The Effect of Multidisciplinary Spinal Care on Back
Pain Related Healthcare Costs
Nynke Buwalda
1Supervisors: Jochen Mierau and Hermien Dijk University of Groningen, the Netherlands
June 6, 2018
Abstract
The high prevalence and high direct and indirect societal costs of back pain increase the need for an intervention that is clinically effective and that reduces associated costs. This paper contains an analysis of the effect on back pain related healthcare costs of an intervention for complex chronic back pain implemented by the Groningen Spine Center, namely multidisciplinary spinal care. For this analysis a clinical dataset provided by the Groningen Spine Center was used, as well as a dataset containing all healthcare declarations of the patients undergoing this treatment provided by the healthcare insurer Menzis. This resulted in a sample of 997 patients and 700.533 declarations. The employed pooled OLS model indicates that multidisciplinary spinal care has a significant negative effect on back pain related healthcare costs. The results are robust to controlling for background characteristics, using multiple analytical methods and unlikely to be driven by regression to the mean. A study of the predictors of treatment outcome shows that age, bmi, duration of pain pre-treatment and comorbidity at baseline significantly influence the treatment effect.
Keywords: Low Back Pain, Multidisciplinary Spinal Care, Healthcare Costs, Groningen Spine Center
JEL-code: H51, I12, I13
1 Corresponding author. Tel: +31621672846
2
1. Introduction
As mortality rates decline and populations age, non-fatal outcomes of diseases and injuries are becoming a larger part of the global burden of disease (Global Burden of Disease Study, 2017). As a result, health systems are facing increasing demand for services that are generally more costly than the interventions that have led to the declines in mortality (Olshansky and Ault, 1986). In order for the health system to adequately respond, attention should be paid to the trends of these non-fatal diseases and their treatments.
According to the Global Burden of Diseases, Injuries, and Risk Factors Study (Vos et al., 2017), out of the non-fatal diseases, back pain is the leading cause of morbidity globally. Lifetime prevalence estimates indicate that 80 percent of the population has reported having back pain at some point in their life, with annual prevalence estimates of severe back pain ranging from 13 to 54 percent (World Health Organization, 2013). The impact of back pain is multi-dimensional and includes decreased quality of life, impaired social functioning, increased sick leave and high direct and indirect societal costs (Anderson, 1999; Breivik, Cohen, Collett, Gallacher and Ventafridda, 2006; de Vroome et al., 2015). High direct and indirect societal costs are mostly due to high persisting healthcare costs and the indirect costs of work absence (absenteeism) or productivity losses at work (presenteeism) (Caro, Dagenais and Haldeman, 2008; de Vroome et al., 2015). In the Netherlands, research indicates that musculoskeletal disorders are the fifth most expensive disease category in terms of healthcare costs and the most expensive in terms of indirect societal costs, with respectively one-third to one-half of these costs due to back pain (Bouter, Koes and Tulder, 1995). The total of these annual direct and indirect societal costs of back pain in the Netherlands is estimated to be 3.5 billion euros (Lambeek et al., 2011). Due to the high prevalence and the high economic burden of back pain there is a pressing need for an effective treatment and several are being developed and tested.
3
Although the evidence suggests that multidisciplinary biopsychosocial treatments are more effective than usual care in decreasing pain and disability for people with chronic back pain (Guzman et al., 2001; Karjalainen et al., 2001; Kamper et al., 2015), the evidence on the effect of these programs on healthcare costs is relatively scarce (Marin et al., 2017). The current paper adds to the existing literature by combining extensive clinical data from the Groningen Spine Center with data containing healthcare declarations of these patients from the healthcare insurer Menzis in order to answer the question: what is the effect of multidisciplinary spinal care for complex chronic back pain patients on back pain related healthcare costs? Furthermore, it uses baseline characteristics to assess whether patient and disease characteristics have predictive value for the effect of multidisciplinary spinal care on back pain related healthcare costs.
The next section includes a literature review, discussing literature on multidisciplinary biopsychosocial treatments, predictive factors for the treatment outcome of these treatments and the effect of these treatments on healthcare use. Section 3 discusses the data and relevant descriptive statistics. Section 4 discusses the methodology employed in this analysis. Section 5 presents the empirical results. Section 6 contains a discussion of the results and section 7 is the conclusion.
2. Literature Review
Increasingly widespread acceptance of the biopsychosocial model (Foster, 2011), accompanied by the relatively modest performance of monotherapies in clinical trials (Machado et al., 2009), has led to increased research into the effectiveness of multidisciplinary biopsychosocial treatment. Most evidence suggests that multidisciplinary biopsychosocial programs are more effective than usual care and physical treatments in decreasing pain and disability in people with chronic back pain. However, the evidence is not unanimous (Guzman et al., 2001; Karjalainen et al., 2001; Kamper et al., 2015).
4
value in explaining the variance in clinical outcome measures (Hasenbring, 1998; Hulst, Vollenbroek-Hutten and Ijzerman, 2005; McCracken and Turk, 2002).
Lastly, for other psychological variables (e.g. depression, hypochondria, personality) the evidence is mixed. Hulst, Vollenbroek-Hutten and Ijzerman (2005) performed a systematic review of the available literature on sociodemographic, physical and psychological predictors of multidisciplinary interventions for patients with chronic low back pain. They found no consistent evidence for the predictive value for treatment outcomes of psychological variables. However, other systematic reviews by Hasenbring (1998) and Feuerstein et al. (1994) do show that high levels of depression at baseline are associated with worse clinical treatment outcomes. The aforementioned results indicate that pain intensity, work-related factors, coping skills and depression could influence treatment outcomes of multidisciplinary biopsychosocial treatments.
Multidisciplinary treatments need to be evaluated in terms of clinical effects as well as healthcare costs, since healthcare costs are a major determinant of the societal burden of low back pain (Meatzal, 2002). Multidisciplinary interventions can be resource intensive and time consuming (Kamper et al., 2015). However, if the effectiveness of multidisciplinary interventions decreases the duration or severity of the disease compared to usual care, this could in turn reduce the healthcare costs. Although the literature on this topic is relatively scarce (Marin et al., 2017), several studies have tried to estimate the effect of multidisciplinary interventions on healthcare use and costs.
In a meta-analysis of the existing literature, Flor, Fydrich and Turk (1992) found that the beneficial effects of multidisciplinary treatment were not only limited to pain, mood or interference in daily life, but extended to behavioral variables such as return to work and decreased use of the healthcare system. According to their results, the healthcare use of the treatment groups considered decreased by thirty-five percent. However, they stated that their analysis was limited by the fact that the included studies often had no or an inappropriate control group, as well as poor descriptions of patient populations, designs and analysis.
5
Another outcome measure for healthcare use was employed by Simmons et al. (1988), who compared medical insurance declarations one year pre-treatment to one year post-treatment. Their analysis included a small sample of sixteen patients with chronic low-back pain taking part in a multidisciplinary treatment program in the United States. Compared to one year pre-treatment a reduction in medical costs of fifty-nine percent was found post-treatment.
To summarize, in general the literature suggest that multidisciplinary treatments are more effective than usual care and physical treatments for treating chronic back pain in terms of clinical outcomes. However, new treatments not only need to be evaluated in terms of clinical effects, but in terms of healthcare costs as well. Although the literature on the effect of multidisciplinary biopsychosocial treatment for chronic back pain on healthcare costs is relatively scarce and of limited quality, the evidence suggest that multidisciplinary treatments for chronic back pain decrease healthcare use and costs. Furthermore, since research has shown that certain factors have predictive value for clinical treatment outcomes of multidisciplinary treatment for patients with chronic back pain, these factors might also influence the effect of treatment on healthcare costs. Therefore, this research paper will complement the existing literature by exploring the relationship between multidisciplinary spinal care for chronic back pain and back pain related healthcare costs. Additionally, it will add to the existing literature by assessing the predictive value of patient and disease characteristics at baseline for the effect of multidisciplinary spinal care on back pain related healthcare costs.
3. Data and Descriptive Statistics
6
Table 1: Patient Descriptive Statistics (N=997)
A second dataset was provided by the healthcare insurer Menzis containing healthcare declarations of these patients. Menzis gathered declarations filed from the first of January 2011 until the 31st of December 2017, thereby trying to gather declarations from two years pre-treatment to at least two years post-pre-treatment for every patient. This resulted in a total of 700.533 declarations for 997 patients. The Medical Ethics Committee Groningen approved the use of these records for research purposes. To my knowledge, this study is the first that makes use of such an extensive dataset of healthcare declarations to study the effect of multidisciplinary treatment on healthcare costs for a large patient population.
3.1 Patient Selection A total of 1830 patients were admitted to the Groningen Spine Center and invited to participate in the study. To be eligible for participation, patients needed to be over 18 years, insured at the healthcare insurer Menzis and to have reported back pain as their primary pain site. Out of these 1830 patients, 442 participants did not sign informed consent for unknown reasons, 264 patients were deemed to have complaints not related to their spine after the first consult (e.g. widespread pain, leg or arm pain unrelated to the spine), 92 patients never showed up at first consultation, 15 patients did not fill out the baseline data and 20 persons deceased because of reasons unrelated to their spine. This resulted in a final sample of 997 patients.
3.2 Procedures After referral by a general practitioner or secondary care specialist, patients were sent an online questionnaire containing biopsychosocial questions. Additionally,
Name Min Max Mean Std. Dev. Missing
SocioDemographic Characteristics
Age 18 80 51.726 14.432 59
Gender (male/female) 0 1 414/586 - 0
Education High (high/low) 0 1 276/778 - 61
Financial Worries 0 1 0.140 0.347 21
Bmi 16.096 49.219 27.142 5.091 131
Work Related Factors
7
former imaging from other medical facilities was retrieved. All information was gathered by four physician assistants who performed triaging on the patients. Subsequently, patients consulted one or multiple departments at the Groningen Spine Center and were referred to the most optimal treatment based on the clinicians view. Interventions were chosen based on patient preference and medical status, which included outpatient multidisciplinary rehabilitations with psychology, physiotherapy and occupational therapy or anesthesiology to patients with sacroiliac-blocks or clear sensitization in well-described dermatomes. Surgery was provided to those patients with a specific operation indication (i.e. instable fractures, stenosis or failed conservative rehabilitation in patients with disc herniation). A minimal intervention was provided for diagnostic purposes (such as second opinions) or when patient and specialist disagreed on the intervention indication. The minimal intervention consisted of a single consultation with advice and reassurance on how to cope with complaints. Table 2 contains information on the distribution of patients to the treatment types.
Table 2: Patients per Treatment (N=997)
Lastly, to assess the effectiveness of care, another biopsychosocial questionnaire was sent to the patients at discharge, three months post discharge and twelve months post discharge. The response rate for the clinical dataset questionnaires at discharge, three months post discharge and twelve months post discharge was too low to use in a meaningful analysis. Therefore, only baseline data of the clinical dataset was used for this analysis.
3.3 Healthcare Costs The healthcare declarations were stratified into blocks of one year. Six blocks were formed, starting at two years pre-treatment and ending at three years post-treatment. The healthcare costs declared between baseline and discharge at the Groningen Spine Center were aggregated into one block (T=0). Because the interest of this analysis was to compare pre-treatment to post-treatment healthcare costs, block T=0 was not included in the models. Furthermore, based on ATC codes attached to declarations, a division in healthcare costs was made based on relevance to the treatment of back pain as well as the type of care that was involved. Table 3 contains an overview of the mean back pain related costs and mean back pain related costs per relevant care category over the years in euros.
8
Table 3:Mean Back Pain Related Healthcare Costs in Euros
Menzis attempted to gather all declarations ranging from at least two years pre- to two years post-treatment. However, due to the fact that some patients possibly changed health insurer during that time period or started treatment close to the first of January 2011 or the 31st of December 2017, there were no declarations available for the entirety of that range for some patients, whilst for other patients declarations for a larger range of blocks were included. To assure the completeness of the declaration blocks included in the analysis, the difference was taken between the start of treatment and the first of January 2011 as well as the discharge date and the 31st of December 2017 per patient. A patient declaration block was only included in the models if that block could contain a full year of declarations, so for example the block of a patient for three years post-treatment was included if declarations were made during that time period and the discharge date of that patient was at least three years prior to the 31st of December 2017.
Figure 1 contains a graph with the mean back pain related healthcare costs over time. The graph seems to indicate that back pain related healthcare costs are rising pre-treatment and are lower, at least compared to one year pre-treatment, and decreasing post-treatment. To assess whether this difference is significant and not driven by other effects, such as omitted variable bias and regression to the mean, a regression analysis is needed. This analysis is described in the methodology section below.
-2 Years -1 Years During
Treatment + 1 Years +2 Years +3 Years
9
Figure 1
3.4 Treatment Indicator The main independent variable in this regression analysis is whether the patient has experienced treatment. The data obtained from the Groningen Spine Center contains no control group and consists only of patients exposed to the treatment. Therefore, a dummy variable was created, indicating whether that patient was pre- or post-treatment. A value of 0 was assigned to a patient that was pre-treatment and a value of 1 was assigned to a patient was post-treatment.
3.5 Patient and Disease Characteristics As mentioned before, the range of declaration blocks differed per patient. This entailed that not all patients were included both pre-treatment and post-treatment and that pre- and post-treatment patient populations may differ on characteristics that influence back pain related healthcare costs themselves. To capture the true effect of treatment on back pain related healthcare costs, controls were included for patient and disease characteristics at baseline. These characteristics can be classified into five main categories: individual characteristics, other sociodemographic characteristics, psychological factors, work-related factors and disease characteristics.
The individual characteristics are age and gender (0 indicating a female and 1 indicating a male). The category of other sociodemographic characteristics include: a dummy for a high level of attained education (no education, primary school, lower vocational education, and high school was assigned a value of 0 and secondary vocational education, pre-university education, and higher professional education was assigned a value of 1), bmi and a dummy variable for financial worries, which serves as a proxy for income since no income data was available. The indicators for psychological factors are: mental health as measured by the Rand-36 for emotional wellbeing questions (the score was adjusted by a factor of one-tenth to create a
0 500 1000 1500 2000 2500 3000 -2 -1 0 1 2 3 Hea lth ca re C os ts in E ur oś
Time to Treatment in Years
10
variable value range from 0 to 10) (Hays, Mazel and Sherbourne, 1993) and coping skills as measured by the answer to question twelve of the Örebro musculoskeletal pain questionnaire (based on all things you do to cope, or deal with your pain, on an average day, how much are you able to decrease it on a 0-10 scale?) (Linton, 1999). The indicators for work-related factors are: job satisfaction as measured by question seventeen of the Örebro musculoskeletal pain questionnaire (if you take into consideration your work routines, management, salary, promotion possibilities and work mates, how satisfied are you with your job on a 0-10 scale?) (Linton, 1999) and how strenuous a patient’s job is as measured by question eight of the Örebro musculoskeletal pain questionnaire (is your work heavy or monotonous on a 0-10 scale?) (Linton, 1999). Lastly, the included disease characteristics are: pain intensity as measured by question ten of the Örebro musculoskeletal pain questionnaire (in the past three months, on average, how bad was your pain on a 0-10 scale?) (Linton, 1999), baseline EQ-5D score, a dummy for duration of pain pre-treatment (0 indicating less than one year and 1 indicating more than one year) and a dummy variable for comorbidity (0 indicating no comorbidity and 1 indicating comorbidity).
Furthermore, the reviewed literature indicated that several patient and disease characteristics are associated with treatment outcomes of multidisciplinary biopsychosocial treatments. To assess whether patient and disease characteristics influence the effect of multidisciplinary spinal care on back pain related healthcare costs, the aforementioned variables were interacted with the treatment dummy in a regression analysis.
3.6 Missing
11
variable. Subsequently, the missing values are replaced by a random draw from the corresponding posterior predictive distribution. This process is repeated for all other variables using the appropriate regression model with missing values, as well as for several cycles to produce an imputed dataset. The entire procedure is performed m, in this case ten, times to produce m imputed datasets, which can be used in a subsequent analysis.
The variables included in the multiple imputation model were: age, gender, bmi, high education, financial worries, job satisfaction, hard work, mental health, coping skills, pain intensity, EQ-5D baseline, comorbidity and duration. Out of these variables, all except for gender were imputed. This means that a total of ten imputations was performed. This resulted in a final dataset that contains 997 values for all patient and disease characteristics.
4. Methodology
The data supplied by the Groningen Spine Center and Menzis contains only patients admitted to the treatment. No control group is available containing patients admitted to usual or no care. This entails that the back pain related healthcare costs of patients had the treatment not occurred are unknown, which means that pre-treatment and post-treatment back pain related healthcare costs may differ due to reasons unrelated to the treatment. This research paper will employ different methods to evaluate the effect of multidisciplinary spinal care on back pain related healthcare costs and by doing so increase the validity that the observed effect is indeed due to the treatment and not due to other effects, such as regression to the mean.
4.1 Paired Sample T-Test Firstly, following the method used by Simmons et al. (1989), a comparison was made of back pain related healthcare costs pre-treatment to these costs post-treatment. A paired sample t-test was used to test for the equality of the mean back pain related healthcare costs of pre-treatment and post-treatment patient populations. A comparison was made between one year as well as two years pre-treatment with one, two and three years post-treatment. Only patients were included in the paired sample t-tests for which a complete declaration block was available for both of the comparison years.
4.2 Pooled OLS Model Secondly, to assess the effect of multidisciplinary spinal care on back pain related healthcare costs whilst taking patient and disease characteristics into account, the following pooled OLS model was used:
(1) ln (𝐻𝐶𝑖𝑡) = 𝛽1+ 𝛽2𝐼𝐶𝑖 + 𝛽3𝑇𝑟𝑖𝑡+ 𝛽4𝑃𝐶𝑖 + 𝛽5𝐷𝐶𝑖 + 𝜀𝑖𝑡
pre-12
treatment until two years post-treatment were used, because this range was indicated to be the most complete and reliable by Menzis. The dependent variable is HCit and indicates back pain
related healthcare costs of patient i at time t. The log of HCit was used to regularize the skewed
data, that is due to a long right tail. ICi stands for the individual characteristics that were added
to each regression, namely: age, age squared and the gender of person i. Age squared was used due to the likelihood that the effect of age is non-linear. Trit was constructed for each individual
over time and indicated 0 if the patient i was pre-treatment at time t and 1 if the patient i was post-treatment at time t. Based on the reviewed literature, I expect that treatment has a negative effect on back pain related healthcare costs and therefore 𝛽3 to be negative and significant.
As stated in the data section, pre- and post-patient patient populations may differ on characteristics that influence back pain related healthcare costs themselves which might cause omitted variable bias. Therefore, patient characteristics (PCi ) and disease characteristics (𝐷𝐶𝑖)
of patient i were added to the regression. Due to the fact that only baseline data was used in this analysis ICi, PCi and 𝐷𝐶𝑖 are assumed to remain constant over time. Therefore, only the dependent variable 𝐻𝐶𝑖𝑡 and the main independent variable Trit vary over time. Lastly, 𝜀𝑖𝑡 indicates the residual term.
4.3 Interaction Effects To assess whether patient characteristics influence the effect of multidisciplinary spinal care on back pain related healthcare costs, the following model was used:
(2) ln (𝐻𝐶𝑖𝑡) = 𝛽1𝑖+ 𝛽2𝐼𝐶𝑖+ 𝛽3𝑇𝑟𝑖𝑡+ 𝛽4𝑃𝐶𝑖+ 𝛽5 𝐷𝐶𝑖+ 𝛽6𝑃𝐶𝑖𝑥𝑇𝑟𝑖𝑡 + 𝛽7 𝐷𝐶𝑖𝑥𝑇𝑟𝑖𝑡+ 𝜀𝑖𝑡
The aforementioned pooled OLS model with cluster robust standard errors was altered to include individual characteristics, an interaction term consisting of the treatment dummy and a patient or disease characteristic and the constitutive terms of the interaction term. Based on the reviewed literature, I expect lower pain intensity, higher rated work-related factors, better coping skills and better mental health at baseline to increase the negative effect of multidisciplinary spinal care on back pain related healthcare costs. I expect other sociodemographic and physical variables, such as age, gender, bmi and education, to have no or little influence on treatment effect on healthcare costs.
13
(3) ln (𝐻𝐶𝑖𝑡) = 𝛽1𝑖+ 𝛽2𝑇𝑟𝑖𝑡+ 𝜀𝑖𝑡
An F-test was performed and rejected the null hypothesis of no fixed effects at a one percent significance level. The fixed effects model employed in this analysis does not estimate coefficients of time-invariant variables, because the coefficient estimates depend only on the variation of the dependent and explanatory variables within individuals. Therefore, this model only estimates the effect of multidisciplinary spinal care (𝛽2) on back pain related healthcare costs. Similar to the pooled OLS model, the log of back pain related healthcare costs (ln (𝐻𝐶𝑖𝑡) was used. Furthermore, Trit indicates 0 if the patient i was pre-treatment at time t and 1 if
patient i was post-treatment at time t and 𝜀𝑖𝑡 indicates the residual term. I expect the effect of treatment (𝛽2) on back pain related healthcare costs to remain negative and significant in this model specification.
Lastly, when looking at the trend in healthcare costs in Figure 1, there might be an indication of regression to the mean. If patients have been admitted to multidisciplinary spinal care based on high back pain related healthcare costs one year pre-treatment, regression to the mean is likely to occur, which entails that on average a second measure of healthcare costs will be less than the first measure (Galton, 1886). As a result of regression to the mean, the aforementioned models would overestimate the treatment effect. To increase the internal validity of the results adjusted patient samples were used in both models to compare solely the patients two years pre-treatment to two and three years post-treatment. If these comparisons also indicate that mean post-treatment back pain related healthcare costs are significantly lower than mean pre-treatment back pain related healthcare costs, the result is unlikely to be driven by regression to the mean, assuming that patients are not admitted to multidisciplinary spinal care based on high healthcare costs two years pre-treatment.
5. Empirical Results
14
post-treatment and increasing to a significant difference of 377.64 euros, or 32 percent, compared to three years post-treatment.
Table 4: Paired Sample T-Test of Back Pain Related Healthcare Costs
Note: */**/*** stands for statistical significance at the 10%/5%/1% level. The number of patients included in the paired sample t-test is between brackets.
These results confirm the indication from the preliminary view that healthcare costs are lower post-treatment compared to one year pre-treatment, as well as increasing pre-treatment and decreasing post-treatment, which is likely to be the reason for the non-significant negative difference when comparing two years pre-treatment to one and two years post-treatment that becomes significant when comparing two years pre-treatment to three years post-treatment.
5.2 Pooled OLS Regression Results The estimation results for the pooled OLS regression with cluster robust standard errors are presented in table 5 and more extensive in table 6 (see appendix). To assess the effect of adding control variables to the model, each category of patient and disease characteristics was added separately. The first column contains the regression results including only age, age squared, gender and a treatment dummy. The results of the first column indicate that treatment has a significant negative effect on back pain related healthcare costs. Subsequently, in the second column other sociodemographic characteristics were added to the regression. In the third column psychological and work-related factors were added and lastly disease characteristics were added. As can be seen in table 5, after addition of multiple patient and disease characteristics, treatment continues to have a significant negative effect on back pain related healthcare costs. According to the final pooled OLS regression model, containing all possible controls, treatment decreases back pain related healthcare costs by 45.6 percent.
Time Post-Treatment Compared to -1 Years Compared to -2 Years
15
Table 5: Regression Output: Pooled OLS with Cluster Robust Standard Errors
Note: */**/*** stands for statistical significance at the 10%/5%/1% level. The standard errors are between
brackets.
5.3 Interaction Effects Table 7 (see appendix) contains the results from the pooled OLS regression including the different interaction terms. Each interaction term was added and reported separately in the aforementioned altered pooled OLS model (2).
In accordance with the reviewed literature, the results indicate that gender and education do not significantly influence the effect of treatment on back pain related healthcare costs. Contrary to the reviewed literature, pain intensity, work-related factors, coping skills, and mental health do not significantly interact with the treatment indicator. Furthermore, two sociodemographic variables did have a significant interaction with the treatment indicator, namely age and bmi. Following the method proposed by Brambor, Clark and Golder (2005), figure 2 and 3 show the marginal effect of treatment and the corresponding standard errors across a meaningful range of age and bmi. Figure 2 indicates that age decreases the negative effect of treatment, but that treatment continues to have a significant negative effect on back pain related healthcare costs over the included range of age. Figure 3 indicates that bmi decreases the negative effect of treatment and that for patients with a bmi over 35 treatment no longer has a significant negative effect on back pain related healthcare costs. The disease characteristics comorbidity and duration of pain pre-treatment also significantly interact with the treatment dummy. Table 8 contains the marginal effects of treatment on back pain related costs and the corresponding standard errors, when dummies for comorbidity and duration are 0 as well as 1. The estimates show that comorbidity and a longer duration of pain pre-treatment, respectively, decrease and increase the negative effect of treatment. The marginal effects of treatment remain significant and negative for both values of both dummies. Lastly, the variables for having financial worries and the EQ-5D value do not significantly interact with the treatment
Back Pain Related Costs
1 2 3 4 Treatment -0.456*** (0.056) -0.457*** (0.056) -0.454*** (0.056) -0.456*** (0.056) Individual
Characteristics YES YES YES YES
Other SocioDemographic
Characteristics NO YES YES YES
Psychological and
Work-Related Factors NO NO YES YES
Disease
Characteristics NO NO NO YES
16
dummy. For each insignificant interaction, marginal effects or marginal plots were computed, which indicated that these variables do not alter the significance of the negative effect of treatment.
Figure 2: Marginal Effect of Multidisciplinary Spinal Care on Back Pain Related Healthcare Costs
Figure 3: The Marginal Effect of Multidisciplinary Spinal Care on Back Pain Related Healthcare Costs -1,2 -1 -0,8 -0,6 -0,4 -0,2 0 20 30 40 50 60 70 Ma rg in al E ff ec t o f T rea tm en t Patient Age 95 % Confidence Interval -1,5 -1 -0,5 0 0,5 1 15 20 25 30 35 40 45 50 Ma rg in al E ff ec t o f T rea tm en t
Body Mass Index
17
Table 8: Marginal Effect of Treatment on Back Pain Related Healthcare Costs
Note: */**/*** stands for statistical significance at the 10%/5%/1% level. The standard errors are between brackets
5.4 Sensitivity Analysis The results of the fixed effects model are presented in table 9. The estimation results for the fixed effects model also resulted in a significant negative effect of treatment on back pain related healthcare costs, indicating a decrease of 52.7 percent. Additionally, table 9 contains the estimates of using the first and second adjusted sample in both a pooled OLS and a fixed effects model. Treatment continuously has a negative effect on back pain related healthcare costs in each regression model. When using the adjusted sample 1, comparing two years pre- to two years post-treatment, the effect is negative but not significant. For this sample a decrease of 6.3 or 10.4 percent was found depending on the regression model. However, when using adjusted sample 2, comparing two years pre- to three years post-treatment the results are negative and significant, with a decrease of 18.2 or 25.9 percent depending on the regression model.
Table 9:Regression Results Sensitivity Analysis
Note: */**/*** stands for statistical significance at the 10%/5%/1% level. The standard errors are between brackets.
The regression results all indicate that multidisciplinary spinal care has a negative effect on back pain related healthcare cost. The negative effect remains when using adjusted samples, but is only significant for adjusted sample 2. These results confirm the indication of the
Dummy Value Treatment
Comorbidity 0 1 -0.583*** (0.079) -0.294*** (0.074) Duration 0 1 -0.312*** (0.100) -0.522*** (0.067) Pooled OLS Regression Fixed Effect Regression Adjusted Sample 1 Pooled OLS Adjusted Sample 2 Pooled OLS Adjusted Sample 1 Fixed Effects Adjusted Sample 2 Fixed Effects Treatment -0.456*** (0.056) -0.527*** (0.049) -0.063 (0.085) -0.182** (0.090) -0.104 (0.092) -0.259** (0.105) Patient and Disease
Characteristics
YES NO YES YES NO NO
18
preliminary view, similar to the paired sample t-test, that healthcare costs are rising pre-treatment and decreasing post-pre-treatment. Furthermore, when using the adjusted samples, the negative effect of treatment on healthcare costs is unlikely to be driven by regression to the mean, assuming patients do not get referred to the Groningen Spine Center on the basis of high healthcare costs two years pre-treatment.
6. Discussion
A preliminary look at the back pain related healthcare costs of patients admitted to multidisciplinary spinal care at the Groningen Spine Center indicated that rising back pain related healthcare costs pre-treatment were lower and decreasing post-treatment. This indication implied a possible cost-saving effect of multidisciplinary spinal care, which was subsequently tested and confirmed using several analytical methods. The methods employed include: a paired sample t-test, a pooled OLS model with cluster robust standard errors to control for observed patient and disease characteristics, a fixed effects model to allow for unobserved individual heterogeneity and adjusted samples to control for regression to the mean.
Although the estimates are unanimously negative, the different methods produce different estimates of the percentage decrease in back pain related healthcare costs due to treatment. The most conservative estimate was an insignificant decrease in back pain related healthcare costs of 6.3 percent. This was computed by comparing patients two years pre-treatment to patients two years post-treatment in a pooled OLS model including all patient and disease controls. The largest difference was found when comparing patients one and two years pre-treatment to patients one and two years post-treatment in the fixed effects model, which estimates a significant decrease of 52.7 percent. The most meaningful and reliable estimate is a significant decrease of 25.9 percent two years pre-treatment compared to three years post-treatment using the fixed effects model. This estimate is unlikely to be driven by regression to the mean, unobserved heterogeneity and is an indication that back pain related healthcare costs continue to decline post-treatment. Considering the mean back pain related healthcare costs two years pre-treatment are 1189 euros, this estimate would imply a decrease of 307,95 euros three years post-treatment compared to two years pre-treatment per patient and a decrease of 307.026,15 euros for the entire patient sample.
19
have cost saving effects. Although for conclusive evidence on cost-effectiveness a control group receiving usual or no care is needed, the significant estimated decrease warrants attention and further trials to further validate the effect of multidisciplinary spinal care on back pain related healthcare costs. If further trials identify a similar decrease in associated healthcare costs, the implementation of multidisciplinary spinal care for complex chronic back pain could significantly decrease the burden of back pain on the healthcare system and society.
Lastly, a study of the predictors of treatment outcome showed that a higher age, a higher bmi and comorbidity decrease the negative effect of multidisciplinary spinal care on back pain related healthcare costs. Furthermore, a longer duration of pain pre-treatment increase the negative effect of multidisciplinary spinal care on back pain related healthcare costs. Treatment has a significant marginal effect across the entire meaningful range of age, comorbidity and duration. However, for patients with a bmi over 35, treatment no longer has a significant effect on back pain related healthcare costs. No significant predictive value of gender, financial worries, pain intensity, EQ-5D value, work-related factors, coping skills and mental health was found. These results are relevant for clinicians in determining subgroups of patients eligible for treatment, which could and should be further explored in subsequent trials.
As previously stated, the main drawback of this paper is that no control group was available that received usual or no care. Although this paper used adjusted samples to make it unlikely that regression to the mean is the sole driver behind the negative effect, it is not possible to exclude with certainty that healthcare costs would not have gone down had treatment not occurred. Further research should aim at finding a (non)-randomized control group to provide further evidence on the effect of multidisciplinary spinal care.
20
dates or whether patients are still registered at Menzis. Subsequent research should take this into account and request the relevant information on these topics.
Lastly, this research paper focused solely on back pain related healthcare costs. However, when looking at non-back pain related healthcare costs no decrease in costs was found. There was even an indication that non-back pain related healthcare costs rise post-treatment. It would be of interest to explore this relationship further and find out what happens to non-back pain related healthcare costs post-treatment.
7. Conclusion
21
References
Andersson, G. B. (1999). Epidemiological features of chronic low-back pain. The Lancet, 354(9178), 581-585.
Bouter, L.M., Koes, B.W., van Tulder, M. W. (1995). A cost-of-illness study of back pain in The Netherlands. Pain, 62(2), 233-240.
Brambor, T. (2006). Understanding Interaction Models: Improving Empirical Analyses. Political Analysis, 14, 63-82.
Breivik, H., Collett, B., Ventafridda, V., Cohen, R., & Gallacher, D. (2006). Survey of chronic pain in Europe: prevalence, impact on daily life, and treatment. European Journal of Pain, 10(4), 287-287.
Carosella, A. M., Lackner, J. M., & Feuerstein, M. (1994). Factors associated with early discharge from a multidisciplinary work rehabilitation program for chronic low back pain. Pain, 57(1), 69-76.
Cassisi, J. E., Sypert, G. W., Salamon, A., & Kapel, L. (1989). Independent evaluation of a multidisciplinary rehabilitation program for chronic low back pain. Neurosurgery, 25(6), 877-883.
Caudill, M., Schnable, R., Zuttermeister, P., Benson, H., & Friedman, R. (1991). Decreased clinic use by chronic pain patients: Response to behavioral medicine intervention. The Clinical Journal of Pain, 7(4), 305-310.
Cook, T. D., Campbell, D. T., & Shadish, W. (2002). Experimental and quasi-experimental designs for generalized causal inference. Boston: Houghton Mifflin.
Dagenais, S., Caro, J., & Haldeman, S. (2008). A systematic review of low back pain cost of illness studies in the United States and internationally. The Spine Journal, 8(1), 8-20. Flor, H., Fydrich, T., & Turk, D. C. (1992). Efficacy of multidisciplinary pain treatment
centers: a meta-analytic review. Pain, 49(2), 221-230.
Foster, N. E. (2011). Barriers and progress in the treatment of low back pain. BMC Medicine, 9(1), 108.
Galton, F. (1886). Regression towards mediocrity in hereditary stature. The Journal of the Anthropological Institute of Great Britain and Ireland, 15, 246-263.
22
Guzmán, J., Esmail, R., Karjalainen, K., Malmivaara, A., Irvin, E., & Bombardier, C. (2001). Multidisciplinary rehabilitation for chronic low back pain: systematic review. BMJ: British Medical Journal, 322(7301), 1511-1516.
Hasenbring, M. (1998). Predictors of efficacy in treatment of chronic low back pain. Current Opinion in Anesthesiology, 11(5), 553-558.
Harris, A. D., McGregor, J. C., Perencevich, E. N., Furuno, J. P., Zhu, J., Peterson, D. E., & Finkelstein, J. (2006). The use and interpretation of quasi-experimental studies in medical informatics. Journal of the American Medical Informatics Association, 13(1), 16-23.
Hays, R. D., Mazel, R. & M. Sherbourne, C. D. (1993). The rand 36‐item health survey 1.0. Health Economics, 2(3), 217-227.
van der Hulst, M., Vollenbroek-Hutten, M. M., & IJzerman, M. J. (2005). A systematic review of sociodemographic, physical, and psychological predictors of
multidisciplinary rehabilitation or, back school treatment outcome in patients with chronic low back pain. Spine, 30(7), 813-825.
Kamper, S. J., Apeldoorn, A. T., Chiarotto, A., Smeets, R. J. E. M., Ostelo, R. W. J. G., Guzman, J., & van Tulder, M. W. (2015). Multidisciplinary biopsychosocial rehabilitation for chronic low back pain: Cochrane systematic review and meta-analysis. BMJ: British Medical Journal, 350, h444.
Karjalainen, K., Malmivaara, A., van Tulder, M., Roine, R., Jauhiainen, M., Hurri, H., & Koes, B. (2001). Multidisciplinary Biopsychosocial Rehabilitation for Subacute Low Back Pain in Working-age Adults. Spine, 26(3), 262-269.
Koes, B. W., Van Tulder, M. W., & Thomas, S. (2006). Diagnosis and treatment of low back pain. BMJ: British Medical Journal, 332(7555), 1430.
Lambeek, L. C., van Tulder, M. W., Swinkels, I. C., Koppes, L. L., Anema, J. R., & van Mechelen, W. (2011). The trend in total cost of back pain in The Netherlands in the period 2002 to 2007. Spine, 36(13), 1050-1058.
Linton, S. J. (1999). Manual for the Örebro Musculoskeletal Pain Screening Questionnaire: the early identification of patients at risk for chronic pain. Department of
Occupational and Environmental Medicine, Örebro Medical Center.
Maetzel, A., & Li, L. (2002). The economic burden of low back pain: a review of studies published between 1996 and 2001. Best Practice & Research Clinical
23
Machado, L. A. C., Kamper, S. J., Herbert, R. D., Maher, C. G., & McAuley, J. H. (2008). Analgesic effects of treatments for non-specific low back pain: a meta-analysis of placebo-controlled randomized trials. Rheumatology, 48(5), 520-527.
Marin, T. J., van Eerd, D., Irvin, E., Couban, R., Koes, B. W., Malmivaara, A, van Tulder M. W. & Kamper, S. J. (2017). Multidisciplinary biopsychosocial rehabilitation for subacute low back pain. The Cochrane Database of Systematic Reviews, 6, CD002193 McCracken, L. M., & Turk, D. C. (2002). Behavioral and cognitive–behavioral treatment for
chronic pain: outcome, predictors of outcome, and treatment process. Spine, 27(22), 2564-2573.
Olshansky, S. J., & Ault, A. B. (1986). The fourth stage of the epidemiologic transition: the age of delayed degenerative diseases. The Milbank Quarterly, 355-391.
Simmons, J. W., Avant, J. W., Demski, J. O. Y. C. E., & Parisher, D. (1988). Determining successful pain clinic treatment through validation of cost effectiveness. Spine, 13(3), 342-344.
Turk, D. C., & Okifuji, A. (2002). Psychological factors in chronic pain: Evolution and revolution. Journal of Consulting and Clinical Psychology, 70(3), 678.
Vos, Theo, Amanuel Alemu Abajobir, Kalkidan Hassen Abate, Cristiana Abbafati, Kaja M. Abbas, Foad Abd-Allah, Rizwan Suliankatchi Abdulkader et al. (2017) "Global, regional, and national incidence, prevalence, and years lived with disability for 328 diseases and injuries for 195 countries, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016." The Lancet 390, no. 10100: 1211-1259. de Vroome, E. M., Uegaki, K., van der Ploeg, C. P., Treutlein, D. B., Steenbeek, R., de Weerd, M., & van den Bossche, S. N. (2015). Burden of sickness absence due to chronic disease in the Dutch workforce from 2007 to 2011. Journal of Occupational Rehabilitation, 25(4), 675-684.
Waddell, G. (2004). The back pain revolution. London, United Kingdom: Elsevier Health Sciences.
24
APPENDIX .
Table 6
Regression Output: Pooled OLS with Cluster Robust Standard Errors
Note: */**/*** stands for statistical significance at the 10%/5%/1% level. The standard errors are between
brackets.
Back Pain Related Costs
25
Table 7
Pooled OLS Model with Interaction Terms (N=997)
Note: */**/*** stands for statistical significance at the 10%/5%/1% level. The standard errors are between brackets.
Treatment Other Constitutive Term Interaction Term Individual Characteristics Age -1.078*** (0.206) 0.009*** (0.003) 0.012*** (0.004) Gender -0.394 (0.069) -0.475*** (0.099) -0.149 (0.115) SocioDemographic Characteristics BMI -1.324*** (0.305) 0.019** (0.009) 0.032*** (0.011) Education High -0.431*** (0.063) -1.332 (0.115) -0.120 (0.141) Financial Worries -0.429*** (0.060) 0.203 (0.137) -0.190 (0.170) Work Related Factors
