Background

In 2008, according to the Medical Expenditure Panel Survey (MEPS), about 100 million adults in the United States were affected by chronic pain, including joint pain or arthritis. For those who suffer pain, it limits their functional status and adversely impacts their quality of life. Pain is costly to the nation because it sometimes requires medical treatment. Pain also complicates medical care for other ailments, and it hinders one’s ability to work and function in society.

Objective

We estimated (1) the annual economic costs of pain in the United States and (2) the annual costs of treating patients with a primary diagnosis of pain.

Data

We used the 2008 MEPS to compute the economic costs of pain in the United States. The analytic sample was restricted to adults, ages 18 years or older, who were civilians and noninstitutionalized. To compute the annual economic cost of pain, we defined persons with pain as those who reported having “severe pain,” “moderate pain,” “joint pain,” “arthritis,” or functional limitation that restricted their ability to work. To compute the cost of medical care for patients with a primary diagnosis of pain, we examined adults who were treated for headache, abdominal pain, chest pain, and back pain in 2008.

Methodology

The annual economic costs of pain can be divided into two components: (1) the incremental costs of medical care due to pain, and (2) the indirect costs of pain due to lower economic productivity associated with lost wages, disability days, and fewer hours worked. We estimated the incremental and indirect costs using two-part models consisting of logistic regression models and generalized linear models. We also used different model specifications for sensitivity analysis and robustness. To compute the annual costs of medical treatment for patients with a primary diagnosis of pain, we summed the expenditures for medical encounters for headache, abdominal pain, chest pain, and back pain. We converted the cost estimates into 2010 dollars using the Medical Care Inflation Index of the Consumer Price Index (CPI) for medical costs and the General CPI for wages.

Results

We found that the total incremental cost of health care due to pain ranged from $261 to $300 billion. The value of lost productivity is based on three estimates: days of work missed (ranging from $11.6 to $12.7 billion), hours of work lost (from $95.2 to $96.5 billion), and lower wages (from $190.6 to $226.3 billion). Thus, the total financial cost of pain to society, which combines the health care cost estimates and the three productivity estimates, ranges from $560 to $635 billion. All estimates are in 2010 dollars.

Conclusion

We found that the annual cost of pain was greater than the annual costs in 2010 dollars of heart disease ($309 billion), cancer ($243 billion), and diabetes ($188 billion) and nearly 30 percent higher than the combined cost of cancer and diabetes.

Sample

We used the 2008 MEPS to examine the economic burden of pain in the United States. Cosponsored by the Agency for Health care Research and Quality and the National Center for Health Statistics, the MEPS is a nationally representative longitudinal survey that covers the U.S. civilian noninstitutionalized population (Cohen et al., 1996–1997). For this analysis, we used the Household Component (HC) file of the MEPS—the core component of the survey that collects data on demographic characteristics, health expenditures, health conditions, health status, utilization of medical services, access to care, health insurance coverage, and income for each person surveyed. We combined data from the HC file with data from the Condition and Event files of the MEPS to capture the different pain management services used and associated direct medical costs. The analytic sample for the analysis of incremental health care costs was restricted to 20,214 individuals aged 18 or older. This sample is representative of all noninstitutionalized civilian adults in the United States. The analytic sample for the analysis of indirect costs was restricted to 15,945 individuals aged 24–65 to capture the active labor force in the United States. The analysis of direct medical costs was conducted at the event level. We scanned the Event files for diagnosis of pain and the Prescribed Medicine file for pharmaceuticals used to treat pain. Specifically, we identified medical expenditures associated with headache, abdominal pain, nonspecific chest pain, and back pain that occurred in several settings, including physician and nonphysician office-based visits, hospital outpatient visits, emergency department visits, and hospital inpatient stays. We also identified expenditures associated with prescription drugs. We summed the costs of medical encounters for these diagnoses and the costs of medications used to treat pain.

Key Independent Variables

We defined persons with pain as those who reported that they experienced pain that limited their ability to work, that they were diagnosed with joint pain or arthritis, or that they had a disability that limited their ability to work. The SF-12 pain question of the MEPS asked the respondent whether, during the past 4 weeks, pain interfered with normal work outside the home and housework. The joint pain question inquired whether the person had experienced pain, swelling, or stiffness around a joint in the last 12 months. The question for arthritis determined whether the person had ever been diagnosed with arthritis. The question about functional disability inquired whether the person had any work or housework limitation. We explored whether we could use information from the Event files on persons who were diagnosed with a headache, abdominal pain, chest pain, or back pain. We identified relatively few persons who had medical encounters in which pain was the primary diagnosis. Consequently, we decided not to use the Event files to determine the prevalence of pain in the population. Rather, we expected that persons suffering from these pain conditions would report having moderate or severe pain on the SF-12.

Dependent Variables

We used total expenditures as the dependent variable to predict the incremental costs of care for individuals with selected pain conditions compared with those without these conditions. Total expenditures in the MEPS include both out-of-pocket and third-party payments to health care providers but do not include health insurance premiums. Expenditures for hospital-based services include those for both facility and separately billed physician services. Total expenditures include inpatient, emergency room, outpatient (hospital, clinic, and office-based visits), prescription drugs, and other (e.g., home health services, vision care services, dental care, ambulance services, diagnostic services, medical equipment). The expenditures do not include over-the-counter purchases.

For the analysis of indirect costs, we used the annual number of days of work missed because of pain conditions, the annual number of hours of work missed because of pain conditions, and hourly wages as dependent variables to predict the productivity loss associated with the different pain conditions. Variations in the annual number of days of work missed measure workers’ decisions to use sick days. Variations in the annual number of hours worked measure workers’ decisions whether to work full time, part time, or overtime. Variations in the hourly earnings measure the value of the amount of work workers can perform in an hour.

Control Variables

We used a modified version of Aday and Andersen’s (1974) behavioral health model of health services to estimate direct medical costs for patients with pain compared with those without any pain. This model hypothesizes that health expenditures depend on predisposing, enabling, and health need factors. In this conceptual framework, pain is a health need factor. We estimated the association between pain and health care expenditures. We predicted health care expenditures using demographic, socioeconomic status, health behavior, location, and health need measures. The demographic factors were age, gender, race, and marital status. The socioeconomic factors were education, income, and health insurance status. To measure health behaviors, we used whether respondents smoked or exercised and their obesity status. Census region and urban/rural residence were used to measure location. To measure health needs, we used whether respondents reported they were in fair or poor health and whether they had been diagnosed with diabetes or asthma. Diabetes and asthma were included because they may complicate the treatment of other conditions, and we did not want to attribute these costs to the incremental medical costs of pain. We excluded other chronic conditions, including hypertension, heart disease, emphysema, and stroke because we were concerned about the potential correlation between these other chronic conditions and the SF-12 measures of pain. We estimated preliminary models with the full complement of chronic conditions; however, some conditions were statistically insignificant. Therefore, we elected to use the most parsimonious models that adequately controlled for health needs.

The lost productivity computation was based on the human capital approach of estimating labor supply and earning models (Becker, 1973, 1974; Killingsworth, 1983). Theoretically, hours worked, wages, and labor force participation are based on a set of factors, including age, sex, race, ethnicity, education, health status, and location. We also included the size of the family the person lives with to capture some of the household characteristics that are associated with labor market outcomes.

Health Care Expenditure Models

We estimated a standard two-part expenditure model to address issues of sample selection and heterogeneity and computed the economic burden for patients with the different types of pain conditions noted above compared with those without any pain (Manning, 1998; Mullahy, 1998; Manning and Mullahy, 2001; Buntin and Zaslavsky, 2004; Deb et al., 2006; Cameron and Trivedi, 2008). The first part of the model consisted of estimating logistic regression models to estimate the probability of having any type of health care expenditures. The second part consisted of using generalized linear models with log link and gamma distribution to predict levels of direct expenditures conditional on individuals with positive expenditures. We used a log link and gamma distribution to address the skew in the expenditure data. We eliminated outliers, i.e., observations with expenditures greater than $100,000. We conducted the different diagnostic and specification tests recommended by Manning (1998), Mullahy (1998), and Manning and Mullahy (2001). We estimated the models using the survey regression procedures in STATA 11, which appropriately incorporates the design factors and sample weights.

We developed three models to predict total health care expenditures and conduct sensitivity analyses for robustness, varying the degree to which we controlled for health status. In the first model, we measured pain with indicators for moderate pain, severe pain, joint pain, and arthritis. We controlled for health status using only self-reported general health status and body mass index. In the second model, we added functional disability to our pain measures. In the third model, we included diabetes and asthma in our measures of health status. We conducted sensitivity analyses using several of the chronic condition indicators available in the MEPS and found that diabetes and asthma were significant predictors of expenditures independently of the pain measures. We estimated models with and without an indicator for functional disability. We were concerned that persons with a functional disability who had chronic pain might not be captured by the other pain measures; however, we were also aware that the functional disability variable might capture people with a functional disability but no chronic pain. By conducting the computation both ways, we could see whether including functional disability in our definition of pain conditions mattered.

We computed the incremental costs of pain by using our model to predict health care costs if a person has any type of pain and subtracting the predicted health care costs if a person does not have pain (Deb et al., 2006). To perform this calculation, the probabilities of having health care costs for persons with and without pain must be taken into account. We computed unconditional levels of health care expenditures by multiplying the probabilities obtained from the first part of the model by predicted levels of expenditures from the second part of the model for individuals with and without pain. Subsequently, we computed the incremental values for each type of pain condition by taking the difference between those with and without pain. We converted the cost estimates into 2010 dollars using the medical care index of the CPI.

We computed the impact of the incremental costs of selected pain conditions on the various payers for health care services. The HC file from the MEPS contains 12 categories of direct payment for care provided during 2008: (1) out-of-pocket payments by users of care or family; (2) Medicare; (3) Medicaid; (4) private insurance; (5) the VA, excluding CHAMPVA; (6) TRICARE; (7) other federal sources (includes the Indian Health Service, military treatment facilities, and other care provided by the federal government); (8) other state and local sources (includes community and neighborhood clinics, state and local health departments, and state programs other than Medicaid); (9) workers’ compensation; (10) other unclassified sources (includes such sources as automobile, home-owner’s, and liability insurance and other miscellaneous or unknown sources); (11) other private (any type of private insurance payments); and (12) other public. For each payer category, we computed its proportion of total health care expenditures. We multiplied our estimate of total incremental health care costs due to pain by these proportions to estimate the impact on each payer.

Indirect Cost Models

As with the health care expenditure models, we used two-part models to estimate the indirect costs of pain. The structure of the models depended upon the dependent variables. For missed days of work, we estimated the probability of missing a work day as a result of selected pain conditions during the year. Second, we estimated a log linear regression model in which the dependent variable was the log of the number of disability days for those adults who had positive disability days.

For hours worked and wages, the first equation estimated the impact of pain on the probability that a person is working. The second equation estimated the impact of pain on the number of annual work hours and hourly wages. Combining the results from these different parts of the models, we computed the productivity costs associated with chronic pain for each of the conditions noted above. We used a standard two-step estimator for labor supply to predict lost productivity due to pain (Greene, 2005; Cameron and Trivedi, 2008). As with the incremental cost models, we multiplied the probabilities obtained from the first part of the model by predicted levels of work days missed, lost work hours, or lost wages from the second part of the model for individuals with and without pain. To compute the total cost of missed days, we multiplied the days missed by 8 hours times the predicted hourly wage rate for individuals with the pain condition. To compute the total cost of reduction of hours worked, we multiplied the total of annual hours missed by the predicted hourly wage rate for individuals with the pain condition. To compute the total cost due to a reduction in hourly wages, we multiplied the predicted hourly wage reduction by the predicted annual hours lost for individuals with the pain condition. We converted the cost estimates into 2010 dollars using the general CPI.

The approach of using a two-part model to estimate lost productivity is similar to the use of Heckman selection models, but can be used in the absence of the identifying variables required by Heckman selection models and other limited dependent variables models, such as the Tobit (see Heckman, 1979; Ettner, 1995). Additionally, we conducted a series of tests to determine the appropriate distribution for each of these models. For instance, we used a log link with Gaussian distribution to estimate the models for hours worked.

Incremental Costs of Health Care

Table C-1 displays the dependent and independent variables used in the analysis of the incremental costs of health care. The sample includes 20,214 individuals aged 18 and older, representing 210.7 million adults in the United States as of 2008. The mean health care expenditures were $4,475, and 85 percent of adults had a positive expenditure. The prevalence estimates for selected pain conditions were 10 percent for moderate pain, 11 percent for severe pain, 33 percent for joint pain, 25 percent for arthritis, and 12 percent for functional disability.

TABLE C-1

Dependent and Independent Variables Used in the Incremental Cost Models for Patients Aged 18 or Older for Selected Pain Conditions (N = 20,214, US$2010).

Adults with pain reported higher health care expenditures than adults without pain (see Table C-2). Based on the SF-12 pain measures, a person with moderate pain had health care expenditures $4,516 higher than those of someone with no pain. Persons with severe pain had health care expenditures $3,210 higher than those of a person with moderate pain. We found similar differences for persons with joint pain ($4,048), arthritis ($5,838), and a functional disability ($9,680) compared with persons without these conditions. All of these differences were statistically significant (p < 0.001).

TABLE C-2

Means of Unadjusted Expenditures for Patients Aged 18 or Older for Selected Pain Conditions (US$2010).

The regression results of the logistic regression models and generalized linear models indicate that moderate pain, severe pain, joint pain, arthritis, and functional disability were strongly associated with an increased probability of having a health care expenditure and with higher expenditures (see Table C-3). The coefficients were all statistically significant and positive predictors of whether a person had a health care expenditure and the amount of that expenditure. The coefficients were relatively stable across the three models. The magnitude of the coefficients declined as we included functional disability, asthma, and diabetes in the models.

TABLE C-3

Results of Two-Part Total Expenditure Models for Patients Aged 18 or Older for Selected Pain Conditions.

To interpret the coefficients on pain conditions, we exponentiated the coefficients in the logistic models to compute the odds ratio (OR) of having a health care expenditure for a person with pain relative to a person without pain. For example, the odds of having a health care expenditure increased by 70 percent for persons with joint pain relative to persons without joint pain (OR = 1.70) according to Model 1. Similarly, because the link function in the generalized linear model is a log, we exponentiated the coefficients on the pain variables to compute the percentage increase in health care expenditure for a person with pain relative to a person without pain. For example, among persons with a health care expenditure, spending for persons with joint pain was 16.2 percent higher than that for persons without joint pain based on Model 1.

The coefficients on the control variables had the expected signs. Women were more likely to have a health care expenditure and a higher expenditure than men. The likelihood of an expenditure and the level of expenditures increased with age. Blacks, Hispanics, and Asians were less likely than whites to have a health care expenditure and had lower expenditures. Socioeconomic and health factors had the expected impact. As education, income, and health insurance status increased, health care spending also increased. Health care spending increased for persons who were obese, who reported they were in fair or poor health, who had asthma, and who had diabetes.

We computed the average and total incremental costs of the selected pain conditions (see Tables C-4 and C-5). The average incremental costs of health care for selected pain conditions ranged from $854 for joint pain to $3,957 for severe pain according to Model 1. When functional disability was included in the model, its incremental costs were $3,787, while the estimates for the other pain conditions declined, particularly for severe pain, which fell to $2,573 in Model 2. We estimated that approximately 100 million persons had at least one of the pain conditions based on the 2008 MEPS. Our estimate correlates well with the national estimate of at least 116 million persons given in the main report because the MEPS excludes persons in nursing homes, prisons, and the military. The most prevalent condition was joint pain, affecting more than 70 million adults. We estimated that the incremental costs of health care for these selected pain conditions ranged from $261 billion to $293 billion annually. The most expensive pain condition was severe pain at $89.4 billion annually. However, functional disability was the most expensive when we included it in the model—$93.5 billion in Model 2. One interesting observation is that the incremental costs of severe pain declined to $58 billion when we included functional disability.

TABLE C-4

Average Incremental Costs of Medical Expenditures for Selected Pain Conditions (US$2010).

TABLE C-5

Total Incremental Costs of Medical Expenditures for Selected Pain Conditions (in millions of US$2010 and millions of persons).

Table C-6 shows the distribution of the incremental costs by source of payment. We estimated that private insurers paid the largest share of incremental costs, ranging from $112 billion to $129 billion. Medicare bore 25 percent of the incremental costs due to pain, ranging from $66 billion to $76 billion. Individuals paid an additional $44 billion to $51 billion in out-of-pocket health care expenditures due to persistent pain. Medicaid paid about 8 percent of the incremental costs of pain, ranging from $20 billion to $23 billion.

TABLE C-6

Distribution of Total Incremental Costs of Medical Expenditures by Source of Payment (in millions of US$2010).

Indirect Costs

Table C-7 shows the dependent and independent variables for the analysis of incremental indirect costs. The sample was 15,945 persons ages 24 to 64, representing 156 million working-age adults. The mean number of work days missed was 2.14, and 46 percent of adults missed at least one day of work. The average number of hours the sample worked annually was 1,601, with 81 percent of adults working. The average hourly wage was $14.19. Among working-age adults, 9 percent reported having moderate pain, 10 percent severe pain, 31 percent joint pain, 21 percent arthritis, and 10 percent a functional disability.

TABLE C-7

Dependent and Independent Variables Used in the Indirect Cost Models for Patients Aged 24–64 for Selected Pain Conditions (N = 15,945).

Adults with pain reported missing more days of work than adults without pain (see Table C-8). A person with moderate pain, based on the SF-12 pain measures, missed 2.1 days more than someone with no pain. Adults with severe pain missed 2.6 days more than those with moderate pain. The differences for joint pain, arthritis, and functional disability were 1.3 days, 1.3 days, and 3.3 days, respectively. Pain was associated with fewer annual hours worked (see Table C-9). Persons with functional disability had the largest difference, working 1,203 fewer hours than persons with no functional disability. Compared with persons with no pain, persons with moderate pain worked 291 fewer hours, and persons with severe pain 717 fewer hours. We found similar differences in hours for joint pain (220) and arthritis (384). Wages were lower for persons with pain (see Table C-10). The largest difference was for persons with functional disability, followed by severe pain, moderate pain, arthritis pain, and joint pain. Persons with functional disability earned $11 an hour less than persons without functional disability.

TABLE C-8

Means of Unadjusted Number of Work Days Missed for Adults Aged 24–64 with Selected Pain Conditions.

TABLE C-9

Means of Unadjusted Number of Hours Worked for Adults Aged 24–64 with Selected Pain Conditions.

TABLE C-10

Means of Unadjusted Number of Hourly Wages for Adults Aged 24–64 with Selected Pain Conditions (US$2010).

The regression results for the indirect cost analysis are reported in Tables C-11, C-12, and C-13. As with the health care cost models, we interpreted the coefficients on the pain measures by exponentiating them. The first step models were logistic regressions, so the exponentiated coefficients on the indicator variables were ORs. The second step models were log-linear using the generalized linear model. Thus, the exponentiated coefficients were percent changes in the dependent variables. For example, in Table C-11, Model 1, the coefficients on moderate pain were 0.5 in the logistic model and 0.49 in the generalized linear model. We interpreted these coefficients as follows. Compared with a person with no pain, someone with moderate pain had 64 percent greater odds of having at least one missed day of work during the year, and having moderate pain increased the number of days missed by 63 percent. Tables C-12 and C-13 display the impact of pain conditions on the likelihood of working, the number of hours worked, and hourly wages. The pain conditions had a significant negative impact on the likelihood of working. The impact on hours worked and wages was negative but modest and in several cases insignificant. This means that the negative impact of pain conditions on hours worked and wages occurred largely through the decision to work or not. Persons with pain were less likely to work than persons without pain.

TABLE C-11

Results of Two-Part Missed Days Models for Persons Aged 24–64 for Selected Pain Conditions.

TABLE C-12

Results of Two-Part Missed Hours Models for Persons Aged 24–64 for Selected Pain Conditions.

TABLE C-13

Results of Two-Part Logistic Regression and Generalized Linear Hourly Wages Models for Adults Aged 24–64 for Selected Pain Conditions.

The calculated incremental costs are reported in Tables C-14 to C-19. The average incremental number of days of work missed was greatest for severe pain, with estimates ranging from 5.0 to 5.9 days. Arthritis caused the fewest days of work missed—0.1 to 0.3. Almost 70 million working adults reported having one of the pain conditions. The annual costs for the number of days missed ranged from $11.6 to $12.7 billion. More persons reported joint pain, but severe pain was more costly. Including functional disability in these models did not affect the estimates for the other pain conditions.

TABLE C-14

Average Incremental Number of Days of Work Missed Because of Selected Pain Conditions.

TABLE C-15

Total Incremental Costs of Number of Days of Work Missed Because of Selected Pain Conditions (in millions of US$2010 and millions of persons).

TABLE C-16

Average Incremental Number of Hours of Work Lost Because of Selected Pain Conditions.

TABLE C-17

Total Incremental Costs of Number of Hours of Work Missed Because of Selected Pain Conditions (in millions of US$2010 and millions of persons).

TABLE C-18

Average Incremental Reduction in Hourly Wages Due to Selected Pain Conditions (US$2010).

TABLE C-19

Total Indirect Costs Associated with Reductions in Wages Due to Selected Pain Conditions (in millions of US$2010 and millions of persons).

Pain also was associated with fewer annual hours worked. For Model 1, severe pain was associated with the largest reduction, 204 hours. However, when we included functional disability in the model, the impact of severe pain fell to 30 hours, while the reduction associated with having a functional disability was 740 hours. While the inclusion of functional disability changed the distribution of the costs, it did not change the overall estimate of the costs associated with fewer annual hours worked, which totaled about $95 to $96 billion.

The average reduction in hourly wages for selected pain conditions ranged from $0.26 an hour for joint pain to $3.76 an hour for severe pain according to Model 1. Including functional disability in the models changed the estimates substantially for the other pain conditions—from $0.05 an hour for joint pain to $1.66 an hour for severe pain. Functional disability was associated with a large reduction in wages ($9.36 an hour), which did impact the total estimate of the costs due to wage reductions. The indirect cost associated with reduced wages was $191 billion for Model 1 but $226 and $217 billion for Models 2 and 3, respectively.

Total Direct Cost for Medical Care for Pain Diagnoses

The direct cost of medical treatment for pain diagnoses was almost $47 billion (see Table C-20). The bulk of these costs was for back pain ($34 billion). Office-based services and hospital stays accounted for 36 percent and 33 percent of the total costs, respectively. The difference between the total direct cost and the total incremental health care costs was $214 to $246 billion. This indicates that most of the health care costs were attributable not to a direct diagnosis of pain but to the impact of pain on the treatment of other conditions.

TABLE C-20

Total Direct Costs for Selected Pain Conditions (in millions of US$2010).

In summary, we found that the total incremental costs of health care due to pain ranged from $261 to $300 billion. The value of lost productivity ranged from $11.6 to $12.7 billion for days of work missed, from $95.2 to $96.5 billion for hours of work lost, and from $190.6 to $226.3 billion for lower wages. The total annual costs ranged from $560 to $635 billion.