Insurer and out-of-pocket costs of osteoarthritis in the US: Evidence from national survey data

Authors


Abstract

Objective

Osteoarthritis (OA) is a major debilitating disease affecting ∼27 million persons in the US. Yet, the financial costs to patients and insurers remain poorly understood. The purpose of this study was to quantify by multivariate analyses the relationships between OA and annual health care expenditures borne by patients and insurers.

Methods

Data from the Medical Expenditure Panel Survey (MEPS) for the years 1996–2005 were used. MEPS is a large, nationally representative US database that includes information on health care expenditures, medical conditions, health insurance status, and sociodemographic characteristics. Individual and nationally aggregated cost estimates are provided.

Results

OA was found to contribute substantially to health care expenditures. Among women, OA increased out-of-pocket (OOP) expenditures by $1,379 per annum (2007 dollars) and insurer expenditures by $4,833. Among men, OA increased OOP expenditures by $694 per annum and insurer expenditures by $4,036. Given the high prevalence of OA, the aggregate effects on health care expenditures were very large. OA raised aggregate annual medical care expenditures by $185.5 billion. Of that amount, insurer expenditures were $149.4 billion and OOP expenditures were $36.1 billion. Because of the greater prevalence of OA in women and their more intensive use of health care, total expenditures for this group accounted for $118 billion, or almost two-thirds of the total increase in health care expenditures resulting from OA.

Conclusion

The health care cost burden associated with OA is quite large for all groups examined and is disproportionately higher for women. Although insurers bear the brunt of treatment costs for OA, the OOP costs are also substantial.

Osteoarthritis (OA) is a common and debilitating disease that affects ∼27 million people in the US (1). Given an aging population, the prevalence and costs associated with OA are projected to increase. Forecasts indicate that by the year 2030, 25% of the adult US population, or nearly 67 million people, will have physician-diagnosed arthritis (2). It is thus important from a policy perspective to quantify the direct health care costs associated with this disease.

While a number of studies have estimated the direct costs of OA (3–12), existing studies using US data are regionally based, and the results may not be generalizable. These studies typically use data obtained during the early 1990s. In reviewing the literature on the costs of OA, Xie et al (13) noted that “… cost of OA studies were insufficiently performed in the past decade …” Existing studies typically did not perform multivariate analyses to better isolate the effects of OA on health expenditures. While studies have adjusted for age and sex in comparing cohorts of OA and non-OA patients, Maetzel (14) noted that such matching “… is unlikely to weed out the costs that are attributable to other comorbidities, unless they have been adjusted for.”

Evidence on the direct costs of OA varies greatly. In their literature review, Xie et al (13) noted that direct costs from OA varied 10-fold among studies in the US. Wide variations across other countries were observed as well. These striking variations reflect a host of factors, including different geographic regions assessed, different health care systems across countries, different types of databases used, and different items included in the cost estimates.

In the present study, we sought to contribute to the literature on the direct costs of OA in a number of respects. First, we used nationally representative data for the US population to obtain more generalizable results. Second, we examined a more recent time period than did the earlier studies addressing this topic. Third, we estimated the effect of OA on health care expenditures while controlling for other major comorbidities and sociodemographic and economic characteristics. Fourth, we obtained separate estimates of the costs of OA for men and women. Given sex-specific differences in health care utilization patterns generally, it is important to examine sex-specific differences in OA costs. Fifth, we estimated individual out-of-pocket (OOP) and insurer costs of OA to ascertain how the cost burden of OA is shared between insurers and patients.

METHODS

Data.

Our analyses used pooled data for the years 1996–2005 from the Medical Expenditure Panel Survey (MEPS). Information in the MEPS database is collected by the Agency for Healthcare Research and Quality. The database provides nationally representative estimates of health care utilization and expenditures, health status, health insurance coverage, and sociodemographic and socioeconomic characteristics of the civilian noninstitutionalized population in the US. The MEPS sample was selected from a nationally representative subsample of the ongoing National Heath Interview Survey (NHIS) conducted by the National Center for Health Statistics and may be linked to the NHIS database.

The MEPS surveys have been conducted annually since 1996 and have achieved response rates of ∼75%; all respondents were interviewed in person (15). Our sample included adults ages 18 years and older who had health insurance. We limited our analysis of the insured population to focus on the relative cost burden borne by insurers and consumers. Samples of 84,647 women and 70,590 men were used in logistic regressions to predict the probability of having positive health care expenditures, while observations for 74,603 women and 53,890 men were used in multivariable regressions predicting the amount of health care expenditures among those who have positive health care expenditures.

The MEPS database has a complex survey design that, beyond stratifying by sampling units, includes clustering and oversampling of certain subgroups, such as ethnic minorities. Therefore, our statistical analyses used weights provided in the MEPS to obtain the correct standard errors.

Dependent variables.

We examined the effects of OA on the overall annual expected medical care expenditures. Expenditures for physician, hospital, and outpatient services, as well expenditures for drugs, diagnostic testing, and related medical services were included. We estimated the probability that an individual incurred any medical expenditures during the year as well as conditional expenditures (e.g., expenditures among subjects incurring positive expenditures). We performed separate estimates for OOP expenditures by the patients and for expenditures by the insurers.

To estimate the probability of incurring OOP expenditures, we used a binary variable equal to 1 if the patient incurred OOP expenditures and 0 if not. The probability of insurer costs was estimated using a binary variable equal to 1 if the patients' insurer paid for any medical costs during the year and 0 if not. Conditional OOP expenditures were measured as the natural logarithm of these expenditures, and conditional insurer expenditures were also measured in logarithmic form. The logarithmic transformation was used to help normalize the distributions of the conditional expenditure variables.

Explanatory variables.

Explanatory variables predicting the outcomes of interest included major medical conditions, sociodemographic factors, and regions and years.

Medical conditions.

Our main explanatory variable of interest was a binary variable equal to 1 if the subject had OA and 0 if not. To compare the effects of OA with those of other major chronic illnesses and to control for confounding comorbid conditions, we also included binary variables indicating whether the subject had or did not have (variable of 1 or 0, respectively) each of the following conditions: hypertension, diabetes mellitus, hyperlipidemia, anxiety disorders, or asthma.

Sociodemographic factors.

The sociodemographic factors we investigated included binary variables indicating age group, education, ethnicity, marital status, and type of health insurance. Age variables were grouped as follows: 18–34 years old, 35–49 years old, 50–64 years old, 65–79 years old, or ≥80 years old. Subjects in the age group 35–49 years served as the reference group. Education was measured by binary variables indicating 1 of the following: less than high school, some high school, some college, college graduate or above, with high school serving as the reference group. Ethnicity variables included African American, Hispanic, and other non-Caucasian, with Caucasian serving as the reference cohort. Marital status was measured as a variable indicating whether the subject was currently married or not. We also included a binary variable equal to 1 if the subject had public health insurance and 0 if privately insured.

Regions and years.

Region variables included whether the subject was living in an urban location, as well as whether the subject was living in 1 of the 4 US Census regions (Northeast, Midwest, South, or West, with the Northeast serving as the reference region). Binary variables measuring the year during which the observation was made (from 1996 through 2005) were also included, with 1996 serving as the reference year.

Estimation methods.

We estimated 2-part models (16), in which incurring any expenditures and the natural logarithm of the conditional expenditures were estimated separately by logistic regression and ordinary least squares regression, respectively. The 2-part model is frequently used in health services and health economics studies when many observations are clustered at zero and the remaining observations are skewed to the right (17, 18).

Our logistic regression models were of the form shown in equation 1:

equation image

where ProbOOPExp represents the probability of OOP expenditures, as a binary variable: 1 if the patient incurred OOP medical expenses and 0 if not; OA represents the binary indicator of osteoarthritis; Comorbid represents a vector of other comorbid conditions; X represents a vector of sociodemographic, region, and year variables; α0, α1, β, and Θ represent coefficients to be estimated; and ϵ represents an error term.

The same specification as in equation 1 was used to estimate the probability of insurer expenditures (ProbInsExp). In the second stage, we estimated conditional expenditures as shown in equation 2:

equation image

where lnOOPExp represents the natural logarithm of OOP health care expenditures; the other terms were defined as listed above. The same model as in equation 2 was used to estimate conditional insurer expenditures (lnInsExp). Equations 1 and 2 were estimated separately for men and women.

Health care expenditures were expressed in 2007 dollars using the Medical Care Component of the Consumer Price Index. All models were estimated using SAS software, version 9.1 (SAS Institute, Cary, NC).

RESULTS

Descriptive statistics for the study variables are given in Table 1. Annual insurer health care expenditures (in 2007 dollars) were $3,108.698 for women and $3,040.444 for men. Annual OOP expenditures are also higher for women ($770.077) than for men ($612.120). As indicated in Table 1, women also had a higher probability of incurring OOP expenditures and insurer expenditures than did men.

Table 1. Dependent and explanatory variables and descriptive statistics*
VariableWomen (n = 74,603)Men (n = 53,890)
  • *

    Except where indicated otherwise, values are the prevalence. These descriptive statistics pertain to the sample used in the insurer cost regression estimates. Descriptive statistics for other samples are excluded in the interest of brevity but are available upon request from the authors. Expenditures are expressed in 2007 dollars. OOP = out of pocket.

Dependent variable  
 Total insurer expenditures, $3,108.6983,040.444
 Total OOP expenditures, $770.077612.120
 Probability of insurer costs0.8810.763
 Probability of OOP costs0.8870.774
Explanatory variables  
 Medical conditions  
  Osteoarthritis0.0140.007
  Hypertension0.2040.209
  Diabetes mellitus0.0660.079
  Hyperlipidemia0.0900.125
  Anxiety disorders0.0090.006
  Asthma0.0520.033
 Sociodemographic factors  
  Age  
   18–34 years0.2790.248
   35–49 years0.3050.318
   50–64 years0.2200.246
   65–79 years0.1440.149
   ≥80 years0.0520.039
  Education  
   Less than high school0.0560.055
   Some high school0.1130.111
   High school graduate0.3300.305
   Some college0.2420.221
   College graduate or above0.2590.308
  Ethnicity  
   Caucasian0.7770.807
   African American0.0990.076
   Hispanic0.0730.067
   Other non-Caucasian0.0510.050
  Married0.4980.583
  Public health insurance0.1260.090
 Regions and years  
  Census region of residence  
   Living in an urban area0.8160.814
   Northeast0.2050.204
   Midwest0.2420.247
   South0.3410.331
   West0.2120.218
  Year of observation  
   19960.1040.104
   19970.0980.096
   19980.1010.099
   19990.1030.100
   20000.1040.100
   20010.1090.106
   20020.1100.109
   20030.0900.095
   20040.0900.095
   20050.0910.096

In both men and women, hypertension was the most common of the chronic conditions examined. The prevalence of the chronic conditions in our sample was low relative to the population estimates. These condition codes reflect subjects' responses to specific questions about whether they believed that they had the condition. Thus, those with undiagnosed hypertension, for example, would likely be included among the cohort without hypertension.

Results of logistic regression analyses of the estimated odds of incurring insurer and OOP expenditures are provided in Table 2. The presence of OA significantly raised the odds of incurring insurer expenditures, both for men and for women. In women, this effect was particularly large, ranking second only to hyperlipidemia. Having OA also significantly increased the odds of incurring OOP expenses. Again, this effect was particularly large for women. This could indicate that women have less generous health insurance benefits and must therefore pay more OOP or that women are more willing than men to pay OOP for OA treatment.

Table 2. Logistic regression analysis of the probabilities of positive insurer costs and positive out-of-pocket costs per annum
VariableLogistic regression, odds ratio*
Total insurer costs >0Total out-of-pocket costs >0
WomenMenWomenMen
  • *

    All odds ratios were statistically significant at the 1% level.

  • The following reference groups were used: for age, 35–49 years; for education, high school; for ethnicity, Caucasian; for marital status, not married; and for health insurance, private insurance.

  • The reference group for the Census region of residence was the Northeast, and the reference group for the year of observation was 1996.

Medical conditions    
 Osteoarthritis5.8046.24410.9773.126
 Hypertension5.3666.9899.25213.237
 Diabetes mellitus5.0247.5736.88912.187
 Hyperlipidemia6.2038.5865.33010.131
 Anxiety disorders4.4016.9744.1378.117
 Asthma4.8594.5185.1626.174
Sociodemographic factors    
 Age    
  18–34 years1.0100.7620.9730.767
  50–64 years1.0901.2421.3551.417
  65–79 years1.1682.0161.6912.649
  ≥80 years2.0014.4732.2325.261
 Education    
  Less than high school0.8170.7900.7810.720
  Some high school0.9001.0810.7981.015
  Some college1.2771.3391.3891.337
  College graduate or above1.7841.6681.9671.857
 Ethnicity    
  African American0.5110.5370.3810.471
  Hispanic0.5650.5240.4980.534
  Other non-Caucasian0.4400.5910.3710.556
 Married1.2141.2101.1271.038
 Public health insurance1.2011.2280.5150.724
Regions and years    
 Census region of residence    
  Living in an urban area1.0841.1040.9190.992
  Midwest1.0940.9781.2011.049
  South0.9040.8021.2121.025
  West1.0160.9671.0860.994
 Year of observation    
  19970.8970.8880.8380.866
  19980.9940.9230.9280.933
  19991.1260.9711.0470.896
  20001.0880.8891.0770.810
  20011.2980.9831.1390.917
  20021.2761.0311.1160.911
  20031.3161.0691.2170.951
  20041.1051.0101.0390.851
  20051.2670.9961.0570.832

Older and better-educated subjects were more likely to incur both OOP and insurer expenditures. This may reflect greater need for health care among older subjects and greater access to health care among more educated individuals. African Americans, Hispanics, and other non-Caucasians were less likely to incur OOP or insurer expenditures than were Caucasians (the reference group).

Ordinary least squares estimates of insurer and OOP expenditures are given in Table 3. In men and in women, OA was associated with significantly higher insurer expenditures. The marginal impact of OA on insurer expenditures was exceeded only by anxiety disorders in women and by anxiety disorders and diabetes mellitus in men. OA also significantly raised OOP costs for men and women, although the magnitudes of these effects were somewhat smaller than for the other conditions we examined. This may reflect less generous coverage for treatments, tests, and procedures related to OA as compared with the other conditions we examined.

Table 3. Ordinary least squares analysis of total insurer costs and total out-of-pocket costs per annum
VariableOrdinary least squares (coefficient)
Total insurer costs (natural log)Total out-of-pocket costs (natural log)
WomenMenWomenMen
  • *

    Significant at the 1% level.

  • The following reference groups were used: for age, 35–49 years; for education, high school; for ethnicity, Caucasian; for marital status, not married; and for health insurance, private insurance.

  • Significant at the 5% level.

  • §

    The reference group for the Census region of residence was the Northeast, and the reference group for the year of observation was 1996.

Intercept6.457*6.135*5.502*5.180*
Medical conditions    
 Osteoarthritis0.727*0.627*0.482*0.363*
 Hypertension0.423*0.543*0.436*0.466*
 Diabetes mellitus0.727*0.726*0.582*0.652*
 Hyperlipidemia0.445*0.543*0.420*0.495*
 Anxiety disorders0.959*1.225*0.623*0.918*
 Asthma0.635*0.596*0.436*0.469*
Sociodemographic factors    
 Age    
  18–34 years−0.138*−0.320*−0.240*−0.273*
  50–64 years0.236*0.456*0.401*0.460*
  65–79 years0.309*0.821*0.724*0.796*
  ≥80 years0.594*1.155*0.944*1.099*
 Education    
  Less than high school0.004−0.051−0.188*−0.124*
  Some high school0.0040.070*−0.119*−0.052*
  Some college0.028−0.0150.157*0.074*
  College graduate or above0.084*−0.0430.304*0.178*
 Ethnicity    
  African American−0.240*−0.207*−0.666*−0.496*
  Hispanic−0.221*−0.328*−0.445*−0.383*
  Other non-Caucasian−0.350*−0.276*−0.579*−0.419*
 Married0.060*−0.013−0.073*−0.115*
 Public health insurance0.260*0.275*−0.340*−0.055
Regions and years§    
 Census region of residence    
  Living in an urban area0.071*0.032−0.032*−0.061*
  Midwest0.049*0.0310.120*0.081*
  South−0.059*−0.0390.261*0.205*
  West0.0350.0190.044*0.023
 Year of observation    
  1997−0.043−0.026−0.008−0.011
  19980.029−0.027−0.010−0.077*
  19990.0330.000−0.023−0.106*
  20000.010−0.017−0.001−0.111
  20010.149*0.113*0.120*0.014
  20020.167*0.139*0.129*0.026
  20030.191*0.155*0.185*0.142*
  20040.268*0.200*0.141*0.122*
  20050.235*0.137*0.119*0.076*

Using the model estimates from Table 2, we calculated the predicted probabilities of incurring OOP and insurer health expenditures per year (Table 4). OA increased the likelihood of insurer and OOP expenditures for women from ∼88% to nearly 99%. For men, the likelihood of insurer expenditures rose from ∼77% to almost 95%, while the probability of OOP expenditures increased from ∼79% to ∼91%.

Table 4. Effects of OA on the predicted probability of incurring expenditures per annum*
 With OAWithout OADifference
  • *

    Values are the percentage. OA = osteoarthritis; OOP = out of pocket.

Women   
 Insurer expenditures98.76788.44610.321
 OOP expenditures98.76789.1539.614
Men   
 Insurer expenditures94.69276.84317.849
 OOP expenditures90.98578.53912.446

Table 5 shows the expected annual OOP and insurer expenditures as calculated using the model results reported in Table 3. We can see that the presence of OA increased the annual insurer costs by $4,833 in women and by $4,036 in men. Annual OOP expenditures increased by $1,379 in women with OA and by $694 in men with OA.

Table 5. Effects of OA on expected health care expenditures per annum*
 With OAWithout OADifference
  • *

    Values are expressed in 2007 dollars. OA = osteoarthritis; OOP = out of pocket.

Women   
 Insurer expenditures8,5183,6854,833
 OOP expenditures2,4531,0741,379
Men   
 Insurer expenditures7,1233,0874,036
 OOP expenditures1,7381,044694

To generate aggregate expenditures, we relied upon arthritis prevalence rates reported in the literature. The MEPS survey data on chronic conditions were based on each subject's recall of the information queried and therefore cannot provide a reliable estimate of prevalence rates (19). We obtained arthritis prevalence rates for adult men and women from the report by Helmick et al (1). We also adjusted the prevalence rates downward by 5% to eliminate patients with rheumatoid arthritis (1, 20). We then used information on the US adult insured population to obtain population estimates of the numbers of adult men and women with OA (21). Combining this information with the expenditure results from Table 5, we obtained aggregate annual insurer and OOP expenditure estimates for OA according to sex (Table 6).

Table 6. Effects of osteoarthritis on aggregate expected health care expenditures per annum*
 Out-of-pocket expendituresInsurer expendituresTotal
  • *

    Values are expressed in billions of dollars (2007 dollars).

Women26.291.8118.0
Men9.957.667.5
Total36.1149.4185.5

As indicated in Table 6, health care expenditures related to OA were very large. OA increased insurer expenditures by $149.4 billion and OOP expenditures by $36.1 billion annually, for an aggregate increase of $185.5 billion per year. Women accounted for more of these expenditures ($118.0 billion) than did men ($67.5 billion), reflecting the higher prevalence of this disease among women and the higher per capita health expenditures related to OA among women.

DISCUSSION

Understanding the economic costs of OA is important for payors, providers, patients, and other stakeholders. There has been little recent evidence quantifying how OA affects health care expenditures and none demonstrating how these costs are borne by patients and insurers. Yet, such evidence is important for designing health insurance benefits packages, promoting interventions to direct the right treatments to the right patients, and educating patients about the costs they must bear because of this disease.

We found that OA increased health care expenditures dramatically. Aggregate expenditures increased by $185.5 billion per year (in 2007 dollars). Moreover, women account for almost two-thirds of these expenditures. Although insurers bear the brunt of these expenditures, effects on OOP expenditures by the patient are substantial as well.

The prevalence of OA has risen rapidly in recent years, a trend that is expected to continue (2). Unfortunately, OA often goes unnoticed until the disease has progressed (22). Efforts to increase awareness of this disease and to promote better screening will identify patients with OA earlier in the disease course so that interventions can be initiated sooner, which may help to delay progression of the disease and its debilitating effects. Exercise and proper use of medications, such as nonsteroidal antiinflammatory drugs, are known to be effective in treating the symptoms and adverse effects of this disease (23). For more advanced cases affecting major joints, such as the knee or hip, arthroplasty may be indicated. Our results suggest that patients with OA may benefit from greater efforts to promote exercise, proper medication use, and other treatments for the disease.

Our study has some limitations. First, while we have attempted to control for a variety of disease-specific effects, as well as sociodemographic and regional factors that may affect health expenditures, there may be important unobserved factors that we have been unable to control for. Second, while the MEPS database allows one to control for a number of confounders and to generate nationally aggregated estimates, its large sample size dictates that the information on medical conditions is based upon survey responses rather than documentation by clinicians. Thus, the data on medical conditions, including OA, are based upon patients' responses to queries, which are subject to recall bias, and may exclude those whose condition is undiagnosed as well as those who are otherwise unaware of their condition. Third, our study considered only direct medical care expenditures associated with OA. Other direct health care expenditures, such as physical and occupational therapy and alternative medicine interventions, were not included. We anticipate that including these additional factors would further increase the estimated health care expenditures related to OA.

In their review of the literature on studies that have attempted to quantify costs associated with OA, Xie et al (13) noted considerable variations in the cost estimates. They traced such variation to a number of causes, including the tendency to use regional databases that lack generalizability, recall bias in survey-based studies, differences in the cost components considered, and differences in health care systems across countries. Given these data issues, it is not surprising that experts have noted the considerable challenges in generating national cost estimates related to OA (14, 24). We have attempted to address some of these concerns by analyzing nationally representative data and by explicitly stating the expenditure components that were included and the limitations of our data.

Increasingly, third-party payors as well as patients are confronted with difficult health care choices. Given scarce health care resources, health plans should consider covering the most cost-effective and beneficial interventions. Consumers desire that their health care expenditures provide them the maximum possible health benefits. Understanding the economic burden of major diseases is important for informed decision-making. The findings of our study indicate that direct health care expenditures for OA are extremely large. Both payors and patients may benefit from earlier detection of this disease and increased efforts to delay or minimize the adverse and costly consequences of this illness.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Rizzo had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Kotlarz, Gunnarsson, Fang, Rizzo.

Acquisition of data. Kotlarz, Fang, Rizzo.

Analysis and interpretation of data. Kotlarz, Gunnarsson, Fang, Rizzo.

Ancillary