Estimating Productivity Losses in Health
Economics: the Impact of Method
Course code: EBM877A20
Author: Hermine, H. Dijk, s2168251
Supervisors: Jochen Mierau, Femke Hoekstra Date: June 24, 2016
Organisation: Centre of Human Movement Sciences, University Medical Centre Groningen
Abstract
The proper estimation of productivity losses is an important topic of debate in health economic research. Currently, two different approaches, the human capital (HC) approach and the friction cost (FC) approach are widely in use. As the two methods can give extremely different results, the impact of the estimation method on health care decisions can be substantial. Consequently, this paper investigates when either of the two methods should be used, on the basis of an example of the societal cost of Dutch rehabilitation patients during, and shortly after, rehabilitation. When the number of individuals affected by a disease, or treatment, is small, or when the disease only results in short-term absences from work, the FC approach is sufficient for estimating productivity losses. When groups are large, neither of the two approaches provides satisfactory results. Instead, a macroeconomic model with an explicitly modelled health care sector should be used.
1
1. Introduction
The proper method for estimating productivity losses is an important topic of debate in health economic research. Currently, two different approaches, the human capital (HC) approach and the friction cost (FC) approach are widely in use. The first method assumes that an individual’s wage equals that individual’s marginal labour productivity. Consequently, according to the HC approach, societal cost can be calculated by multiplying the hourly wage by the number of hours a disabled/ill worker works less due to the disability/illness for that individual’s entire life up until retirement. On the other hand, the FC method assumes that productivity losses due to a disability or disease are only there until the firm finds another employee to take on the work. In other words, productivity losses only exist during a friction period (Koopmanschap, et al., 1995).
While the FC approach proposed by Koopmanschap et al. (1995) has gained importance and the support of the Dutch guidelines for pharmacoeconomic research (National Health Care Institute, 2015), internationally the debate seems to have reached an impasse. This is illustrated by the fact that other guidelines, such as the Austrian (Walter & Zehetmayr, 2006) and the Italian guideline (Capri, et al., 2001) advise the use of the HC approach, or in the case of the German guideline (IQWiG, 2009) advise that ideally both methods should be used.
As the two methods can give extremely different results (Van den Hout, 2010; Hanly, et al., 2012; Hutubessy, et al., 1999) the impact of the estimation method on health care decisions should not be underestimated. A therapy that is cost-effective with the HC approach might be ineffective when costs are calculated using the FC approach. Consequently, it is not only important from an academic perspective that the debate is settled, but also from a policy perspective.
As the period of estimation is the main difference between the two approaches, the debate is more relevant for cases where long-term absences from work play an important role. As such, individuals with a disability and/or chronic illness are particularly relevant to the debate, as due to the nature of their condition they are often absent from work, or less productive at work, for a period exceeding the friction period. Since many individuals with a disability and/or chronic disease spend time in a rehabilitation facility, ascertaining the societal cost of rehabilitation patients can provide and excellent example of the extent of the influence of the FC and HC estimation methods.
While research on the societal cost of numerous illnesses and disabilities does exist, it is rarely - if ever - focussed specifically on the rehabilitation setting. However, individuals with a disability
2 productivity losses has not yet been solved, a sensible estimate of societal cost for rehabilitation patients should include both approaches.
Therefore, this study provides an estimate for the societal cost of Dutch individuals with a disability or chronic illness during and after rehabilitation and explores the effect of the HC and FC approach on these societal costs. In addition, it aims at providing guidance on when either of the two methods should be used.
The next section will provide a more detailed overview of the arguments in favour and against both methods, as discussed by earlier research. After that, section 3 will describe the data used for this study, whereas section 4 describes the methods used to estimate societal costs and section 5 provides results and uncertainty analyses. In section 6, the limitations of the study will be discussed. In addition, a comparison with the existing literature will be made, as well as recommendations for further research. Finally, section 7 provides a conclusion.
2. Literature
The FC approach was introduced in a paper by Koopmanschap et al. (1995) in which they argue that the FC method is more realistic than the HC approach. The reasons mentioned are that
individuals might be able to catch up on foregone work upon their return, firms could have internal labour reserves, tasks foregone during the absence are probably the less productive ones and if it is clear that the absence or reduced productivity will be long-term, firms could hire a new employee from the unemployed workforce.
However, Johanneson and Karlsson (1997) have argued that the FC method is not grounded in economic theory while the HC approach is. From a neoclassical point of view, this is certainly the case. According to neoclassical theory the marginal labour productivity of an individual equals his/her gross wage. Consequently, if that individual falls ill, the loss to the firm would also equal that
individual’s gross wage, and thus the HC approach would be correct (Johanesson & Karlsson, 1997). However, the HC approach implicitly assumes a world with no unemployment, which seems quite unrealistic.
On the other hand, the friction-cost approach implicitly assumes that any individual filling a vacancy due to sick-leave comes from the ranks of the unemployed. This is not necessarily true. Even if the filling of the vacancy leads to a chain of vacancies being filled and opened, with at the end of the line a formerly unemployed individual gaining a job, this would lead to multiple friction periods. However, the FC method only assumes one friction period, as pointed out by Johanesson and
3 solved by using a longer friction period in the estimations, as even in this case the HC approach would still be an overestimation of true societal cost.
In a partial equilibrium analysis, the FC method could be sensible, as it assumes an endless supply of (previously unemployed) labour against a certain cost, mostly search cost, denoted by the friction period. However, when looking at the larger picture, where one leaves the realm of partial equilibria, or when the analysis involves a very large group of individuals, the HC approach seems more
sensible, as productivity losses will be large enough that labour related parameters become endogenous and consequently long-term absences may have real world consequences for other individuals in society.
In conclusion, while the Dutch guideline advises the use of the FC method, the academic debate is far from settled. While the HC approach is likely to overestimate true societal cost, as it assumes a world with no unemployment, the FC method could very possibly lead to an underestimation, as it assumes that there exists an endless supply of unemployed labour against a certain search cost. Consequently, the FC approach likely constitutes a lower bound, while the HC approach constitutes an upper bound.
3. Data
3.1. The Rehabilitation, Sports and Exercise Programme
This study focused on individuals participating in the Rehabilitation, Sports and Exercise (RSE) programme. The programme, which is currently implemented in 18 hospitals and rehabilitation centres in the Netherlands, aims at stimulating an active lifestyle for individuals with a disability or chronic illness during and after their rehabilitation treatment (Hoekstra, et al., 2014; Alingh, et al., 2015; ReSpAct, sd), and is based on a similar programme that was proven to be effective by Van der Ploeg (Van der Ploeg, 2005). In the RSE programme, sports and exercise are an integrated part of the rehabilitation treatment. In addition, at the end of the treatment patients receive tailored advice on retaining an active lifestyle from a sports counsellor and after discharge patients receive four counselling calls by means of which the counsellor guides the patient in maintaining an active lifestyle after rehabilitation.
4 close to the moment of discharge from rehabilitation, and the second survey (T1) followed 14 weeks after the first. In between T0 and T1, the four counselling calls took place. As the last waves were not available in time for this research, only data from T0 and T1 was used.
3.2. Study population
While most of the variables in the dataset consisted of survey responses filled in by the patients themselves, some general information was provided by the counsellor in charge of a particular patient. This included the individual’s diagnosis, gender, year of birth, treatment form at T0
(outpatient, inpatient, consults only), treatment location and whether an individual suffered from a progressive disease. Diagnoses were classified into eight different groups, based on the classification presented by the rehabilitation industry report (Revalidatie Nederland, 2012). The different groups consisted of: conditions affecting the muscoskeletal system, amputation, brain related conditions, neurological conditions, paraplegia, organ related diagnoses, chronic pain, and miscellaneous. Appendix A gives a detailed overview of which diagnoses belong to which diagnosis group.
Table 1 provides some general information about the study population. As can be derived from
Table 1: Summary statistics
Percent N Percent N
Age Diagnosis group
18-44 30.7% 528 Muscoskeletal 18.5 % 318 45-64 53.6% 921 Amputation 4.5% 77 65-85 15.4% 265 Brain-related 26.4% 454 (missing) 0.3% 5 Neurological 14.3% 246 Gender Paraplegia 3.4% 58 Male 45.9% 789 Organ-related 11.5% 198 Female 54.1% 930 Chronic pain 16.4% 282 (missing) 0.0% 0 Miscellaneous 3.7% 63
Education Level (missing) 1.3% 23
Primary education 3.4% 59 Gross monthly household Lower vocational education 27.4% 471 Income (T0)
Secondary education 30.9% 532 <€2,290 18.9% 325 Higher professional education or university 19.8% 341 €2,290-€2,800 18.0% 310 (missing) 18.4% 316 >€2,800 24.8% 426
Long-term absences Not revealed 19.1% 328
(Partially) unable to work for > 2 years 19.9% 342 (missing) 19.2% 330
5 the table, the population is quite heterogeneous in terms of diagnosis, age, gross monthly income and education level. In addition, table 1 shows that a large percentage of the study population experienced long-term absences from work due to their condition. Moreover, with the exception of gender, all variables represented in the table contain missing values, ranging up to 18.4% missing for education level and 19.2% missing for gross monthly incomes. In section 4.2. this will be discussed in further detail.
3.3. Survey data
Medical consumption was assessed with the iMTA Medical Consumption Questionnaire (iMCQ) in both surveys for T0 and T1 (Bouwmans, et al., 2013a). The questionnaire includes questions about the number of consults with medical specialists in the last 13 weeks, the number of consults with primary healthcare providers (such as a physical therapist), the number of overnight hospital stays, the number of overnight stays in health facilities other than hospitals (such as rehabilitation centres), day treatments in health facilities other than hospitals (such as rehabilitation centres), and the types and amount of home care the individual received. Questions about which scans and examinations the individual had undergone and the amount of informal care an individual received were added to the iMCQ questionnaire for completeness.
Furthermore, to assess productivity losses, the iMTA Productivity Costs Questionnaire (iPCQ) was used (Bouwmans, et al., 2013b). In the questionnaire, participants were asked if they had paid and/or unpaid work, or whether they were inactive on the labour market due to their condition. In addition, they were asked how many hours of paid and unpaid work they had missed as a result of poor health. If they were present at work during the last three months, individuals were subsequently asked how their health condition influenced their productivity, by asking how many hours someone else would have to work to make up for the work the individual was unable to do due to his/her poor health.
3.4. Medical Cost data
Data on medical cost mainly came from the Dutch Manual of Costing 2015 and if not available in the 2015 edition from 2010 (Hakkaart-Van Roijen, et al., 2015; Tan, et al., 2012). Cost that were not included in the Dutch Manual of Costing were derived by averaging the prices published online of 8 Dutch hospitals1 (Dutch: passantentarieven) in the case of several scans and examinations (see appendix B). In the case of acupuncture, homeopathy and occupation physicians, costs were derived
1
6 by averaging the first five prices found. Prices that were not in €2014 were converted into 2014 prices with the consumer price index provided by Statistics Netherlands (2016a).
3.5. Travel expenses
Average distances to medical facilities and average prices for different modes of transport came from the Dutch Manual of Costing (Hakkaart-Van Roijen, et al., 2015). Individual data on mode of transport came from the ReSpAct survey as part of the iMCQ (Bouwmans, et al., 2013a).
3.6. Labour Market data
Labour market data for the HC method was derived from Statistics Netherlands. It consisted of the average gross wage based on age, gender and education level from 2010 (Statistics Netherlands, 2016b) converted to €2014 by using the CBA wage index from Statistics Netherlands (Statistics Netherlands, 2016c), labour participation rates for the general population based on age, gender and education level from 2013 (Statistics Netherlands, 2016d) and the average hours worked per week among the working population, subdivided by age and gender from 2014 (Statistics Netherlands, 2016e).
For the FC estimation the recommendations from the Dutch Manual of Costing (Hakkaart-Van Roijen, et al., 2015) were followed. The Dutch manual of costing assumes that an employer will wait approximately four weeks before posting a vacancy. Consequently, the total period of absence or reduced productivity that involves productivity losses for the employer according to the FC approach equals the vacancy duration plus four weeks. According to the Manual of Costing this totalled a maximum of 85 days, or approximately 12.1 weeks in 2014 (Hakkaart-Van Roijen, et al., 2015). In addition, the Manual of Costing provided productivity costs subdivided by gender, which were used for FC estimates instead of the gross hourly wages used for the HC estimates.
4. Method
4.1. Perspective and estimation period
7 result, calculating lifetime cost for both datasets and then comparing them would be illogical, since both estimates consider the same individuals. In addition, while lifetime costs could perhaps be estimated using a Markov model, the heterogeneity of the dataset makes this an aspiration outside the scope of this study.
As the ReSpAct questionnaires only inquired about productivity losses within the last four weeks to prevent recall bias, individual productivity data for individuals that were absent from work for less than a month was extrapolated to comprise a period of 13 weeks.
4.2. Missing data
Of only 16% of individuals in this study, all medical and non-medical costs were known. In addition, as several variables in the dataset were strongly correlated with the missingness of cost-data, data was undoubtedly not missing completely at random, indicating that estimation results based on elimination of individuals with missing data would likely produce biased results. To prevent biases as a result of missing data, missing data was imputed using the predictive mean matching multiple imputation technique (Little, 1988). With this method, data is estimated using a
nonparametric regression technique. In other words, after estimating the data with regression techniques, values are randomly drawn from k values of the same variable from the original sample that are closest to the estimated result for an individual. Usually, k is referred to as the number of nearest neighbours. By performing this routine multiple times, m imputations (complete datasets) are obtained on which the eventual analysis can be performed. By later combining these m different analysis outputs into one final output, standard errors can reflect the statistical uncertainty within the complete dataset, and the statistical uncertainty that comes with estimating one’s data by taking the variance between the m analysis outputs into account. Another advantage of this technique is that it allows for a realistic prediction of data with a non-normal distribution (which is usually the case for cost-data) by randomly drawing from k values from the original dataset that are closest to the estimated value. If k is large enough, imputed data will have approximately the same distribution as the original dataset. However, it should be noted that k should not be too large, as this will reintroduce the bias that was present in the original dataset due to missing data.
8 level, what kind of medical professional referred the patient to the counsellor, and whether an individual was unsure whether he/she was interested in being more physically active. After
imputation, cost where grouped together into three different categories: medical cost, productivity losses and travel expenses.
5. Results
5.1. Societal Cost
Average societal costs over the 3-month period before T0 were €5,208.47 (±€321.28; 95% CI) using the FC approach and €6,490.83 (±€375.34; 95% CI) using the HC approach. For T1, average societal costs were €2,352.23 (±€177.764; 95% CI) for the FC approach and €2,937.24 (±€235.89; 95% CI) with the HC approach. These results can also be found in table 2. In addition, table 2 also provides average societal cost estimates for different diagnoses groups. Table 3 provides the estimation results for different cost categories.
As can be derived from table 3, differences in productivity losses when estimated with the HC and the FC approach are substantial. When estimated with the HC approach, productivity losses at T0 are €3,159.49 (±€291.30; 95% CI), but when estimated with the FC approach, they drop by more than one third to €1,877.13 (±€215.29; 95% CI). This is repeated by the estimates for T1, where
Table 2: Average total societal cost and average societal cost for different diagnosis groups
T0 T1
Friction cost Human capital Friction cost Human capital Total societal cost € 5,208.47 (162.81) € 6,490.83 (191.50) € 2,352.23 (90.28) € 2,968.35 (120.35) Muscoskeletal € 5,106.95 (332.95) € 6,054.71 (393.36) € 2,205.15 (193.58) € 2,635.58 (243.11) amputation € 6,040.18 (671.31) € 7,188.71 (846.97) € 2,590.09 (470.14) € 2,918.45 (485.97) Brain-related € 5,678.53 (322.71) € 7,650.31 (388.72) € 2,536.56 (187.91) € 3,529.59 (279.05) Neurological € 5,105.32 (431.14) € 6,493.93 (494.67) € 2,942.79 (296.80) € 3,749.27 (375.41) Paraplegia € 5,581.76 (823.17) € 7,356.68 (961.79) € 2,257.89 (379.79) € 3,233.86 (695.07) Organ-related € 5,450.34 (531.28) € 5,896.19 (534.55) € 1,958.62 (256.61) € 2,248.01 (323.50) Chronic pain € 4,567.30 (340.76) € 5,385.87 (410.48) € 1,934.44 (203.31) € 2,272.05 (254.42) miscellaneous € 3,946.58 (540.95) € 5,757.04 (774.99) € 2,360.51 (448.83) € 2,809.48 (491.63)
9 Table 3: Productivity losses, Medical cost and Travel expenses
T0 T1
Friction cost Human capital Friction cost Human capital Productivity losses € 1,877.13 (109.84) € 3,159.49 (148.62) € 1,010.08 (72.65) € 1,626.20 (97.22) Medical cost € 3,236.95 (105.93) € 1,313.60 (53.88) Travel expenses € 94.39 (6.15) € 28.55 (2.38)
Estimates for the different cost categories. All costs are in €2014. Standard errors are given between brackets.
productivity losses are €1,626.20 (±€190.55; 95% CI) when estimated with the HC approach and €1,010.08 (±€142.39; 95% CI) when estimated with the FC approach.
Figure 1 shows the development of different categories of costs and their 95% CIs over time. As can be seen, all categories dropped significantly between T0 and T1 which is likely a logical consequence from the fact that at T0 most individuals were still in rehabilitation, but had left rehabilitation at T1. In addition, figure 1 also illustrates the stark contrast between productivity losses when measured with the FC approach or with the HC approach.
The following paragraphs will describe which diagnosis groups differed significantly in terms of societal cost. It can be important to know these differences, as average total societal cost might be too broad for policy decisions and practical use in further pharmacoeconomic research. For example,
Figure 1: Development of different cost categories between T0 and T1
Average costs for T0 and T1 in €2014. Productivity losses estimated by means of the Friction Cost approach are denoted by FC, productivity losses estimated by means of the Human Capital approach are denoted by HC.
10 there might be a larger societal gain in implementing new effective health technologies for groups that have particularly high cost. In addition, the in- or exclusion of long-term absences might have different consequences for each group, which could express itself in varying differences between groups when using one approach or the other. This would make the HC and FC debate more relevant for certain groups and could potentially pinpoint where one has to beware of choosing an approach too callously.
Figures 2 to 5 show the average societal costs for different diagnosis groups with the FC
Figure 2: Average societal costs for different diagnoses at T0 with the Friction Cost approach
Average costs for T0 in €2014
Figure 3: Average societal costs for different diagnoses at T1 with the Friction Cost approach
Average costs for T1 in €2014.
11 approach and the HC approach at T0 and T1. As can be derived from table 2 and the figures,
individuals with chronic pain have significantly lower societal cost at T0 (FC approach) than individuals with brain-related diagnoses (p<0.01), individuals with amputation (p<0.05) and individuals with organ-related diagnoses (p<0.10). In addition, individuals in the miscellaneous diagnosis group had significantly lower cost than individuals with muscoskeletal related diagnoses (p<0.05), individuals with amputation (p<0.01), individuals with brain-related diagnoses (p<0.01), individuals with neurological diagnoses (p<0.05), individuals with paraplegia (p<0.10), and individuals
Figure 4: Average societal costs for different diagnoses at T0 with the Human Capital approach
Average costs for T0 in €2014.
Figure 5: Average societal costs for different diagnoses at T1 with the Human Capital approach
Average costs for T1 in €2014.
12 with organ-related diagnoses (p<0.05).
At T1 with the FC approach, individuals with neurological diagnoses have significantly higher cost than individuals with muscoskeletal diagnoses (p<0.05), individuals with paraplegia (p<0.10),
individuals with organ-related diagnoses (p<0.01), and individuals with chronic pain (p<0.01). Furthermore, individuals with brain-related diagnoses have significantly higher societal cost than individuals with organ-related diagnoses (p<0.05) and individuals with chronic pain (p<0.05).
When cost are estimated with the HC approach, at T0 individuals with brain-related diagnoses have significantly higher cost than individuals with muscoskeletal diagnoses (p<0.01), individuals with neurological conditions (p<0.05), individuals with organ-related diagnoses (p<0.01), individuals with chronic pain (p<0.01) and individuals with diagnoses in the category miscellaneous (p<0.05).
Furthermore, individuals with chronic pain have lower societal cost at T0, than individuals with amputation (p<0.10), individuals with neurological diagnoses (p<0.05), and individuals with
paraplegia (p<0.05). Lastly, individuals with paraplegia have slightly significantly higher cost than individuals with organ related diagnoses (p<0.10).
At T1, both individuals with brain-related diagnoses and individuals with neurological diagnoses had significantly higher societal cost than individuals with muscoskeletal diagnoses (p<0.01; p<0.01), individuals with organ-related diagnoses (p<0.01; p<0.01) and individuals with chronic pain (p<0.01; p<0.01). In addition, individuals with neurological conditions had slightly significantly higher cost than individuals with amputation (p<0.10) and individuals that belonged to the miscellaneous diagnosis group (p<0.10).
The fact that the significance of societal cost differences between diagnoses groups differs between the HC approach estimations and the FC approach estimations for T0 and T1, indicates that for certain groups, long-term absences from work might play an important role. This seems to be the case for individuals with brain-related diagnoses, for example. With the FC approach societal cost for this group are only significantly higher than the chronic pain and miscellaneous group at T0.
13
5.2. Uncertainty Analyses
5.2.1. Friction period
The estimated friction period in this study was 85 days for all individuals, regardless of gender, or education level. However, while this greatly simplifies the estimation, results might be unrealistic. The Dutch Employee Insurance Agency stated in a report about Dutch vacancies in 2015 that
especially vacancies looking for highly educated employees were difficult to fill and that this difficulty increased when work experience became more important. Consequently, firms often had to content themselves with underqualified personnel (UWV, 2016). It would therefore be logical to vary the friction period with education level and age. Table 4 explores the effect on productivity losses when using education level adjusted friction periods. Deviation from the mean vacancy duration was calculated using Dutch vacancy data for different education levels from 2008 (Statistics Netherlands, 2009), after which the 2014 friction period was recalculated assuming that relative differences in vacancy duration for different education levels had not changed. As this might be an unrealistic assumption, an uncertainty analysis was performed by gradually decreasing and increasing the percentage deviation from the mean friction period with a factor in steps of 0.1. Consequently, the estimations with factor 1 assumes that relative deviations from the mean vacancy duration for different education levels had not changed between 2014 and 2008. As can be derived from the table, varying the friction period for different education levels does alter estimation results, productivity losses at T0 were €2,101.59 (±246.62; 95% CI) with education level-adjusted (2008) friction periods, and €1,877.13 (±215.29; 95% CI) without. The non-linearity of the results for different factors of the uncertainty analysis likely derives from the fact that vacancies for high education levels as well as vacancies for the lowest education level had an above average vacancy duration.
Table 4: FC productivity losses for varying education-dependent vacancy durations
factor T0 T1 0.9 €2,051.88 (130.64) €1,226.27 (111.96) 1 €2,101.59 (124.90) €1229.02 (123.25) 1.1 €2,054.53 (135.57) €1,232.22 (129.53) 1.2 €2,058.32 (122.59) €1,212.43 (119.88) 1.3 €2,049.98 (121.08) €1,247.17 (120.82) 1.4 €2,095.67 (119.44) €1,242.37 (116.74) 1.5 €2,053.88 (116.85) €1,230.29 (118.99)
14 However, there is an argument for assuming higher differences between education levels in friction periods than the 2008 estimates suggest: vacancies looking for higher educated personnel might have higher search cost (Russo, et al., 2005; Barron, et al., 1997), as well as a longer friction period. Consequently, to take into account potential higher search cost, one could either attach an additional cost to the friction period for highly educated individuals, artificially increase the foregone productivity cost as to include the search cost, or lengthen the friction period in such a way that the additional friction days/hours are the equivalent of the incurred search cost. Therefore, it might not be unlikely that the deviations from the mean friction period for individuals with a high education level calculated with 2008 data, should be multiplied by a factor >1 to include education adjusted search cost.
5.2.2. Productivity cost
In addition to assuming a universal friction period for all ages, genders and education levels, this study also assumed that hourly productivity costs only varied between genders as advised by the costing Manual (Hakkaart-Van Roijen, et al., 2015). This too seems overly simplistic, as highly educated employees are certainly expected to be more productive than their lower educated counterparts and unexperienced individuals just entering the labour market are expected to be less productive. Therefore, table 5 provides results for average societal cost when productivity losses are calculated using the friction method but with use of the labour market data earlier used to estimate HC productivity losses. In addition, table 5 also shows an estimation of average HC productivity losses when estimated with the hourly productivity cost provided by the Manual of Costing (Hakkaart-Van Roijen, et al., 2015). As can be derived from the table, FC productivity losses are lower when estimated using gross wages.
5.2.3. Presenteeism
Finally, all estimates for the FC approach in this paper assume that the friction period for
individuals that are still present at work despite their condition, but are less productive when at work due to their condition (presenteeism), is infinite. However, if an individual’s degree of presenteeism follows a predictable pattern for a longer period of time and when it is reasonable to assume that this pattern will continue in the future, a friction period should be used to estimate FC productivity Table 5: FC productivity losses estimated with age, gender and education dependent gross wages, HC productivity losses estimated with gender dependent productivity cost
T0 T1
FC with HC wages € 1,414.30
(79.77)
€ 750.85 (53.48)
HC with FC productivity cost € 4,486.53 (216.33)
€ 2,284.25 (131.41)
15 Table 6: FC productivity losses without presenteeism
T0 T1
FC productivity losses € 1457.22 (91.27)
€ 608.21 (53.55)
Productivity losses calculated with the friction cost approach without including losses due to presenteeism. Standard errors are between brackets
losses due to that individual’s presenteeism. After all, if the presenteeism is predictable and
expected to continue for a significant duration, an employer can anticipate the resulting productivity losses and act accordingly, for example by hiring an additional part-time employee. As with the current data it is impossible to know for which individuals this might be the case, estimating presenteeism cost with a friction period is extremely problematic. Nevertheless, to still provide an idea of scope regarding presenteeism cost, table 6 provides productivity losses estimated without productivity losses that resulted from presenteeism. When presenteeism is not taken into account, FC productivity losses are €1,457.22 (±178.89; 95% CI) and €608.21 (±104.96; 95% CI) at T0 and T1 respectively. Assuming all the other initial assumptions are valid, true FC productivity losses are likely to lie somewhere between the original estimate and the new estimate.
5.2.4. Travel expenses
As the survey only contained one question about travel expenses - asking which mode of transport the individual had often used – where multiple answers were possible, it was not possible to discern the true extend of travel expenses for each individual. Consequently, the estimations in section 4 constitutes a lower bound, as it was assumed that individuals had only travelled with the least expensive mode of transport they had indicated in the survey. As this assumption might lead to unrealistic results, travel expenses were estimated again with the assumption that individuals had only used the most expensive mode of transport indicated by their survey response. Results can be found in table 7. When considering only the most expensive modes of transport, average travel expenses were €144.38 (± 7.33; 95% CI) at T0 and €46.35 (±3.21; 95% CI) at T1. While these results differ statistically significantly from the low estimate (p < 0.01), they do not differ substantially: the difference between the two means was around €50 at T0 and €20 at T1. Consequently, the
assumptions regarding mode of transport are not likely to influence the outcome of average total societal cost substantially.
Table 7: Travel expenses: most expensive mode of transport
T0 T1
High estimate € 144.38
(7.33)
€ 46.35 (3.21)
16
6. Discussion
6.1. Travel Expenses
In addition to the mode-of-transport assumption, the previous estimates for travel expenses also assumed that there were no opportunity costs to traveling in terms of (leisure) time. This assumption was made, because the Manual of Costing (Hakkaart-Van Roijen, et al., 2015) advises against the valuation of leisure time and because there was no individual data available on time spent traveling. Estimating individual time spent traveling based on average distances to medical facilities and mode of transport is problematic, as often individuals have to ask friends/family to accompany them, or to drive them. As it is almost impossible to know when an individual travelled alone and when that same individual was accompanied by friends or family, estimating the opportunity cost of time spent travelling becomes a precarious undertaking. Therefore, it was not taken into consideration in the previous estimates. As the average distances to different healthcare facilities are quite small (the largest average distance mentioned by the Manual of Costing was 7 km (Hakkaart-Van Roijen, et al., 2015)), it is also unlikely that the inclusion of lost leisure time due to traveling would greatly
influence the results.
6.2. Passantentarieven
Since for several scans and examination the Dutch Manual of Costing (Hakkaart-Van Roijen, et al., 2015) did not provide prices, these costs were estimated by averaging the prices published online - the so-called passantentarieven - of 8 Dutch hospitals. The Dutch Manual of Costing (Hakkaart-Van Roijen, et al., 2015) advises against the use of passantentarieven based on a newspaper article published by a Dutch newspaper in 2013 (Wester, 2013) that compared the prices of seven
treatments for 33 hospitals and found large differences. However, when comparing the prices of the 10 medical examinations for which this research used passantentarieven, differences in price were much smaller (see appendix B). In addition, when prices do vary considerably, the effect of that variance on the accurateness of societal cost is likely be small, as either the cost of the examination was low, or only few individuals in the sample underwent the examination.
6.3. Medication
Because of time constraints, the estimates for medical cost exclude medication use.
17
6.4. Endogeneity
As was already expressed in section 2, when looking at the larger picture, where one leaves the realm of partial equilibria, or when the analysis involves a very large group of individuals, the HC approach seems more sensible than the FC approach. The current study involves a group of individuals with a wide range of different diagnoses and as such represents a considerable share of the Dutch population: in 2014, 12.7% of the Dutch population older than 12 had a physical disability, according to Statistics Netherlands (2016). Consequently, the societal cost of this group is likely to have micro- and macroeconomic consequences that cannot be captured by a partial equilibrium analysis such as performed here. For example, a large group with average medical cost of €3236.95 in just the first three months of this study is likely to influence national health care prices. In addition, with productivity losses of €3159.49 (HC estimate), or €1877.13 (FC estimate), the study population is likely to have an influence on national labour market outcomes, such as wages and unemployment. Therefore, neither the HC approach nor the FC approach is completely satisfactory in this case. To estimate the full extent of societal cost, a macroeconomic model should be used, so that parameters such as health care prices and wages can be allowed to be endogenous. While this approach is not mentioned in the guidelines of the National Health Care Institute (2015) or the Manual of Costing (Hakkaart-Van Roijen, et al., 2015), Koopmanschap et al. (1995) already assumed this approach in the paper in which they presented the FC approach. However, it should be noted that using the FC approach in a macroeconomic model could be problematic as when labour market parameters are assumed to be endogenous, the friction period could be endogenous as well. This advice is also in line with a yet to appear publication of the Dutch Central Planning Bureau on cost-benefit analyses, that argues that the HC approach should be used and that subsequently the estimates should be incorporated into a macroeconomic model.
18
6.5. Comparison with literature
The findings of this study are similar to the previous literature in that estimates for productivity losses can differ greatly, depending on estimation method (Van den Hout, 2010; Hanly, et al., 2012; Hutubessy, et al., 1999). The absolute values of the estimates are difficult to compare to other studies, as it involves a relatively short estimation period and diagnosis groups that consist of a wide range of different diagnoses. In addition, cost-of illness studies are often not comparable across different countries (Drummond, et al., 2015, p. 227).
6.6. Recommendations
As can be derived from the analyses above, it can matter greatly which method is used. The following section aims to provide guidance when determining which method should be used.
When an incremental cost analysis is performed and the treatment does not seem to affect long-term absences, or when the disease itself only causes absences for limited periods of time, the FC and the HC approach will give similar results. Consequently, either method will be appropriate and it is recommended that the researcher uses the method that is easiest to implement.
When the population under study is small, i.e. the disease is rare, or the number of individuals affected by a treatment is small, it is not unreasonable to assume that the population has no influence on wages and health care prices and that there is an unlimited supply of unemployed labour against a certain cost for the individuals and the firms affected by the disease or treatment. Consequently, the FC approach would be appropriate.
On the other hand, when the population under study is large, or when societal costs are expected to be high, neither method is sufficient. In that case, a macroeconomic model should be used in which the health care sector is explicitly modelled, so that health care prices, wages, unemployment and the friction period are allowed to be endogenous. It is strongly recommended that further research aims at developing such a model and making it accessible in a user-friendly manner, so that researchers without a strong economic background can use it as well. In addition, once such a model is developed, it can be investigated in more specific terms when groups, or costs, are too large for the FC method.
19
7. Conclusion
This study aimed at providing an estimation for the societal cost of Dutch individuals with a disability or chronic illness during and after rehabilitation and exploring the effect of the human capital and friction cost approach on these societal costs. The estimated societal cost during the first three months, when most individuals were still in rehabilitation, were €5208.47 (±€321.28; 95% CI) when estimated with the friction cost approach and €6490.83 (±€375.34; 95% CI) when estimated with the human capital approach. In the following three months, societal cost were estimated at €2352.23 (±€177.764; 95% CI) and €2937.24 (±€235.89; 95% CI) respectively. Various uncertainty analyses showed that friction cost productivity losses could lie somewhere between €1414.30 (±€156.35; 95% CI) and €2095.67 (±€119.44; 95% CI)) during the first three months, and between €608.21 (±104.96; 95% CI) and €1247.17 (±€120.82; 95% CI) during the second three months. It should be noted that in the latter case, the lower estimate is very conservative, as it excludes productivity losses due to presenteeism. When productivity losses are estimated with the human capital approach, they lie somewhere between €3159.49 (±€291.30; 95% CI) and €4486.53 (±€424.01; 95% CI) in the first three months, and between €1626.20 (±€190.55; 95% CI) and €2284.25 (±€257.56; 95% CI) for the second three months. True societal cost are likely to lie somewhere between the human capital and the friction cost estimates.
The second aim of the study was to provide guidance on when either of the two methods should be used. When the number of individuals affected by a disease, or treatment is small, or when the disease only results in short-term absences from work, the friction cost approach is sufficient for estimating productivity losses. When groups are large, neither of the two approaches provides satisfactory results. Instead, a macroeconomic model with an explicitly modelled health care sector should be used. Further research should be aimed at developing such a model. When the use of such a model is not possible, productivity losses should be estimated with both the friction cost and human capital approach, as the first provides a lower bound, while the latter provides an upper bound.
References
Alingh, R., Hoekstra, F. & Van der Schans, C., 2015. Protocol of a longitudinal cohort study in physically disabled patients participating in a rehabilitation counselling programme: ReSpAct. BMJ Open, Volume 5.
Barron, J., Berger, M. & Black, D., 1997. Employer Search, Training, and Vacancy Duration. Economic Inquiry, 35(1), pp. 167-192.
20 Bouwmans, C. et al., 2013b. Productivity Cost Questionnaire - Productivity and Health Research Group, Rotterdam: Institute for Medical Technology Assessment, Erasmus University.
Capri, S. et al., 2001. Guidelines for Economic Evaluations in Italy: Recommendations from the Italian Group of Pharmacoeconomic Studies. Drug Information Journal, Volume 35, pp. 189-201.
Drummond, M. et al., 2015. Methods for the Economic Evaluation of Health Care Programmes. 4th red. Oxford: Oxford University Press.
Hakkaart-Van Roijen, L. et al., 2015. Manual of Costing, Erasmus Universiteit Rotterdam: Institute for Medical Technology Assessment.
Hanly, P., Timmons, A., Walsh, P. & Sharp, L., 2012. Breast and Prostate Cancer Productivity Costs: A Comparison of the Human Capital Approach and the Friction Cost Approach. Value in Health , 15(3), pp. 429-436.
Hoekstra, F. et al., 2014. Design of a process evaluation of the implementation of a physical activity and sports stimulation programme in Dutch rehabilitation setting: ReSpAct. Implementation Science, 127(9).
Hutubessy, R., Van Tulder, M., Vondeling, H. & Bouter, L., 1999. Indirect costs of back pain in the Netherlands: a comparison of the human capital method with the friction cost method. Pain, Volume 80, pp. 201-207.
IQWiG, 2009. General Methods for the Assessment of the Relation of Benefits and Costs, Cologne: Institute for Quality and Efficiency in Health Care.
Johanesson, M. & Karlsson, M., 1997. The friction cost method: A comment. Journal of Health Economics, Volume 16, pp. 249-255.
Koopmanschap, M., Rutten, F., Van Ineveld, B. & Van Roijen, L., 1995. The friction cost method for measuring indirect costs of disease. Journal of Health Economics, Volume 14, pp. 171-189.
Koopmanschap, M., Rutten, F., Van Ineveld, B. & Van Roijen, L., 1997. Reply to Johanneson's and Karlsson's comment. Journal of Health Economics, Volume 16, pp. 257-259.
Little, R., 1988. Missing-Data Adjustments in Large Surveys. Journal of Business & Economic Statistics, 6(3), pp. 287-296.
National Health Care Institute, 2015. Richtlijn voor het uitvoeren van economische evaluaties in de gezondheidszorg, Diemen: National Health Care Institute.
ReSpAct, sd www.respact.nl. [Online] Available at: www.respact.nl
[Accessed 15 June 2016].
21 Statistics Netherlands, (., 2016d. Arbeidsdeelname; kerncijfers. [Online]
Available at: http://statline.cbs.nl/Statweb/publication/?DM=SLNL&PA=82309ned&D1=22-23&D2=1-2&D3=11-22&D4=a&D5=54&HDR=G4&STB=G1,G2,G3,T&VW=T
[Accessed 18 March 2016].
Statistics Netherlands, (., 2016e. Werkzame beroepsbevolking; arbeidsduur. [Online]
Available at: http://statline.cbs.nl/Statweb/selection/?VW=T&DM=SLNL&PA=82647NED&D1=0,3-4,6-9&D2=1-2&D3=1-11&D4=59&HDR=G3,G2&STB=G1,T
[Accessed 28 April 2016].
Statistics Netherlands, (., 2016. Gezondheid en Zorggebruik; persoonskenmerken. [Online]
Available at: http://statline.cbs.nl/Statweb/publication/?VW=T&DM=SLNL&PA=83005NED&D1=39-51&D2=0-2,5-13,37-41&D3=0&D4=l&HD=150430-1403&HDR=G2,G3,T&STB=G1
[Accessed 11 February 2016].
Statistics Netherlands, 2009. Vacatures; bedrijfsgrootte, beroep, opleiding, 2000 - 2008. [Online] Available at:
http://statline.cbs.nl/Statweb/publication/?DM=SLNL&PA=70238ned&D1=1,10,19,28,37,46,246,250, 254,258,262,266,270-272,298-299&D2=0&D3=4-5,l&D4=a&HDR=G2,T&STB=G1,G3&VW=T
[Accessed 12 June 2016].
Statistics Netherlands, 2016a. Consumentenprijzen; prijsindex 2006 = 100, 1996 - 2015. [Online] Available at:
http://statline.cbs.nl/StatWeb/publication/?VW=T&DM=SLNL&PA=71311NED&D1=0,2,4,6&D2=0-
1,61,70,87,108,137,145,172,176,221-222,230,255,l&D3=(l-34)-l&HD=081020-1310&HDR=T&STB=G1,G2 [Accessed 20 March 2016].
Statistics Netherlands, 2016b. Banen en lonen; werknemers 15 tot 65 jaar, onderwijsniveau en baankenmerken. [Online]
Available at: http://statline.cbs.nl/Statweb/publication/?DM=SLNL&PA=82049ned&D1=1&D2=1-2&D3=1-5&D4=a&D5=0&D6=l&HDR=T%2cG1%2cG2&STB=G3%2cG4%2cG5&VW=T
[Accessed 28 April 2016].
Statistics Netherlands, 2016c. Cao-lonen, contractuele loonkosten en arbeidsduur; indexcijfers (2010=100). [Online]
Available at: http://statline.cbs.nl/StatWeb/publication/?DM=SLNL&PA=82838ned [Accessed 28 April 2016].
Tan, S., Bouwmans-Frijters, C. & Hakkaart-van Roijen, L., 2012. Handleiding voor kostenonderzoek: methoden en referentieprijzen voor economische evaluaties in de gezondheidszorg. Tijdschrift voor gezondheidswetenschappen, 90(6), pp. 367-372.
UWV, 2016. Vacatures in Nederland 2015, sl: UWV, department of labour market information and advise.
22 Van der Ploeg, H., 2005. Promoting physical activity in the rehabilitation setting. Amsterdam: Vrije Universiteit Amsterdam.
Walter, E. & Zehetmayr, S., 2006. Guidelines on Health Economic Evaluation, Vienna: Institute for Pharmacoeconomic Research.
23
Appendix A
Diagnosis groups:Diagnosis group 1 = muscoskeletal diagnoses
Subgroup A: Congenital muscoskeletal diseases Subgroup B: Miscellaneous disorders upper extremity Subgroup C: Miscellaneous disorders lower extremity Subgroup D: Disorders in vertebral column en torso Subgroup E: Rheumatic disorders
Subgroup F: Polytrauma (=multiple bone fractures) Subgroup G: Miscellaneous muscoskeletal diseases Diagnosis group 2 = Amputation
Subgroup A: Amputation upper extremity Subgroup B: Amputation lower extremity Diagnosis group 3 = Brain
Subgroup A: Congenital brain defects
Subgroup B: Cerebrovascular accident (CVA) Subgroup C: Acquired brain injury: miscellaneous Diagnosis group 4 = Neurology
Subgroup A: Cerebrospinal diseases Subgroup B: Neuropathy
Subgroup C: Neuromuscular diseases
Subgroup D: Miscellaneous neurological diseases Diagnosis group 5 = Paraplegia
Subgroup A: Paraplegia Subgroup B: Spina bifida Diagnosis group 6 = Organs
Subgroup A: Cardiovascular diseases Subgroup B: Respiratory diseases
Subgroup C: Miscellaneous organ diseases Diagnosis group 7 = Miscellaneous
Subgroup A: Oncology Subgroup B: Mental disorders
Subgroup C: miscellaneous disorders and symptoms (incl. somatic symptom disorder and chronic fatigue syndrome)
24
Appendix B
Average prices for passantentarieven, standard deviation and 95% confidence intervals of eight Dutch hospitals and the number of individuals in the sample
