0 10 14 T2 36 40 T5 62 66 T8
The first questionnaire requires patients to report data on demographic characteristics, quality of life, work ability, productivity and work situation. The Institute for Medical Tech-nology Assessment (Bouwmans et al., 2014) developed the second and third questionnaires on health care consumption and costs that arise from sick days leave (absenteeism) and productivity losses (presenteeism). Beemster et al. (2018) adjusted the original versions to be better suited for vocational rehabilitation. The Treatment Inventory of Costs in Patients with psychiatric disorders (TIC-P) measures direct health care consumption (Bouwmans et al., 2013). Similarly, the productivity cost questionnaire is adjusted for Vocational Rehabil-itation (iPCQ-VR) and measures absenteeism and presenteeism (Bouwmans et al., 2014).
work ability scores ranging from 0 to 10, where a score of 5 means that a patient is only for 50% productive at work. Both absenteeism and presenteeism variables are then multiplied by the productivity value which equals the average hourly wage in the Netherlands and amounts to AC31.60 for women and AC37.90 for men (Hakkaart-van Roijen, Van der Linden, Bouwmans, Kanters, & Tan, 2015). National averages are provided by the statistics monitor for the working environment (Hooftman et al., 2018). They can be used as the dataset is fairly representative for the Dutch population in terms of education as the main predictor of wages, shown in the appendix in table 6 and table 7. The direct health care costs are constructed by multiplying the utilization of a specific form of health care by its reference price. These reference prices are provided by the Dutch Institute of Health Care (Hakkaart-van Roijen et al., 2015) and shown in table 8 in the appendix. The price multiplied by the number of consultations sum to an aggregate medical consumption amount.
A total of 1272 patients reported from 17 November 2014 until 30 July 2018 on all three questionnaires. Table 1 shows the descriptive statistics at the start of the treatment and is divided into a column of patients who followed vocational rehabilitation (VR) and a column of patients complemented with a work module (VR+WM).
Table 1: Descriptive statistics at baseline
All VR VR+WM
Mean Obs Mean Obs Mean
Male 38.7% 286 46.2% 733 36.8% Age (years) 45.80 (10.83) 283 44.18 (11.44) 721 46.52 (10.43) Education 2.028 (0.730) 272 1.871 (0.745) 690 2.087 (0.720) EQ5D 0.585 (0.254) 289 0.570 (0.265) 738 0.600 (0.247) PDI4 6.888 (2.246) 288 6.917 (2.201) 736 6.879 (2.209) WAI 3.831 (2.416) 241 3.905 (2.399) 719 3.860 (2.396) Work time (hours) 19.27 (10.61) 139 20.70 (11.62) 399 18.77 (10.27) Medical Costs (AC) 527.0 (402.9) 285 477.0 (370.8) 732 539.1 (372.0) Absenteeism (AC) 2303 (2571) 240 2251 (2907) 713 2276 (2393) Presenteeism (AC) 1407 (731) 157 1618 (1278) 438 1379 (1147)
Male is a dummy variable that scores a 0 if the patient is a woman and 1 if the patient is a man. The educational level is divided into the categories low (1), medium (2) and high (3). The mean score of around 2 indicates a medium level of education, see table 6 in the appendix for more information. EQ5D measures quality of life by using the EuroQol five dimensions questionnaire and calculates a score using the method by Lamers, Stalmeier, McDonnell, Krabbe, et al. (2005). The scores range from -0.329 to 1, a negative score means that quality of life is considered worse than death and a score of 1 means quality of life is the best possible. PDI4 is the work-specific item of the pain-disability index (PDI), a score of 0 translates to no disability and 10 to maximum disability (Soer, Reneman, Vroomen, Stegeman, & Coppes, 2012). Work ability is assessed by the work ability index (WAI), the current work ability compared to lifetime’s best work ability can be scored on a 0 to 10 response scale, where 0 represents “completely unable to work” and 10 represents “work ability at its best” (Ahlstrom, Grimby-Ekman, Hagberg, & Dellve, 2010). Work time is the number of self-reported hours a patient worked on average per week during the last month. Inspecting the change in these variables from baseline (T0) to discharge (T2) and beyond (T5,T8), it is possible to compare the group of patients who underwent vocational rehabili-tation versus those who had the work module. The effect parameters are displayed in figure 2 and the costs in 3.
Figure 2: Mean effects per patient
(a) VR 0 2 5 8 0 20 40 60 80 100 Observations P ercen tage (%)
EQ5D PDI4 WAI Work time 0 20 40 60 80 100 W ork time (hours) (b) VR+WM 0 2 5 8 0 20 40 60 80 100 Observations P ercen tage (%)
The effect parameters (EQ5D, PDI4 and WAI) are displayed as a percentage of their maximum attainable score on the left y-axis, the right y-axis shows the number of hours worked per week for the work time variable. It becomes clear that for the patients treated with the work module the line is steeper between T0 and T2 than for the vocational re-habilitation group. The percentage change between baseline and treatment is thus higher for the work module group than for the stand-alone vocational rehabilitation. This might suggest that the work module is more effective directly after the treatment. In the follow-up period after T2, the VR group seems to increase more than the VR+WM group which even decreases for some parameters at T8. The evolution of the costs in figure 3 follows a similar pattern to the mean effects. Moreover, medical costs are not measured at T5.
Figure 3: Mean costs per patient
(a) VR 0 2 5 8 0 500 1,000 1,500 2,000 2,500 Observations Euro (AC)
Medical Absenteeism Presenteeism
(b) VR+WM 0 2 5 8 0 500 1,000 1,500 2,000 2,500 Observations Euro (AC)
Medical Absenteeism Presenteeism
4
Methodology
Vocational rehabilitation is extended by the work module to evaluate if it adds value and whether it is worth the extra costs. The primary outcome of interest is whether there is a statistical relationship between the work module and these costs and effects. Next, the monetary value of this relationship is estimated by using a return on investment metric in accordance with van Dongen et al. (2014). This information is necessary as employers funding the treatment are more likely to pay for the work module if it is financially viable. Subsequently, the point of view of insurers and society are included to advise policymakers on whether it is possible to include the work module in the basic insurance package.
4.1
Statistical analysis
The work module is regressed as a dummy variable on the level of costs to test whether there is a relationship. The effect of the work module is evaluated for all perspectives separately as relationships between the work module and the level of costs might differ per point of view. Moreover, the work module treatment is regressed at moments after discharge to look for the impact it has on the level of costs per patient. The follow-up measurements are used as the effects of the work module might take some time to process and translate to a change in costs. The general regression format will look as follows for all three perspectives (i = employer, insurer, society) and evaluation moments (t = T2, T5, T8):
COST Sit = β0it+ β1it ∗ W M + it (1)
The variable WM is a dummy that scores 0 if no work module has been undertaken and 1 if a patient is treated with the work module. Our parameter of interest: the β1it coefficient
measures the stand-alone effect the work module has on the costs. The constant β0it is the
intercept of the regression. The error-term it includes all determinants of the dependent
4.2
Return on investment
In this subsection, the monetary value of the tested relationship is calculated using a return on investment metric. The cost variables described differ from those in other sections, as only monetary expenditures for the treatment are now considered costs. The benefits per perspective are now constructed by using direct medical costs, indirect costs and total costs. Kok, Houkes, Niessen, et al. (2008) argue that a decrease in health care consumption before and after treatment can be considered a benefit to an insurer. If a patient becomes more productive after applying the work module, the indirect costs decline, which benefits the employer (Tuomi, Huuhtanen, Nykyri, & Ilmarinen, 2001). Whenever the aggregate decline in both direct costs and indirect costs together is positive, society benefits as well.
In the Netherlands, health care insurers reimburse AC5000 for vocational rehabilitation as it is considered health care and included int eh basic insurance package. Contrary, work module interventions are not reimbursed, even though they might be effective and its costs are thus solely a burden faced by employers. However, if the work module is effective and reduces the medical costs consumed by patients, this benefits insurers as they do not have to reimburse these costs anymore. Employers have to pay a fixed fee of AC1250 for work module treatments. The benefit is the increase in value employees add to the firm, related to the treatment effect of the work module. From a financial perspective, employers find it worthwhile to fund the work module supplementary to vocational rehabilitation if the benefits exceed the costs. The societal perspective sums the insurance perspective and the employer perspective. Thus, AC5000 + AC1250 are the total costs ofAC6250 for the treatment. The benefits consist of a decrease in medical spending as well as a decrease for absenteeism and presenteeism costs. The metric in equation (2) expresses the return on investment (ROI) in percentages.
ROI = (Benef its − Costs)
As mentioned previously, the surveys are designed with one month to re-call, so between the start and the end of the treatment after 14 weeks exists a gap of 10 weeks without data on absenteeism and presenteeism. Moreover, until the next evaluation point at 6 months follow-up there are 21 weeks. Similarly, before the final evaluation after 12 months, there are again 21 weeks. To calculate the aggregate benefits of workplace intervention a linear interpolation method is used to fill these gaps of missing data by using the known data observations. Equation (3) expresses the interpolated costs for the gap period (t) by using the previous period known costs (T − 1) and the next period known costs (T + 1) for the respective periods:
C(t) =
C(T −1)∗ [W eek(T +1)− W eek(t)] + C(T +1)∗ [W eek(t) − W eek(T −1)]
[W eek(T +1)− W eek(T −1)]
(3)
Figure 4 shows an example for a medical cost consumption of AC2000 at T0 and AC1500 at T2. This translates to AC1857 after four weeks using equation (3) and gradually changes during the ten-week gap between the two observations, as can be seen in figure 4. For the period between week eight and ten, the duration is only two weeks before the next costs are known, hence the calculated costs from (3) are divided by two.
Figure 4: Linear interpolation example Costs: Week: -4 0 AC2000 4 AC1857 8 AC1714 AC821 AC1500 10 14 Benefit: AC143 AC284 AC178 AC495
et al. (2015). Between week four and week eight the interpolated costs amount to AC1714, the difference of AC286 is then discounted toAC284.
Aggregating these benefits for all evaluation periods and using the respective fees per perspective in the return on investment metric calculates the monetary value of the work module. For the calculation of the monetary benefits the human capital approach is used which calculates productivity losses using the method described in the data section. Alter-natively, the friction cost method assumes workers on sick leave can be replaced, leading to lower productivity losses. See table 9 in the appendix for a further discussion.
5
Results
Starting with the regression from equation (1), table 2 shows the test results of the relation-ship between the work module and the costs per perspective and evaluation moment.
Table 2: General regression results work module
Variable Coefficient Constant
Medical Costs T2 -29.67 (32.67) 305.1*** (27.72) Medical Costs T8 66.96 (59.18) 217.3*** (51.97) Indirect Costs T2 -554.9*** (209.6) 1895*** (180.5) Indirect Costs T5 -259.1 (301.7) 1411*** (264.8) Indirect Costs T8 721.1** (334.0) 522.4* (298.3) Total Costs T2 -361.8* (198.4) 1889*** (168.2) Total Costs T8 701.7** (297.6) 592.1** (262.1)
Standard errors are between parentheses. ***, ** and * denote the statistical significance at the 1, 5 and 10 percent levels respectively
complemented with the work module the level of indirect cost will be AC554.9 lower, keeping all else equal. At the first follow-up measurement, this effect is not significant anymore. However, at T8 the sign has changed which means the level of indirect costs increases with AC721 a year after the treatment for employers funding the work module. This finding might be explained by the lower level of observations and thus less power in the test. This issue will be addressed in the further statistical analysis. For the societal perspective, the level of total costs at T2 is significant at the 10 percent level. This means that patients using the work module put a lower financial burden on society. The total costs at T8 are positive again similar to the indirect costs.
5.1
Return on investment
Only the significant regression results at discharge are included for calculating the return on investment. The insurer perspective and evaluation moments after discharge are not taken into account.
Table 3: Return on investment at discharge
Benefit Return on investment Work Module Employer AC4197 (301.7) 235.8% (24.14) No Work Module Employer AC2378 (574.1)
-Work Module Society AC4758 (306.5) -23.87% (4.905) No Work Module Society AC2809 (523.2) -43.82% (10.46)
Standard deviations are between parentheses.
vocational rehabilitation alone. Investing in the work module thus places a lower burden on society.
6
Cost-effectiveness analysis
To assess the cost-effectiveness of the work module, the incremental cost-effectiveness ratio (ICER) provides guidance on whether the effects are worth the opportunity costs. Including the work module in an insurance package means scarce resources will no longer be available to benefit other patients and treatments. Therefore, the treatment must be sufficiently effective to be worth the opportunity costs otherwise available for other treatments.
Quality adjusted life years (QALYs) is the main variable of interest in health economic decision making as an effect parameter. The EQ-5D score, as indicator for quality of life, is adjusted following the method of Prieto and Sacristán (2003). The difference in EQ-5D score between baseline and discharge shows if a patient’s quality of life improved during the treatment period. The number of QALYs gained is calculated by multiplying the 14 week treatment period with the change in quality of life and dividing by 52 weeks as QALYs are measured in years. The costs in this section differ from previous sections. They are evaluated from a societal perspective and include the direct medical costs, as well as the indirect costs from absenteeism and presenteeism. Moreover, the societal treatment fee of AC5000 for vocational rehabilitation and AC6250 for vocational rehabilitation with the work module are added. The difference in the total costs between the start and the end of the treatment is used, together with the gain in QALYs to calculate the ICER in equation (4) to compare both treatments. The ICER is calculated using 136 observations from the vocational rehabilitation group and 361 observations from the work module group.
ICER = ∆COST S
∆EF F ECT S =
∆COST SV R+W M − ∆COST SV R
∆EF F ECT SV R+W M − ∆EF F ECT SV R
In equation (4) ∆Costs is the difference between the change in total costs from the vocational rehabilitation with and without workplace interventions. ∆Effects is the difference in QALY change between both treatments.
6.1
Cost-effectiveness plane
A single ICER as an outcome measure is not sufficient for decision making due to the uncertainty regarding the representativeness of the dataset for the whole population (Briggs, Wonderling, & Mooney, 1997). Therefore, a statistical method known as bootstrapping is used to replicate the sample in a random fashion and increase the accuracy of inference. The concept behind bootstrapping is to treat the study sample as if it were the population. According to Campbell and Torgerson (1999) it is better to draw inferences from the sample rather than to make potentially unrealistic assumptions about the underlying population. Thereby, bootstrapping improves the probability of sound economic decision making from a societal perspective.
After bootstrapping the sample observations, the ICERs are graphed in a cost-effectiveness plane. This figure provides guidance on whether the work module is cost-effective. This plane plots the costs of the treatment against the effects. The cost-effectiveness results in figure 5 are generated by bootstrapping the input for the incremental cost effectiveness ratio with 1000 replications.
Figure 5: Cost-effectiveness plane
6.2
Cost-effectiveness acceptability curve
The monetary value per QALY remains debatable until today and there is no evident thresh-old to assess cost-effectiveness (Neumann, Cohen, & Weinstein, 2014). However, by using multiple threshold values the cost-effectiveness plane can still provide useful insight. If the bootstrapped observations are located mainly below the specified threshold values, the work module is relatively more effective than costly. For summary measures, a cost-effectiveness acceptability curve plots the probability of the treatment being cost-effective against the range of threshold values per QALY. To assess whether the extra quality of life-effects is worth the additional costs, a cost-effectiveness acceptability curve (CEAC) is plotted in figure 6. This graph is suggested by Fenwick, O’Brien, and Briggs (2004) to evaluate the cost-effectiveness plane for the different quality of life thresholds. The y-axis describes the probability the work module is cost-effective against the willingness to pay threshold per quality of life effect.
Figure 6: Cost-effectiveness acceptability curve
is generated by calculating the net benefits of the work module from the non-bootstrapped sample mean, using the method of Hounton and Newlands (2012). Subsequently, standard errors of the net benefits are calculated and assuming a normal distribution probabilities are assigned, as suggested by Van Hout, Al, Gordon, and Rutten (1994).
If the ceiling ratio is AC10.000 it is not likely the work module is cost-effective as this probability is lower than 20 percent. At a suggested threshold ofAC80.000 in the Netherlands “Council for Public Health and Health Care” (2006) this probability increases to 87.4% for the non-parametric approach and 91.7% for the parametric curve.
7
Further statistical analysis
7.1
Multiple imputation
As the questionnaires suffer from an increasing level of non-response after discharge, the data may not be missing completely at random. This issue is addressed by applying multiple imputation as suggested by Vroomen et al. (2016). This decline in response can be explained, as treated patients might not feel the urge anymore to fill in questionnaires. To overcome this problem, multiple imputation is used to fill in missing values multiple times based on available other information of the patient. As suggested by Azur, Stuart, Frangakis, and Leaf (2011) multiple imputation by chained equations is used to estimate missing values on the basis of other patient information. This information comes from the variables with the least missing values, see table 10 in the appendix. The predictive mean matching method for continous variables can be a helpful tool for dealing with the skewed distribution of cost data. Predictive mean matching preserves the distribution of the data and is robust against violations of the normality assumption (Van Buuren, 2018).
7.2
Regression
After imputing the data and increasing the number of observations, control variables are needed to account for the different demographic features of patients. The group of patients who underwent vocational rehabilitation might differ from the work module group. Adding control variables to the general regression removes the effect the control variables have on the change in costs from our β1 coefficient of interest. These adaptations make the regression in
equation (5) statistically more robust and lead to better inference. The regression tests the treatment effect of the work module for all three perspectives (i= employer, insurer, society) following the difference-in-difference method of Angrist and Pischke (2008).
In this regression, ∆COST S is defined as CostsT 2− CostsT 0 to control for the different
levels of costs at baseline and only test the effect the work module has on the change in costs at discharge. WM is again the dummy variable indicating the presence of the work module. The control variable MALE describes gender and is included to account for the fact that men and women might have different direct or indirect costs. Consequently, gender might influence the choice for the work module. The AGE variable copes with the fact that older people are likely to consume more direct health due to an increase in health problems related to age. EDU is another control for the highest level of education patients completed, it is included as research by (Ross & Wu, 1995) has shown that a lower level of education relates to a higher level of direct medical costs and indirect costs like absenteeism and presenteeism. The effectiveness of the treatment is measured by performing sensitivity analyses with several outcome measures. The effect parameters that are evaluated are the EQ-5D score, PDI, WAI and the hours of working time per week. The same control variables are included as the cost regression. At first, the EQ-5D score is used as a general metric for quality of life. Secondly, item four of the pain-disability index (PDI4) is applied as a specific indicator for the work disability. Thirdly, the work ability score (WAI) to measure productivity at work. Finally, the hours of working time per week are used as a useful indicator for employers. The regression in equation (6) to test for the significance of the effect parameters (i= EQ5D, PDI4, WAI, Work time) has the same structure as equation (5).
∆EF F ECT Si = β0i + β1i ∗ W M + β2it∗ GEN + β3i ∗ AGE + β4i ∗ EDU + i (6)
Here, ∆EF F ECT S is defined as EF F ECT ST 2− EF F ECT ST 0. The regression results
of the more sophisticated regressions are separately shown in table 4 and table 5.
Table 4: Costs per perspective between T1 and T2
Variables Direct costs Indirect costs Total costs Work module 62.64 (40.46) 657.9** (350.5) 718.7* (366.3) Age -0.028 (1.371) -13.47 (9.797) -13.34 (9.921) Education -2.023 (9.719) -66.71 (71.64) -68.34 (72.84) Male 20.55 (32.77) 930.9*** (219.0) 952.3*** (225.8) Constant 207.9** (86.79) 2130*** (678.6) 2329*** (688.2) Observations 1195 1182 1182 Largest FMI 0.443 0.662 0.676 Imputations 10 10 10
Standard errors are between parentheses. ***, ** and * denote the statistical significance at the 1, 5 and 10 percent levels respectively. FMI is the fraction of missing information.
there exists an effect between the change in direct costs and the age, educational level and gender of the patient in this regression. The work module has a significant relation at the five percent level to the change in total indirect costs between baseline and discharge. The coefficient of 657.9 means the work module is successful in explaining indirect costs. It can be interpreted as a AC657.9 difference in the level of absenteeism and presenteeism between the control and treatment group evaluated at discharge. In this regression, gender also has a highly significant influence on the change in cost. The positive coefficient with a value of 930 means that a male has a much higher change in indirect costs than a woman. The total costs estimation aggregates both direct and indirect costs and repeatedly confirms the positive relationship. The work module is significant at the ten percent level and has a AC718.7 higher impact than the vocational rehabilitation treatment. Furthermore, gender has a positive relationship and confirms the greater effect for men as explained in the previous regression.
Table 5: Effects per parameter between T1 and T2
Variables EQ5D PDI-4 WAI Worktime
Work module 0.051* (0.276) 0.670* (0.363) 0.618** (0.238) -0.465 (1.151) Age -0.004*** (0.001) -0.015 (0.012) 0.003 (0.011) -0.027 (0.059) Education -0.008 (0.006) 0.002 (0.064) 0.004 (0.067) 0.397 (0.359) Male -0.014 (0.018) 0.134 (0.226) 0.104 (0.196) 0.756 (1.284) Constant 0.305*** (0.058) 2.030*** (0.611) 0.982*** (0.745) 0.119 (3.878) Observations 1200 1,200 1,190 1,184 Largest FMI 0.628 0.738 0.625 0.814 Imputations 10 10 10 10
Standard errors are between parentheses. ***, ** and * denote the statistical significance at the 1, 5 and 10 percent levels respectively. FMI is the fraction of missing information.
in EQ5D. Probably because older patients have a lower quality of life to start with due to age-related health issues. The Pain disability index related to the work environment is significant at the ten percent level, meaning the work module is effective in decreasing the experienced level of pain at the workplace. The coefficient of 0.670 shows the impact is quite large considering that the PDI4 scale ranges from 0 to 10. The work ability score measured by the WAI variable indicates that the patients in the treatment group face an increase in work ability 0.618 points more than the control group on a 10 point scale. Finally, the number of hours of working time per week are regressed as the dependent variable. However, none of the tested variables seem to causally relate to the change in hours of time worked.
8
Discussion
have to recall their medical expenses for the last month in the survey it is likely they for-get some appointments (Battistin, 2003). Including these forgotten medical expenses would result in higher costs.
The creation of direct cost variables is somewhat influenced by the fact that the used reference prices might not be the true costs individuals pay for their medical expenses. The calculation of presenteeism for evaluative purposes is put to question by Beemster et al. (2018). Moreover, the average hourly wage for men and women are used to construct indirect cost variables. Ideally, patients should provide their hourly wage or at least a distinction should be made per occupation type. This distinction could help deciding whether the human capital approach or the friction cost method is best applied. Some employers might find it straightforward to replace their workers and can use a short friction period for evaluative purposes. For employers who can not easily replace their workers, the human capital approach is more suitable.
Another issue is that the group of patients is not randomly divided between the control group that receives vocational rehabilitation and the treatment group that additionally re-ceives the work module. This decision depends on the willingness of employers to fund the work module and therefore leads to a selection bias. Employers funding the work module might be more cooperative in the treatment process, making the treatment more successful. A complete randomized controlled trial solves this issue.
9
Conclusion
times their investment. Therefore, the results suggest that it is beneficial for employers to fund the work module if one of their employees requires vocational rehabilitation. Combining the employer perspective and the insurer perspective, the work module decreases the level of costs directly after treatment and increases various effect parameters like the quality of life. Therefore, the work module is a cost-effective treatment from a societal point of view.
10
References
Ahlstrom, L., Grimby-Ekman, A., Hagberg, M., & Dellve, L. (2010). The work ability in-dex and single-item question: Associations with sick leave, symptoms, and health–a prospective study of women on long-term sick leave. Scandinavian journal of work,
environment & health, 404–412.
Angrist, J. D. & Pischke, J.-S. (2008). Mostly harmless econometrics: An empiricist’s
com-panion. Princeton university press.
Arboned. (2016). Activiteiten- en tarievenoverzicht 2016. Retrieved from https : / / www . arboned.nl/uploads/inline/ArboNed_Vrij_Tarieven_2016.pdf
Azur, M. J., Stuart, E. A., Frangakis, C., & Leaf, P. J. (2011). Multiple imputation by chained equations: What is it and how does it work? International journal of methods
in psychiatric research, 20 (1), 40–49.
Battistin, E. (2003). Errors in survey reports of consumption expenditures. IFS Working Papers, Institute for Fiscal Studies (IFS).
Beemster, T. T., van Velzen, J. M., van Bennekom, C. A., Frings-Dresen, M. H., & Reneman, M. F. (2015). Cost-effectiveness of 40-hour versus 100-hour vocational rehabilitation on work participation for workers on sick leave due to subacute or chronic musculoskeletal pain: Study protocol for a randomized controlled trial. Trials, 16 (1), 317.
Beemster, T. T., van Velzen, J. M., van Bennekom, C. A., Reneman, M. F., & Frings-Dresen, M. H. (2018). Test–retest reliability, agreement and responsiveness of productivity loss (ipcq-vr) and healthcare utilization (ticp-vr) questionnaires for sick workers with chronic musculoskeletal pain. Journal of occupational rehabilitation, 1–13.
Bekkering, G., Bala, M., Reid, K., Kellen, E., Harker, J., Riemsma, R., . . . Kleijnen, J. (2011). Epidemiology of chronic pain and its treatment in the netherlands. Neth J
Med, 69 (3), 141–153.
Bouwmans, C., De Jong, K., Timman, R., Zijlstra-Vlasveld, M., Van der Feltz-Cornelis, C., Tan, S. S., & Hakkaart-van Roijen, L. (2013). Feasibility, reliability and validity of a questionnaire on healthcare consumption and productivity loss in patients with a psychiatric disorder (tic-p). BMC health services research, 13 (1), 217.
Bouwmans, C., Krol, M., Brouwer, W., Severens, J., Koopmanschap, M., & Hakkaart, L. (2014). Imta productivity cost questionnaire (ipcq). Value Health, 7 (17), A550. Briggs, A. H., Wonderling, D. E., & Mooney, C. Z. (1997). Pulling cost-effectiveness analysis
up by its bootstraps: A non-parametric approach to confidence interval estimation.
Health economics, 6 (4), 327–340.
Campbell, M. & Torgerson, D. (1999). Bootstrapping: Estimating confidence intervals for cost-effectiveness ratios. Qjm, 92 (3), 177–182.
Council for Public Health and Health Care. (2006). Sensible and sustainable care (in Dutch). Retrieved from https : / / www . raadrvs . nl / documenten / publications / 2006 / 6 / 27 / sensible-and-sustainable-care
Drummond, M. F., Sculpher, M. J., Claxton, K., Stoddart, G. L., & Torrance, G. W. (2015).
Methods for the economic evaluation of health care programmes. Oxford university
Escorpizo, R., Reneman, M. F. [Michiel F], Ekholm, J., Fritz, J., Krupa, T., Marnetoft, S.-U., . . . Stucki, G., et al. (2011). A conceptual definition of vocational rehabilitation based on the icf: Building a shared global model. Journal of occupational rehabilitation,
21 (2), 126–133.
Fenwick, E., O’Brien, B. J., & Briggs, A. (2004). Cost-effectiveness acceptability curves–facts, fallacies and frequently asked questions. Health economics, 13 (5), 405–415.
Hakkaart-van Roijen, L., Van der Linden, N., Bouwmans, C., Kanters, T., & Tan, S. S. (2015). Kostenhandleiding. Methodologie van kostenonderzoek en referentieprijzen voor
economische evaluaties in de gezondheidszorg. In opdracht van Zorginstituut Nederland. Geactualiseerde versie.
Hoefsmit, N., Houkes, I., & Nijhuis, F. J. (2012). Intervention characteristics that facili-tate return to work after sickness absence: A systematic literature review. Journal of
occupational rehabilitation, 22 (4), 462–477.
Hooftman, W., Mars, G., Janssen, B., De Vroome, E., Ramaekers, M., & van den Boss-che, S. (2018). Nationale enquête arbeidsomstandigheden 2017. methodologie en globale
resultaten. TNO; CBS.
Hounton, S. & Newlands, D. (2012). Applying the net-benefit framework for assessing cost-effectiveness of interventions towards universal health coverage. Cost Effectiveness and
Resource Allocation, 10 (1), 8.
Kok, L., Houkes, A., Niessen, N., et al. (2008). Kosten en baten van revalidatie. Amsterdam:
NL: SEO Economisch Onderzoek.
Lambeek, L. C., Anema, J. R., van Royen, B. J., Buijs, P. C., Wuisman, P. I., van Tulder, M. W., & van Mechelen, W. (2007). Multidisciplinary outpatient care program for patients with chronic low back pain: Design of a randomized controlled trial and cost-effectiveness study [isrctn28478651]. BMC public health, 7 (1), 254.
Lamers, L., Stalmeier, P., McDonnell, J., Krabbe, P., et al. (2005). Measuring the qual-ity of life in economic evaluations: The dutch eq-5d tariff. Nederlands tijdschrift voor
geneeskunde, 149 (28), 1574–1578.
Neumann, P. J., Cohen, J. T., & Weinstein, M. C. (2014). Updating cost-effectiveness–the curious resilience of the $50,000-per-qaly threshold. New England Journal of Medicine,
371 (9), 796–797.
Norlund, A., Ropponen, A., & Alexanderson, K. (2009). Multidisciplinary interventions: Review of studies of return to work after rehabilitation for low back pain. Journal of
rehabilitation medicine, 41 (3), 115–121.
Palmer, K. T., Harris, E. C., Linaker, C., Barker, M., Lawrence, W., Cooper, C., & Coggon, D. (2011). Effectiveness of community-and workplace-based interventions to manage musculoskeletal-related sickness absence and job loss: A systematic review.
Rheuma-tology, 51 (2), 230–242.
Picavet, H. & Schouten, J. (2003). Musculoskeletal pain in the netherlands: Prevalences, consequences and risk groups, the dmc3-study. Pain, 102 (1-2), 167–178.
Prieto, L. & Sacristán, J. A. (2003). Problems and solutions in calculating quality-adjusted life years (qalys). Health and quality of life outcomes, 1 (1), 80.
Ross, C. E. & Wu, C.-l. (1995). The links between education and health. American
sociolog-ical review, 719–745.
Soer, R., Reneman, M. F., Vroomen, P. C., Stegeman, P., & Coppes, M. H. (2012). Re-sponsiveness and minimal clinically important change of the pain disability index in patients with chronic back pain. Spine, 37 (8), 711–715.
Sullivan, T. J. (2000). Injury and the new world of work. UBC Press. Retrieved from http: //search.ebscohost.com.proxy- ub.rug.nl/login.aspx?direct=true&db=nlebk&AN= 382728&site=ehost-live&scope=site
Tuomi, K., Huuhtanen, P., Nykyri, E., & Ilmarinen, J. (2001). Promotion of work ability, the quality of work and retirement. Occupational medicine, 51 (5), 318–324.
Uwv. (2019). Uwv normbedragen voorzieningen 2019. Retrieved from https://www.uwv.nl/ werkgevers/Images/normbedragen-voorzieningen-2019.pdf
Van Buuren, S. (2018). Flexible imputation of missing data. Chapman and Hall/CRC. 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 economics, 3 (5), 309–319.
van Beurden, K. M., Vermeulen, S. J., Anema, J. R., & van der Beek, A. J. (2012). A participatory return-to-work program for temporary agency workers and unemployed workers sick-listed due to musculoskeletal disorders: A process evaluation alongside a randomized controlled trial. Journal of occupational rehabilitation, 22 (1), 127–140. van Dongen, J. M., van Wier, M. F., Tompa, E., Bongers, P. M., van der Beek, A. J., van
Tulder, M. W., & Bosmans, J. E. (2014). Trial-based economic evaluations in occupa-tional health: Principles, methods, and recommendations. Journal of occupaoccupa-tional and
environmental medicine, 56 (6), 563.
van Vilsteren, M., van Oostrom, S. H., de Vet, H. C., Franche, R.-L., Boot, C. R., & Anema, J. R. (2015). Workplace interventions to prevent work disability in workers on sick leave. Cochrane Database of Systematic Reviews, (10).
Vroomen, J. M., Eekhout, I., Dijkgraaf, M. G., van Hout, H., de Rooij, S. E., Heymans, M. W., & Bosmans, J. E. (2016). Multiple imputation strategies for zero-inflated cost data in economic evaluations: Which method works best? The European Journal of
Health Economics, 17 (8), 939–950.
Waddell, G., Burton, A. K., & Kendall, N. A. (2008). Vocational rehabilitation–what works,
11
Appendix A
Table 6: Education
Highest level completed Observations Percentage Classification
1. No schooling 5 0.40% low
2. Elementary school 17 1.37% low
3. Domestic education 131 10.6% low
4. Secondary education 144 11.6% low
5. Intermediate vocational school 466 37.7 % medium 6. Pre-university school 82 6.63 % medium 7. Higher vocational school 275 22.2 % high
8. University 55 4.45 % high
9. Other 62 5.01%
-Total 1237 100%
The level of education of the dataset is compared to the Dutch national average calcu-lated by Hooftman et al. (2018). Table 7 shows the dataset contains slightly more medium educated individuals but is sufficiently comparable to use mean wages for calculating indirect costs.
Table 7: Education
Classification Dataset NEA
Low 25.28% 21.8%
Medium 46.64% 42.9%
High 28.09% 35.3%
12
Appendix B
Table 8: Reference prices
Reference price
General Practioner Appointments AC33
Consultation visits AC33 Consult by phone AC17 Home visits AC50 Physiotherapy Physiotherapist AC33 Exercise therapist AC34 Manual therapist AC33 Occupational therapist AC33 Company doctor AC199
Reintegration Reintegration advisor AC117
Reintegration expert AC117
Jobcoach AC76
Work expert AC168
Insurance doctor AC208
Medical specialist Outpatient clinic AC91
Hospitalized General Hospital AC443
Academic Hospital AC642 Rehabilitation centre AC460 Psychology Psychologist AC94 Psychiater AC94 Social worker AC65 Dietician AC27
Home care General care AC50
Rehabilitation specific care AC120
Alternative care AC100
13
Appendix C
The Institute of Healthcare (Hakkaart-van Roijen et al., 2015) suggests using a friction cost method which assumes that employers suffer from productivity losses for just twelve weeks or 85 days. The rationale behind this method is that employers can replace their employees within this time frame and do not suffer from productivity losses anymore. So, if patients are sick longer than 12 weeks their benefits beyond this period are not taken into account anymore. Beemster et al. (2015) argue that a friction cost period of 160 days is more suitable. In table 9 a sensitivity analysis for the three approaches is shown.
Table 9: Friction cost method employer perspective
Human capital Friction cost 160 Friction cost 85 Work module benefit AC4197 (301.7) AC2137 (253.4) AC885.6 (166.3) Work module ROI 258.8% (24.14) 70.93% (20.27) -29.15% (13.30) No work module benefit AC2378 (574.1) AC1541 (411.7) AC807.2 (274.3)
No work module ROI - -
-Replaced sick workers 0 300 380
Observations 476 476 476
Standard deviations are between parentheses. ROI is the return on investment metric calculated using equation (2).
Table 10: Imputation of dataset
Variable Observations (T0;T2) Description
Advice 1272 Regular
Distress 1239 Regular
Complaints 1206 Regular
Work pressure 1194 Regular
Arrangements 1194 Regular Dissatisfaction 1192 Regular Uncertainty 1217 Regular Diligence 1215 Regular Home-situation 1223 Regular Rehabilitation-centre 1272 Regular
Work module 1038 Imputed
Education 1237 Imputed Age 1225 Imputed Gender 1243 Imputed WAI 1176;529 Imputed EQ5D 1261;562 Imputed PDI4 1254;542 Imputed
Work time 661;396 Imputed
General Practitioner costs 1241;582 Imputed
Physiotherapy costs 1231;578 Imputed
Company doctor costs 1241;514 Imputed
Reintegration costs 1229;576 Imputed
Insurance doctor costs 1241;580 Imputed Medical specialist costs 1206;563 Imputed
Hospitalized costs 1233;579 Imputed
Psychology costs 1231;579 Imputed
Social worker costs 1230;580 Imputed
Dietician costs 1225;580 Imputed
Home care costs 1233;580 Imputed
Alternative care costs 1229;580 Imputed
Absenteeisme costs 1171;518 Imputed
Presenteeisme costs 731;312 Imputed
