Ambulatory visit utilization in a national, population-based sample of adults with osteoarthritis

Authors


Abstract

Objective

To estimate the proportion of adults with osteoarthritis (OA) seeing various medical providers and ascertain factors affecting the likelihood of a patient seeing an OA specialist.

Methods

We used data from the Medical Expenditures Panel Survey, a stratified random sample of the noninstitutionalized civilian population. We classified adults as having symptomatic OA if their medical conditions included at least 1 occurrence of the International Classification of Diseases, Ninth Revision Clinical Modification, codes 715, 716, or 719, and if they reported joint pain, swelling, or stiffness during the previous 12 months. For the purpose of our analysis, we defined rheumatologists, orthopedists, and physical therapists as OA specialists. We first estimated the proportion of OA individuals seen by OA specialists and other health care providers in a 1-year period. We then used logistic regression to estimate the impact of demographic and clinical factors on the likelihood of an individual seeing an OA specialist.

Results

A total of 9,933 persons met the definition of OA, representing 22.5 million adults in the US. Of these persons, 92% see physicians during the year, 34% see at least 1 OA specialist, 25% see an orthopedist, 11% see a physical therapist, and 6% see a rheumatologist. Higher educational attainment, having more comorbidities, and residing in the northeastern US are significant positive predictors for a patient seeing an OA specialist. Significant negative predictors for seeing an OA specialist are being unmarried but previously married and having no health insurance.

Conclusion

Most adults with OA do not visit OA specialists. Those without insurance and with lower levels of education are less likely to see these specialists.

INTRODUCTION

Osteoarthritis (OA) is the most prevalent form of arthritis, second only to ischemic heart disease as a cause of US work disability in men age >50 years (1, 2). The most recently published prevalence estimate of OA was 26.9 million among US adults age ≥25 years in 2005 (2).

OA treatment results in substantial direct costs. In 2006, OA was listed as the primary diagnosis for 735,087 hospitalizations (1.9% of all discharges in that year) (3). Ambulatory visits listing OA as the primary diagnosis accounted for 7.1 million ambulatory visits (0.7% of all ambulatory visits) in 1997 (4). The most recent US study of direct costs attributable to OA was restricted to adults ages 18–64 years and presented in 2004 US dollars; the 2006 equivalent per-patient annual total is $4,091 (5, 6). The most recent OA-attributable cost estimate for the elderly population, based on a managed-care sample, represents a per-patient annual total of $4,719 attributable to OA in 2006 dollars (6, 7).

While many of the risk factors for the development and worsening of symptoms in OA cannot be prevented (female sex, increasing age, genetic predisposition), others (excess body mass, activities that promote stress on joints, and injury) are potentially modifiable through interventions such as weight loss and muscle strengthening exercise programs (8). Understanding patterns of care for the population affected by OA is crucial for targeting interventions, but to our knowledge no national population-based study of ambulatory care patterns in OA patients has been published. The objectives of this study were to use a population-based US data source to ascertain the percentage of adults with OA seeing various medical providers, including specialists who provide arthritis-related care (rheumatologists, orthopedists, and physical therapists), and which factors affect the likelihood of individuals with OA seeing specialists focused on arthritis-related care.

MATERIALS AND METHODS

Data source.

We used the Medical Expenditure Panel Survey (MEPS) household component (HC), a nationally representative survey of the US civilian noninstitutionalized population, to estimate the proportion of adults with OA seeing various providers in ambulatory settings. MEPS is a joint endeavor of the Agency for Healthcare Research and Quality and the National Center for Health Statistics. The MEPS-HC collects data on health care use, demographic characteristics, and health status from 5 interviews over a 2-year period (9, 10). The presence of medical conditions is ascertained primarily by prompting HC respondents for the causes of medical events and disability episodes, but also for conditions “bothering” the person during the reference period. Conditions identified by one or more of these methods are then coded using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) system at the 3-digit level (11). For example, patient-reported “generalized osteoarthritis of the hand,” which corresponds to a 5-digit ICD-9-CM code of 715.04, would be coded in MEPS as 715.

The MEPS Medical Provider Component (MPC) collects diagnostic, procedural, and expenditure data, as well as the date, from medical providers for visits made by a subsample of the MEPS-HC participants. The accuracy of some self-reported condition data from MEPS-HC respondents has been assessed by comparison with MEPS-MPC data, using the diagnostic data in the MEPS-MPC as the gold standard (12). MEPS has an overlapping panel design that allows for pooling of estimates from a number of years to provide stable estimates. The present study uses 2002 to 2005 MEPS data.

Definition of OA.

We classified adults as having OA if the following 2 requirements were met: their medical conditions listed in the MEPS-HC included at least 1 occurrence of OA (ICD-9-CM 715), arthropathies and related conditions (ICD-9-CM 716), or other and unspecified disorders of the joint (ICD-9-CM 719), and they reported pain, swelling, or stiffness around a joint during the previous 12 months. Originally, the medical conditions were restricted to ICD-9-CM 715, the discrete code for OA, but this requirement was broadened because patient-reported ICD-9-CM 715 has demonstrated low agreement with ICD-9-CM 715 recorded in the MEPS-MPC subsample (κ = 0.15, 95% confidence interval [95% CI] 0.12–0.19) (12). Recent research has demonstrated that the addition of ICD-9-CM 719 and 716 to the OA definition resulted in substantial increases in OA sensitivity (from 17–81%), with only minor decreases in specificity (from 99–89%), when using ICD-9-CM 715 recorded in the MPC as the gold standard (Murphy L: personal communication). By adding ICD-9-CM codes 716 and 719 to our definition, we estimated that our study population included 81% of all OA adults in the MEPS sample with a doctor-provided diagnosis of OA in a year, but that 19% of individuals included in our study population did not have a doctor-provided diagnosis.

Data elements.

Demographic factors.

Demographic variables included age in years, female sex, Hispanic ethnicity, white race, marital status (married, widowed/separated/divorced, never married), and highest level of education attained (less than high school, high school graduate, some college, college graduate, or some graduate school). Other demographic variables included geographic region of residence (northeast, midwest, south, west), and living within a metropolitan statistical area (MSA).

Clinical characteristics.

Comorbid condition indicator variables were created by flagging ICD-9-CM codes for cancer, ischemic heart disease/chronic heart failure, chronic obstructive pulmonary disease/asthma, depression, diabetes mellitus, and non-OA musculoskeletal disease (see specific ICD-9-CM codes listed in Appendix A). Obesity was defined as a body mass index (BMI) of ≥30 kg/m2. These binary comorbid condition variables were summed to create a variable, which recorded whether the study subject had 0, 1, 2, or >2 comorbid conditions. Binary variables for patient-perceived health status (poor/fair versus excellent/very good/good) (13) and activity limitation (any limitation at work, home, or school versus none) were also created.

Measures of health care utilization.

We combined visits to offices and hospital outpatient facilities to estimate ambulatory care visits. Physician providers included primary care (family practice, general practice, and internal medicine), orthopedists, rheumatologists, and all other specialties. Nonphysician health care providers included nurse/nurse practitioner/physician's assistant, physical therapist, and all other specialties. We defined rheumatologists, orthopedists, and physical therapists as OA specialists.

Statistical analysis.

Regression models.

The dependent variables for these regressions were the odds of subjects seeing each of the OA specialists. Logistic regression was used to ascertain which of the potential demographic and clinical factors were associated with seeing these providers. Three types of regression models were created for each outcome. Univariate analyses were performed with each of the potential explanatory variables as an independent variable. We then created multivariable main-effects models by including all potential factors into a single multivariable model. Because demographic characteristics are known to influence access to care, we decided to implement the following multivariable model building strategy: demographic variables were included in all models regardless of statistical significance, whereas nondemographic covariates required P values less than 0.05 to be retained. The exclusion of nondemographic covariates was done one variable at a time, starting with the least significant, until all nondemographic variables that remained in the model were significant at P less than 0.05. Our final step was to investigate whether important interactions existed between variables included in the main-effects models and any of the following: obesity, activity limitation, age, sex, race, ethnicity, marital status, education status, region, MSA status, and number of comorbidities, all of which we hypothesized a priori could interact with specialist utilization. Interactions that were statistically significant at P less than 0.05, did not include sparse cells (sample n ≤ 10), and added important information to the interpretation of the model when compared with the main-effects model were retained to yield the interaction model for each outcome. While odds ratios (ORs) and 95% CIs were presented in tables for all covariates included in the models, only those statistically significant at α < 0.05 are discussed in the text.

Accounting for complex survey design.

Population sampling weights were applied in all analyses. We used SAS, version 9.2 survey procedures (Surveyfreq, Surveymeans, and Surveylogistic; SAS Institute, Cary, NC) to adjust standard error estimates for the clustered sampling design of the MEPS (14). Annual estimates were obtained by dividing sampling weights in the pooled file by 4, the number of years of MEPS data used.

Missing data.

For the 416 (4%) adult OA observations with missing data, we imputed data using multiple imputation procedures (SAS, version 9.1.3 PROC MI and PROC MIANALYZE, SAS Institute). SAS multiple imputation procedures are implemented on the data missing at random (MAR) assumption, which states that the probability of missing values for any variable of interest is conditioned on the value of other variables in the analysis, not on the value of this variable itself (14). Specific variables imputed were the highest level of education attained, marital status, BMI, self-perceived health status, and joint pain during the year. Because the missing data for these variables were arranged in a general pattern, we used the Markov chain Monte Carlo imputation method.

RESULTS

Population characteristics.

The population for this study included the 9,933 adults in the MEPS meeting the study definition of OA; these adults represent an annual prevalence of 22.5 million or 10% of the adult population. General characteristics of the OA population used in the current analysis are shown in Table 1. Individuals in the cohort had a mean age of 60 years, and 63% were women. Most (78%) identified themselves as non-Hispanic white; 7% were Hispanic, 11% were non-Hispanic African American, while the remaining 4% identified with some other racial group. The majority (77%) resided in an MSA, were married (56%), and had at least some private insurance coverage (65%). Twenty-five percent had not obtained a high school diploma, 35% had graduated from high school, and the remaining 40% had at least some college education. Thirty-three percent perceived their health as fair or poor and 30% reported activity limitation. Fifty-five percent of respondents indicated that pain limited their normal work at least moderately. Obesity was present for 39% of OA respondents, and 72% of study subjects were classified as having at least 1 of the 6 comorbidities.

Table 1. General population characteristics for OA cohorts, MEPS 2002–2005 (pooled), age ≥18 years with joint pain (n = 9,933)*
CharacteristicValue95% CI
  • *

    Values are the number (percentage) unless otherwise indicated. OA = osteoarthritis; MEPS = Medical Expenditure Panel Survey; 95% CI = 95% confidence interval; MSA = metropolitan statistical area; BMI = body mass index; CHF = chronic heart failure; COPD = chronic obstructive pulmonary disease.

  • Individuals could have 0, 1, or ≥2 comorbidities.

Age, years  
 18–441,539 (16)16–16
 45–644,340 (43)43–44
 ≥654,054 (41)41–41
Mean age, years6060–61
Women6,572 (63)63–63
Race/ethnicity  
 Non-Hispanic white6,571 (78)77–78
 Hispanic1,344 (7)7–7
 Non-Hispanic African American1,583 (11)11–11
 Non-Hispanic other435 (4)4–4
Region  
 Northeast1,435 (17)16–17
 Midwest2,152 (24)24–24
 South4,230 (39)39–39
 West2,115 (20)20–21
Living within MSA7,310 (77)77–77
Marital status  
 Married5,261 (56)56–56
 Widowed, separated, divorced3,782 (35)35–35
 Never married890 (9)9–9
Insurance type  
 Private (reference group)5,556 (65)65–66
 Public3,620 (28)28–29
 None757 (6)6–6
Education  
 <High school3,247 (25)25–25
 High school graduate3,282 (35)35–35
 Some college1,863 (21)21–21
 College graduate908 (11)11–11
 Some graduate school632 (8)8–8
Perceived overall health fair/poor3,864 (33)32–33
Activity limitation3,406 (30)30–30
Pain limits normal work  
 Not at all1,542 (18)17–18
 A little bit2,523 (27)27–28
 Moderately2,161 (22)22–22
 Quite a bit2,576 (23)23–24
 Extremely1,130 (10)9–10
Obese, BMI ≥30 kg/m24,109 (39)38–39
Specific high-cost comorbidities  
 Cancer680 (7)7–7
 Ischemic heart disease/CHF1,031 (10)10–10
 COPD/asthma1,599 (15)15–15
 Depression2,916 (27)27–28
 Diabetes mellitus1,928 (17)17–17
 Non-OA musculoskeletal disease4,501 (46)45–46
 None2,658 (28)28–28
 13,626 (37)37–37
 ≥23,649 (35)35–35

Ambulatory visit utilization.

Virtually all OA patients (92%) visited a physician at least once annually (Figure 1). Primary care physicians were seen by 80% of this group; 25% visited an orthopedist, whereas only 6% were seen by a rheumatologist. Other physician specialists were seen by 65% of OA patients. The majority of patients (65%) saw a nonphysician at least once annually. Twenty-two percent were seen by a nurse, nurse practitioner, or physician's assistant, 11% by a physical therapist, and 58% by some other nonphysician provider. In all, 34% saw an OA specialist at least once annually. As expected, individuals who saw an OA specialist were much more likely to see a primary care physician (OR 29, 95% CI 13–63) than those who did not (data not shown).

Figure 1.

Percent and 95% confidence intervals (95% CIs) of persons with osteoarthritis in MEPS seeing various medical specialties in an ambulatory setting annually, US 2002–2005. Respondents can be seen by multiple specialties. 95% CIs are all within ±2 of the estimate. The MEPS sample reflects the noninstitutionalized civilian population. Analyses limited to individuals age ≥18 years. NP = nurse practitioner; PA = physician's assistant; MEPS = Medical Expenditure Panel Survey.

Factors associated with specialist visits.

Seeing a rheumatologist.

Results of logistic regression models predicting visits to rheumatologists are presented in Table 2. Unadjusted logistic regression models demonstrated that younger age (18–44 years versus >65 years), identification as belonging to an “other” racial group, being widowed, separated, or divorced (versus being married), living in the midwest or west (versus the northeast), and not having insurance coverage were associated with significantly lower odds of seeing a rheumatologist. Uninsured individuals with OA were 35% as likely to see a rheumatologist when compared with those who were privately insured. The univariate models also highlighted factors associated with increased likelihood of seeing a rheumatologist: being female (OR 1.85), having attended some college (OR 1.62) or graduate school (OR 1.87; compared with not completing high school), residing within an MSA (OR 1.41), the presence of 1 comorbidity (OR 1.65) and ≥2 comorbidities (OR 2.28; compared with no comorbidities), and perceived overall health being poor or fair (OR 1.58).

Table 2. Models predicting rheumatologist visits annually by adults with OA, MEPS 2002–2005 (pooled), n = 9,935*
CharacteristicUnivariate OR (95% CI)Main-effects only OR (95% CI)Interaction OR (95% CI)
  • *

    OA = osteoarthritis; MEPS = Medical Expenditure Panel Survey; OR = odds ratio; 95% CI = 95% confidence interval; MSA = metropolitan statistical area; BMI = body mass index.

  • ORs adjusted for all other characteristics in the model.

  • Not retained in the final model because P values ≥0.05.

Age, years  See interactions
 18–440.59 (0.42–0.84)0.58 (0.39–0.86) 
 45–640.96 (0.77–1.21)0.82 (0.64–1.05) 
 ≥65 (reference group)   
Women1.85 (1.46–2.35)1.96 (1.53–2.51)1.97 (1.54–2.53)
Race/ethnicity   
 Non-Hispanic white (reference group)   
 Hispanic1.25 (0.89–1.74)1.56 (1.09–2.22)1.58 (1.10–2.25)
 Non-Hispanic African American1.07 (0.79–1.46)1.13 (0.81–1.59)1.13 (0.91–1.59)
 Non-Hispanic other0.44 (0.23–0.87)0.52 (0.27–1.01)0.53 (0.27–1.02)
Marital status   
 Married (reference group)   
 Widowed, separated, divorced0.75 (0.60–0.93)0.62 (0.49–0.79)0.61 (0.49–0.78)
 Never married0.88 (0.59–1.30)0.96 (0.61–1.53)0.96 (0.60–1.52)
Education   
 <High school (reference group)   
 High school graduate1.34 (1.00–1.79)1.52 (1.13–2.04)1.49 (1.11–2.01)
 Some college1.62 (1.16–2.28)1.94 (1.38–2.71)1.92 (1.38–2.67)
 College graduate1.41 (0.95–2.11)1.83 (1.22–2.74)1.81 (1.21–2.70)
 Graduate school1.87 (1.20–2.90)2.49 (1.57–3.95)2.51 (1.59–3.95)
Geographic region   
 Northeast (reference group)   
 Midwest0.62 (0.42–0.92)0.67 (0.46–0.99)0.67 (0.45–0.98)
 South0.81 (0.60–1.10)0.86 (0.62–1.19)0.86 (0.62–1.19)
 West0.55 (0.39–0.78)0.55 (0.38–0.78)0.54 (0.38–0.77)
Reside within MSA1.41 (1.07–1.86)1.43 (1.07–1.91)1.44 (1.08–1.92)
No. comorbid conditions   
 None (reference group)   
 11.65 (1.24–2.19)1.47 (1.10–1.98)1.44 (1.07–1.93)
 ≥22.28 (1.71–3.05)1.91 (1.39–2.61)1.85 (1.35–2.55)
Obese, BMI ≥30 kg/m21.02 (0.82–1.26)
Activity limitation1.11 (0.90–1.38)
Perceived overall health fair/poor1.58 (1.27–1.95)1.69 (1.32–2.16)See interactions
Insurance type   
 Private (reference group)   
 Public only0.86 (0.66–1.12)0.80 (0.60–1.08)0.79 (0.59–1.05)
 None0.35 (0.19–0.62)0.44 (0.24–0.79)0.42 (0.23–0.76)
Interactions, age (years), perceived health   
 18–44 vs. ≥65, fair/poor  0.90 (0.45–1.35)
 18–44 vs. ≥65, good or better  0.44 (0.21–0.67)
 45–64 vs. ≥65, fair/poor  2.45 (0.38–4.52)
 45–64 vs. ≥65, good or better  0.63 (0.43–0.83)
 18–44, fair/poor vs. good or better  2.39 (1.03–3.75)
 45–64, fair/poor vs. good or better  2.24 (1.49–2.98)
 ≥65, fair/poor, vs. good or better  1.17 (0.76–1.58)

ORs from the multivariable main-effects model predicting rheumatologist visits were similar to those of the univariate models, with a few exceptions. Hispanic ethnicity became significant (OR 1.56), and ORs for all of the education categories showed a positive association between higher education levels and the likelihood of seeing a rheumatologist when compared with study subjects without a high school diploma. The effects of 1 comorbidity (OR 1.47) and 2 comorbidities (OR 1.91) when compared with individuals with no comorbidities was slightly lower in the main-effects model than the univariate model.

Two main effects predicting rheumatologist visits interacted in important ways: age and perceived health. Whereas only the 18–44 years age group was statistically significantly less likely to see a rheumatologist in the main-effects model, for individuals with good or excellent self-perceived health, the interaction model yielded significantly lower likelihoods for both the 18–44 years (OR 0.44) and the 45–64 years (OR 0.63) age groups. Similarly, while the main-effects model predicted increased odds of seeing a rheumatologist for individuals with self-perceived fair or poor health (OR 1.69), the interaction model predicted even stronger likelihoods for individuals with perceived fair or poor health in the 18–44 years (OR 2.39) and the 45–64 years (OR 2.24) age groups, but no significant difference between fair or poor versus good or better health among older individuals.

Seeing an orthopedist.

ORs estimated from logistic models predicting at least 1 orthopedist visit in a year are shown in Table 3. Significant univariate predictors of decreased orthopedist utilization included no insurance (OR 0.28) or only public insurance (OR 0.68), self-identification as Hispanic (OR 0.55), non-Hispanic African American (OR 0.60), or “other” race (OR 0.52), residence in western (OR 0.61) or southern (OR 0.72) states, and being widowed, separated, or divorced (OR 0.73). Univariate predictors associated with the increased odds of seeing an orthopedist included increasing levels of education (ORs ranged from 1.55 for high school graduate to 1.79 for graduate school when compared with individuals with less than a high school education), 1 comorbid (OR 1.31) and ≥2 comorbid (OR 1.49) conditions when compared with individuals with no such conditions, and obesity (OR 1.25). The main-effects model exhibited similar ORs when compared with the univariate models, with one exception: activity limitation was associated with a significant 30% increase in the likelihood of seeing an orthopedist, whereas in the univariate model, no significant relationship was observed.

Table 3. Models predicting orthopedist visits annually by adults with OA, MEPS 2002–2005 (pooled), n = 9,935*
CharacteristicUnivariate OR (95% CI)Main-effects only OR (95% CI)Interaction OR (95% CI)
  • *

    OA = osteoarthritis; MEPS = Medical Expenditure Panel Survey; OR = odds ratio; 95% CI = 95% confidence interval; MSA = metropolitan statistical area; BMI = body mass index.

  • ORs adjusted for all other characteristics in the model.

  • Not retained in the final model because their P values ≥0.05.

Age, years  See interactions
 18–440.86 (0.72–1.04)0.91 (0.75–1.11) 
 45–640.99 (0.86–1.13)0.93 (0.80–1.08) 
 ≥65 (reference group)   
Women1.04 (0.92–1.18)1.08 (0.94–1.24)1.09 (0.96–1.25)
Race/ethnicity   
 Non-Hispanic white (reference group)   
 Hispanic0.55 (0.44–0.73)0.74 (0.58–0.94)0.75 (0.59–0.95)
 Non-Hispanic African American0.60 (0.50–0.73)0.68 (0.55–0.82)0.68 (0.55–0.83)
 Non-Hispanic other0.52 (0.39–0.69)0.63 (0.47–0.86)0.64 (0.47–0.86)
Marital status   
 Married (reference group)   
 Widowed, separated, divorced0.73 (0.64–0.84)0.76 (0.65–0.89)0.76 (0.65–0.89)
 Never married0.83 (0.66–1.04)0.95 (0.74–1.21)0.95 (0.75–1.21)
Education   
 <High school (reference group)   
 High school graduate1.55 (1.31–1.84)1.40 (1.17–1.69)1.38 (1.15–1.66)
 Some college1.49 (1.23–1.81)1.38 (1.12–1.69)1.37 (1.11–1.68)
 College graduate1.71 (1.38–2.12)1.62 (1.28–2.05)1.61 (1.27–2.03)
 Graduate school1.79 (1.40–2.29)1.67 (1.28–2.18)1.67 (1.28–2.18)
Geographic region   
 Northeast (reference group)   
 Midwest0.84 (0.68–1.05)0.84 (0.68–1.05)0.84 (0.67–1.04)
 South0.72 (0.59–0.88)0.78 (0.63–0.95)0.77 (0.63–0.94)
 West0.61 (0.49–0.76)0.65 (0.52–0.80)0.64 (0.52–0.79)
Reside within MSA0.98 (0.83–1.15)
No. comorbid conditions   
 None (reference group)   
 11.31 (1.12–1.53)1.26 (1.08–1.48)1.25 (1.06–1.47)
 ≥21.49 (1.28–1.73)1.42 (1.21–1.67)1.38 (1.17–1.62)
Obese, BMI ≥30 kg/m21.25 (1.11–1.40)1.25 (1.11–1.41)1.26 (1.12–1.42)
Activity limitation1.14 (1.00–1.29)1.30 (1.12–1.50)See interactions
Perceived overall health fair/poor1.00 (0.88–1.13)
Insurance type   
 Private (reference group)   
 Public only0.68 (0.60–0.76)0.74 (0.64–0.86)0.71 (0.62–0.82)
 None0.28 (0.21–0.38)0.34 (0.25–0.47)0.33 (0.24–0.45)
Interactions, age (years), activity limitation   
 18–44 vs. ≥65, with activity limitation  1.64 (1.03–2.24)
 18–44 vs. ≥65, without activity limitation  0.72 (0.57–0.87)
 45–64 vs. ≥65, with activity limitation  1.34 (1.03–1.65)
 45–64 vs. ≥65, without activity limitation  0.76 (0.63–0.90)
 18–44, with activity limitation vs. without  2.13 (1.42–2.85)
 45–64, with activity limitation vs. without  1.65 (1.30–2.01)
 ≥65, with activity limitation vs. without  0.94 (0.75–1.13)

An important interaction from the orthopedist model was identified in the interactive model-building process. Younger ages were associated with increased likelihood of seeing an orthopedist among those with activity limitation (ages 18–44 years [OR 1.64] and ages 45–64 years [OR 1.34], when compared with age ≥65 years), but with a decreased likelihood (ages 18–44 years [OR 0.72] and ages 45–64 years [OR 0.76]) among those without an activity limitation. The effect of activity limitation on utilization was likewise modified by age. While we did not see a significant impact of activity limitation in the elderly adult group, it was associated with an increased likelihood of orthopedist utilization in the youngest (ages 18–44 years [OR 2.13]) and middle (ages 45–64 years [OR 1.65]) age groups.

Seeing a physical therapist.

Univariate models showed a significantly decreased likelihood of seeing a physical therapist for individuals with no insurance (OR 0.35) or public insurance only (OR 0.67) when compared with privately insured patients (Table 4). Self-identification as Hispanic (OR 0.53), non-Hispanic African American (OR 0.53), or “other” race (OR 0.54), in addition to widowed, separated, or divorced status (OR 0.63) and residing in the south (OR 0.69) were also associated with decreased odds of incurring physical therapy visits. Female sex (OR 1.19), increasing levels of education (ORs ranged from 1.75 for high school graduates to 3.09 for graduate school when compared with individuals with less than a high school education), and 1 comorbidity (OR 1.61) or ≥2 comorbidities (OR 1.99) were associated with an increased likelihood of seeing a physical therapist.

Table 4. Models predicting physical therapy visits annually by adults with OA, MEPS 2002–2005 (pooled), n = 9,935*
CharacteristicUnivariate OR (95% CI)Main-effects only OR (95% CI)Interaction OR (95% CI)
  • *

    OA = osteoarthritis; MEPS = Medical Expenditure Panel Survey; OR = odds ratio; 95% CI = 95% confidence interval; MSA = metropolitan statistical area; BMI = body mass index.

  • ORs adjusted for all other characteristics in the model.

  • Not retained in the final model because P values ≥0.05.

Age, years   
 18–441.08 (0.85–1.36)1.10 (0.86–1.40) 
 45–641.05 (0.84–1.32)0.96 (0.76–1.21) 
 ≥65 (reference group)   
Women1.19 (1.01–1.40)1.30 (1.10–1.54)1.32 (1.11–1.55)
Race   
 Non-Hispanic white (reference group)   
 Hispanic0.53 (0.40–0.71)0.73 (0.51–1.06)0.74 (0.51–1.07)
 Non-Hispanic African American0.53 (0.40–0.71)0.67 (0.51–0.89)0.67 (0.51–0.88)
 Non-Hispanic other0.54 (0.36–0.80)0.61 (0.41–0.90)0.61 (0.41–0.90)
Marital status   
 Married (reference group)   
 Widowed, separated, divorced0.63 (0.52–0.76)0.63 (0.51–0.77)0.63 (0.52–0.77)
 Never married1.03 (0.78–1.37)1.06 (0.80–1.40)1.05 (0.79–1.39)
Education   
 <High school (reference group)   
 High school graduate1.75 (1.40–2.17)1.56 (1.23–1.98)1.54 (1.21–1.96)
 Some college1.97 (1.55–2.52)1.77 (1.34–2.34)1.76 (1.34–2.33)
 College graduate2.48 (1.86–3.31)2.31 (1.70–3.14)2.28 (1.68–3.10)
 Graduate school3.09 (2.36–4.05)2.82 (2.06–3.86)2.84 (2.07–3.89)
Geographic region   
 Northeast (reference group)   
 Midwest1.00 (0.80–1.25)1.01 (0.81–1.26)1.00 (0.80–1.25)
 South0.69 (0.56–0.84)0.73 (0.60–0.89)0.72 (0.60–0.88)
 West0.84 (0.64–1.10)0.84 (0.63–1.12)0.83 (0.63–1.11)
Reside within MSA1.05 (0.90–1.24)
No. comorbid conditions   
 None (reference group)   
 11.61 (1.31–1.98)1.60 (1.29–1.97)1.58 (1.28–1.95)
 ≥21.99 (1.62–2.43)2.06 (1.67–2.55)2.02 (1.63–2.49)
Obese, BMI ≥30 kg/m21.03 (0.89–1.19)
Activity limitation1.08 (0.93–1.27)1.28 (1.07–1.52)See interactions
Perceived overall health fair/poor1.06 (0.90–1.24) 
Insurance type   
 Private (reference group)   
 Public only0.67 (0.56–0.81)0.81 (0.66–1.00)0.78 (0.64–0.96)
 None0.35 (0.21–0.61)0.47 (0.27–0.81)0.45 (0.26–0.79)
Interactions, age (years), activity limitation   
 18–44 vs. >65, with activity limitation  1.46 (0.77–2.16)
 18–44 vs. >65, without activity limitation  0.95 (0.70–1.21)
 45–64 vs. ≥65, with activity limitation  1.43 (0.96–1.91)
 45–64 vs. ≥65, without activity limitation  0.78 (0.56–0.99)
 18–44, with activity limitation vs. without  1.44 (0.82–2.06)
 45–64, with activity limitation vs. without  1.73 (1.26–2.20)
 ≥65, with activity limitation vs. without  0.94 (0.69–1.19)

Results from the main-effects model were similar to those of the univariate models with 2 exceptions: Hispanic ethnicity was no longer a significant predictor, and activity limitation became a significant predictor of increased likelihood of physical therapy utilization (OR 1.28). The multivariable model uncovered an important interaction between activity limitation and age, i.e., activity limitation was associated with increased odds of seeing a physical therapist, but only among individuals ages 45–64 years (OR 1.73). Although we did not find age to be significantly associated with the odds of physical therapy utilization in the main-effects model, the interaction model showed the middle age group was less likely than the older age group to incur physical therapy utilization among those without an activity limitation (OR 0.78).

DISCUSSION

We estimate that 22.5 million adults between 2002 and 2005 had ≥1 interaction with health care providers that documented OA diagnosis or reported symptoms evident of OA. This is similar to the most recent estimate of 26.9 million, based on clinically defined OA from the National Health and Nutrition Survey I (NHANES I) applied to the 2005 population (2). Because the 26.9 million figure is considered by its authors to be conservative (2), and current research concerning the reliability of patient-reported OA in the MEPS indicates high specificity and sensitivity for the ICD-9-CM codes included in our definition, we feel this estimate captures most of the individuals with physician-diagnosed, symptomatic OA. The 4.4 million (16%) difference between the NHANES I estimate and the current study may be attributed to the different years and methodologies used to obtain each estimate. The NHANES I prevalence rate is derived from samples of physical examinations and symptoms collected between 1971 and 1975, and then applied to the 2005 US Census population estimates. Whether the 1971–1975 NHANES I estimates reflect the 2005 US population prevalence is not known. The high rates of comorbid conditions we observed in our OA population mirror those of other US-based OA studies (15–17) as do the proportion of women and the mean age (18).

Although, to our knowledge, no other study has examined ambulatory utilization patterns on a national basis, comparisons of annual specialist utilization found here (primary care at 80%, orthopedics at 25%, rheumatology at 6%, and physical therapy at 11%) with that of a regional study is informative. Lanes and colleagues analyzed health utilization data for OA subjects from a central Massachusetts managed-care organization incurred between July 1993 and June 1994. Sixty-seven percent of OA subjects had ≥1 OA-related office visit during that 1-year period; orthopedists, rheumatologists, and physical therapists were seen by 23%, 16%, and 13% of patients for OA-related treatment, respectively (19). An analysis of survey data collected between 1996 and 1998 for OA patients derived from an outpatient rheumatology clinic in Wichita, Kansas, showed a similar low rate for rheumatologist utilization: during a 6-month period, only 6% of OA patients consulted a rheumatologist (18).

Our analysis of factors associated with seeing the specialists most trained to provide OA care shows similarities among the specialists. Having no health insurance, or only public insurance, or being widowed, separated, or divorced is associated with a much lower likelihood of seeing any of the arthritis specialists in the adjusted models. Conversely, higher levels of education, the presence of ≥1 comorbid conditions, and female sex are associated with increased odds of specialist utilization. Older ages are associated with higher odds of seeing either physician specialist, but only among those subjects who perceived their health as being good to excellent. Interestingly, obesity is associated with increased odds of orthopedics utilization, but not the other 2 specialties; this may be worthy of further study. Many of the factors predicting increased utilization of ambulatory OA care found here have been reported in other studies (20–23).

Results of our study should be interpreted within data limitations. First of all, utilization of specialists is derived through patient self-report. It is possible that some individuals visit “arthritis specialists” without being aware of the providers' medical specialty (24). Second, visits to these specialists may not have been related to the subjects' OA. Finally, our definition of OA relied on patient self-report, which is less accurate than a provider-identified definition, and provides no information as to the severity of the respondent's OA.

The present study highlights important associations of low OA specialist utilization with a lack of health insurance coverage, being widowed, separated, or divorced, and lower levels of education, but it does not inform us as to why utilization rates to arthritis specialists are so low overall in the US, nor does it explain why some groups are more or less likely to see a specialist than others. A 1999 survey of adults from all 4 US regions with self-reported doctor-diagnosed rheumatoid arthritis (RA) or OA enrolled in either a private or public health care plan assessed the unmet need for specialist care in the previous 6 months, and rehabilitation (physical therapy, occupational therapy, or speech therapy) during the previous 3 months. In the latter study, the most common reason given for this lack of access was that the individual's insurance plan would not cover the service. Other reasons included the service being too expensive or that a referral could not be obtained (25). This supports the hypothesis that insurance coverage, or the lack thereof, is one of the most important barriers to specialist care. The lack of adequate insurance coverage, as well as socioeconomic barriers such as poverty and lack of fluency in English, are also known to be associated with decreased access to more costly care across most medical specialty areas, including diabetes mellitus (26), cancer (27), orthopedics (28), and vascular disease (29).

There is evidence that rheumatologists achieve superior outcomes treating RA patients relative to other providers (30); it is unknown whether this also applies to OA patients. What is known, however, is that the supply of rheumatologists in the US is increasing at a slower rate than the demand for their services (31). It seems clear that nonrheumatologists will need to become knowledgeable about risks associated with common OA medications such as nonsteroidal antiinflammatory drugs as well as evidence-based interventions for OA, including weight reduction through diet modification and exercise. Because weight loss for overweight and obese populations can prevent or decrease progression of other prevalent chronic conditions, such as heart disease and diabetes mellitus, in addition to preventing OA and reducing OA symptoms, an opportunity exists for primary care providers and health care plans to encourage healthy behavior through implementation of evidence-based weight loss programs. In fact, there is evidence to support the hypothesis that implementation of such programs can actually decrease health care expenditures above and beyond the initial investment (32).

We are aware that OA is much too prevalent to be cared for exclusively or even primarily by specialists. The general practitioner has a central role in recognizing OA and beginning treatment. A small percentage of general practitioners may develop expertise similar to that of a specialist, but currently, most do not. Given OA patients' low utilization rates of specialists trained to provide OA care in general and specific second-line treatments (injections, bracing, and total joint arthroplasty) in particular, in addition to the variation in the provision of guideline-concordant care both within and among specialties (33, 34), studies are needed to determine which types of providers are providing effective care and communicating OA-related treatment information. If effective care is not the norm, targeted educational initiatives should be implemented. For example, it is known that primary care providers often avoid administering injections for arthritis (35, 36), but training can increase provider self-confidence and injection skill (37). In the interim, the current study describes relatively low rates of usage of OA specialists and disparities by insurance coverage and educational status associated with such usage.

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 submitted for publication. Ms Cisternas 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. Cisternas, Yelin, Katz, Solomon, Losina.

Acquisition of data. Cisternas, Yelin.

Analysis and interpretation of data. Cisternas, Yelin, Katz, Solomon, Wright, Losina.

APPENDIX A

OSTEOARTHRITIS POPULATION COMORBIDITY DEFINITIONS

Cancer: ICD-9-CM 140–208 or evidence of chemotherapy; Ischemic heart disease or chronic heart failure: ICD-9-CM 410–414, 428 or a prescription for anti–anginal agents; nonOA musculoskeletal disease: ICD-9-CM 274, 710–714, 716–739; ICD-9-CM 490–496 or prescriptions for montelukast sodium; diabetes mellitus: ICD-9-CM 250 or prescriptions for antidiabetic agents; ICD-9-CM 296 or 311, prescriptions for various antidepressants that are uniquely indicated for depression, or an indication of being downhearted or depressed most or all of the time in the past 4 weeks.

Ancillary