Ambulatory Visit Utilization in a National, Population-based Sample of Adults with Osteoarthritis

Abstract

Objectives

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

Methods

We use data from the Medical Expenditures Panel Survey (MEPS), a stratified random sample of the civilian non-institutionalized population. We classify adults as having symptomatic OA if: (1) their medical conditions include at least one occurrence of ICD-9-CM 715, 716, or 719, and (2) they report joint pain, swelling, or stiffness during the previous twelve months. For the purpose of our analysis we define rheumatologists, orthopedists and physical therapists as OA specialists. We first estimate the proportion of OA individuals seen by OA specialists and other health care providers in a one-year period. We then use logistic regression to estimate the impact of demographic and clinical factors on the likelihood of seeing an OA specialist.

Results

9,933 persons meet the definition of OA, representing 22.5 million adults in the U.S. Virtually all (92%) see physicians during the year; 34% see at least one OA specialist; 25% see an orthopedist, 11% a physical therapist, and 6% a rheumatologist. The following are significant positive predictors for seeing an OA specialist: higher educational attainment, having more comorbidities, and residing in the Northeastern U.S. Significant negative predictors for seeing OA specialists are being unmarried but previously married, and having no health insurance.

Conclusions

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.

Osteoarthritis (OA) is the most prevalent form of arthritis, second only to ischemic heart disease as a cause of U.S. work disability in men over 50^{1, 2}. The most recently published prevalence estimate of OA was 26.9 million among U.S. adults age 25 and over in 2005².

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 1997⁴. The most recent U.S. study of direct costs attributable to OA was restricted to adults 18 – 64 and presented in 2004 U.S. 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^{7, 6}.

While many of the risk factors for the development and worsening of symptoms in OA cannot be prevented (female gender, 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 programs⁸. Understanding patterns of care for the population affected by OA is crucial for targeting interventions, but no national, population-based study of ambulatory care patterns in OA patients has been published.

The objectives of this study are to use a population-based US data source to ascertain: (1) the percentage of adults with OA seeing various medical providers, including specialists who provide arthritis-related care (rheumatologists, orthopedists, and physical therapists), and (2) which factors affect the likelihood of individuals with OA seeing specialists focused on arthritis-related care.

Materials and Methods

Data Source

We use the Medical Expenditure Panel Survey (MEPS) household component (HC), a nationally representative survey of the United States civilian, non-institutionalized population, to estimate the proportion of U.S. adults with OA seeing various providers in ambulatory settings. MEPS is a joint endeavor of the Agency for Healthcare Research and Quality (AHRQ) and the National Center for Health Statistics.

The MEPS-HC collects data on healthcare use, demographic characteristics, and health status from five interviews over a two-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 as “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, Clinical Modification, Ninth Revision (ICD-9-CM) system at the three-digit level.¹¹ For example, patient-reported “generalized osteoarthritis of the hand”, corresponding to a five-digit ICD-9-CM of 715.04, would be coded in MEPS as 715.

The MEPS Medical Provider Component (MPC) collects date, diagnostic, procedure, and expenditure data from medical providers for visits made by a sub-sample of the MEPS-HC participants. The accuracy of some self-reported condition data from MEPS-HC respondents has been assessed by comparison to 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 classify adults as having OA if the following two requirements are met: (1) their medical conditions listed in the MEPS-HC include at least one 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 (2) and they report pain, swelling, or stiffness around a joint during the previous twelve 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 (kappa=0.15, 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 results in substantial increases in OA sensitivity (from 17% to 81%) with only minor decreases in specificity (99% to 89%) when using ICD-9-CM 715 recorded in the MPC as the gold standard (Louise Murphy, Centers for Disease Control and Prevention, personal communication, 11/12/2008). By adding ICD-9-CM codes 716 and 719 to our definition, we estimate that our study population includes 81% of all OA adults in the MEPS sample with a doctor diagnosis of OA in a year, but that 19% of individuals included on our study population did not have a doctor-provided diagnosis.

Data elements

Demographic factors

Demographic variables include age in years, female gender, Hispanic ethnicity, white race, marital status (married, widowed/separated/divorced, never married), and highest level of education attained (less than high school, high school grad, some college, college grad, some graduate school). Other demographic variables include geographic region of residence (Northeast, Midwest, South, West), and living within a metropolitan statistical area (MSA).

Clinical characteristics (specific ICD-9-CM codes provided in the electronic appendix)

Comorbid condition indicator variables are created by flagging ICD-9-CM codes for: cancer, ischemic heart disease/chronic heart failure, COPD/asthma, depression, diabetes, and non-OA musculoskeletal disease. Obesity is defined as a body mass index (BMI) of 30 kg/m² or higher. These binary comorbid condition variables are summed to create a variable recording whether the study subject had zero, one, or two or more comorbid conditions. Binary variables for patient-perceived health status (poor/fair vs. excellent/very good/good)¹³ and activity limitation (any limitation at work, home, or school vs. none) are also created.

Measures of healthcare utilization

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

Statistical analysis

Regression models

The dependent variables for these regressions are the odds of subjects seeing each of the OA specialists. Logistic regression is used to ascertain which of the potential demographic and clinical factors are associated with seeing these providers.

Three types of regression models are created for each outcome. Univariate analyses are performed with each of the potential explanatory variables as an independent variable. We then create 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 are included in all models regardless of statistical significance, whereas non-demographic covariates require a P<.05 significance level to be retained. The exclusion of non-demographic covariates is done one variable at a time, starting with the least significant variable, until all non-demographic variables that remain in the model are significant at P<.05. Our final step is to investigate whether important interactions exist between variables included in the main effects models and any of the following: obesity, activity limitation, age, gender, 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 are statistically significant at P<0.05, do not include sparse cells (sample N≤10), and add important information to the interpretation of the model when compared to the main effects model are retained to yield the interaction model for each outcome. While odds ratios (ORs) and 95% confidence intervals (CIs) are 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 are applied in all analyses. We use SAS version 9.2 Survey procedures (Surveyfreq, Surveymeans, and Surveylogistic) to adjust standard error estimates for MEPS’ clustered sampling design¹⁴. Annual estimates are obtained by dividing sampling weights in the pooled file by four, the number years of MEPS data used.

Missing data

For the 416 (4%) of adult OA observations with missing data, we impute data using multiple imputation procedures (SAS version 9.1.3 PROC MI and PROC MIANALYZE). SAS multiple imputation procedures are implemented on the 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 are 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 are arranged in a general pattern, we use the Markov Chain Monte Carlo imputation method.

Results

Population Characteristics

The population for this study includes the 9,933 adults in MEPS meeting the study definition of OA; these 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 exhibited in Table 1. Individuals in the cohort are 60 years of age on average, and 63% are female. Most (78%) identify themselves as nonHispanic white; 7% are Hispanic, 11% are Black, while the remaining 4% identified with some other racial group. The majority (77%) resides in a MSA, are married (56%), and have at least some private insurance coverage (65%). One quarter have not obtained a high-school diploma; 35% have graduated from high school, while the remaining 40% have at least some college education. One third perceive their health as fair or poor, and 30% report activity limitation. Fifty-five percent of respondents indicate that pain limits their normal work at least moderately. Obesity is present for 39% of OA respondents, and 72% of study subjects are classified as having at least one of the six comorbidities.

Table 1.

General population characteristics for OA Cohorts, MEPS 2002–2005 (pooled) Age 18+ with Joint Pain N=9,933

Characteristics	Sample N	Percent Estimate	(95% CI)
Age in categories
18 – 44	1,539	16	(16,16)
45 – 65	4,340	43	(43,44)
65+	4,054	41	(41,41)
Mean age in years	9,933	60	(60,61)
Percent Female	6,572	63	(63,63)
Race/ethnicity
White (nonHispanic)	6,571	78	(77,78)
Hispanic	1,344	7	(7,7)
Black (nonHispanic)	1,583	11	(11,11)
Other (nonHispanic)	435	4	(4,4)
Region
Northeast	1,435	17	(16,17)
Midwest	2,152	24	(24,24)
South	4,230	39	(39,39)
West	2,115	20	(20,21)
Percent Living within MSA	7,310	77	(77,77)
Marital Status
Married	5,261	56	(56,56)
Widowed, Separated, Divorced	3,782	35	(35,35)
Never married	890	9	(9,9)
Insurance Type
private (ref.)	5,556	65	(65,66)
public	3,620	28	(28,29)
none	757	6	(6,6)
Education
Less than high school	3,247	25	(25,25)
High school grad	3,282	35	(35,35)
Some college	1,863	21	(21,21)
College grad	908	11	(11,11)
Some graduate school	632	8	(8,8)
Percent with perceived overall health fair/poor	3,864	33	(32,33)
Activity Limitation	3,406	30	(30,30)
Pain limits normal work
not at all	1,542	18	(17,18)
a little bit	2,523	27	(27,28)
moderately	2,161	22	(22,22)
quite a bit	2,576	23	(23,24)
extremely	1,130	10	(9,10)
Obese (BMI 30 kg/m2 and over)	4,109	39	(38,39)
Specific high-cost comorbidities*
cancer	680	7	(7,7)
ischemic heart disease/chronic heart failure	1,031	10	(10,10)
COPD/asthma	1,599	15	(15,15)
depression	2,916	27	(27,28)
diabetes	1,928	17	(17,17)
non-OA musculoskeletal disease	4,501	46	(45,46)
none	2,658	28	(28,28)
one	3,626	37	(37,37)
two or more	3,649	35	(35,35)

^*Individuals could have zero, one, or several such comorbidities

Ambulatory Visit Utilization

Virtually all OA patients (92%) visit a physician at least once annually (Figure 1). Primary care physicians are seen by 80% of this group; one quarter visit an orthopedist, whereas only 6% are seen by a rheumatologist. Other physician specialists are seen by 65% of OA patients.

Figure 1.

Percent and 95% CIs of Persons with OA in MEPS Seeing Various Medical Specialties in an Ambulatory Setting over a One-Year Period, U.S., 2002–2005

Respondents can be seen by multiple specialties. NP=Nurse Practitioners, PA=Physician’s Assistants.

95% CIs are all within ±2% of the estimate.

The MEPS sample reflects the non-institutionalized civilian population; analyses limited to individuals age 18 and over.

The majority of patients (65%) see a nonphysician at least once annually. Twenty-two percent are 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% see an OA specialist at least once annually. As expected, individuals who see an OA specialist are much more likely to see a primary care physician (OR 29; 95% CI 13 – 63) than those who do not (data not shown).

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 demonstrate that younger age (18 – 44 vs. 65 and over), identification as belonging to an “other” racial group, being widowed, separated or divorced (vs. married), living in the Midwest or West (vs. Northeast), and not having insurance coverage are associated with significantly lower odds of seeing a rheumatologist. Uninsured individuals with OA are about one third as likely to see a rheumatologist when compared to those who are privately insured. The univariate models also highlight 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 to not completing high school), residing within an MSA (1.41), the presence of one (OR 1.65) and two or more (OR 2.28) comorbidities (compared to no comorbidities), and perceived overall health being poor or fair (OR 1.58).

Table 2.

Models Predicting Rheumatologist Visits in a Single Year by OA Adults, MEPS 2002–2005 (pooled) N=9,935

	Univariate	Main-effects Only	Interaction
Characteristic	OR (95% CI)	Adjusted* OR 95%CI)	Adjusted* OR (95%CI)
Age
18 – 44	0.59 (0.42,0.84)	0.58 (0.39,0.86)	see interactions
45 – 64	0.96 (0.77,1.21)	0.82 (0.64,1.05)
65 and over (ref.)
Female	1.85 (1.46,2.35)	1.96 (1.53,2.51)	1.97 (1.54,2.53)
Race/ethnicity
White (nonHispanic) (ref.)
Hispanic	1.25 (0.89,1.74)	1.56 (1.09,2.22)	1.58 (1.10,2.25)
Black (nonHispanic)	1.07 (0.79,1.46)	1.13 (0.81,1.59)	1.13 (0.91,1.59)
Other (nonHispanic)	0.44 (0.23,0.87)	0.52 (0.27,1.01)	0.53 (0.27,1.02)
Marital status
Married (ref.)
Widowed, Separated, Divorced	0.75 (0.60,0.93)	0.62 (0.49,0.79)	0.61 (0.49,0.78)
Never married	0.88 (0.59,1.30)	0.96 (0.61,1.53)	0.96 (0.60,1.52)
Education
Less than high school (ref.)
High school grad	1.34 (1.00,1.79)	1.52 (1.13,2.04)	1.49 (1.11,2.01)
Some college	1.62 (1.16,2.28)	1.94 (1.38,2.71)	1.92 (1.38,2.67)
College grad	1.41 (0.95,2.11)	1.83 (1.22,2.74)	1.81 (1.21,2.70)
Graduate school	1.87 (1.20,2.90)	2.49 (1.57,3.95)	2.51 (1.59,3.95)
Geographic region
Northeast (ref.)
Midwest	0.62 (0.42,0.92)	0.67 (0.46,0.99)	0.67 (0.45,0.98)
South	0.81 (0.60,1.10)	0.86 (0.62,1.19)	0.86 (0.62,1.19)
West	0.55 (0.39,0.78)	0.55 (0.38,0.78)	0.54 (0.38,0.77)
Reside within MSA	1.41 (1.07,1.86)	1.43 (1.07,1.91)	1.44 (1.08,1.92)
Number of Other Comorbid Conditions
None (ref.)
One	1.65 (1.24,2.19)	1.47 (1.10,1.98)	1.44 (1.07,1.93)
Two or more	2.28 (1.71,3.05)	1.91 (1.39,2.61)	1.85 (1.35,2.55)
Obese (BMI ≥ 30 kg/m2 )	1.02 (0.82,1.26)	—†	—†
Activity limitation	1.11 (0.90,1.38)	—†	—†
Perceived overall health fair/poor	1.58 (1.27,1.95)	1.69 (1.32,2.16)	see interactions
Insurance Type
Private (ref.)
Public only	0.86 (0.66,1.12)	0.80 (0.60,1.08)	0.79 (0.59,1.05)
None	0.35 (0.19,0.62)	0.44 (0.24,0.79)	0.42 (0.23,0.76)
Interactions
age 18– 44 vs. 65+ among fair/poor perceived health			0.90 (0.45,1.35)
age 18– 44 vs. 65+ among perceived health good or better			0.44 (0.21,0.67)
age 45– 64 vs. 65+ among fair/poor perceived health			2.45 (0.38,4.52)
age 45– 64 vs. 65+ among perceived health good or better			0.63 (0.43,0.83)
fair/poor perceived health among age 18 – 44			2.39 (1.03,3.75)
fair/poor perceived health among age 45 – 64			2.24 (1.49,2.98)
fair/poor perceived health among age 65+			1.17 (0.76,1.58)

^*Odds ratios are adjusted for all other characteristics in the model.

^†These candidate variables were not retained in the final model because their P-values did not approach 0.05.

Odds ratios from the multivariable main-effects model predicting rheumatologist visits are similar to those of the univariate models, with a few exceptions. Hispanic ethnicity becomes significant (OR 1.56), and ORs for all of the education categories show a positive association between higher education levels and the likelihood of seeing a rheumatologist when compared to study subjects without a high school diploma. The effects of one (OR 1.47) and two (OR 1.91) comorbidities when compared to individuals without any comorbidities is slightly lower in the main-effects model than the univariate model.

Two main effects predicting rheumatologist visits interact in important ways: age and perceived health. Whereas only the 18–44 age group is 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 yields significantly lower likelihoods for both the 18–44 (OR 0.44) and 45–64 (OR 0.63) groups. Similarly, while the main effects model predicts increased odds of seeing a rheumatologist for individuals with self-perceived fair or poor health (OR 1.69), the interaction model predicts even stronger likelihoods for individuals with perceived fair or poor health in the 18 – 44 (OR 2.39) and 45 – 64 (OR 2.24) age groups, but no significant difference between fair or poor vs. good or better health among older individuals.

Seeing an orthopedist

Odds ratios estimated from logistic models predicting at least one orthopedist visit in a year are displayed in Table 3. Significant univariate predictors of decreased orthopedist utilization include no (OR 0.28) or only public (OR 0.68) insurance, self-identification as Hispanic (OR 0.55), nonHispanic Black (OR 0.60), or “other” race (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 include increasing levels of education (ORs ranging from 1.55 for high-school grad to 1.79 for graduate school when compared to individuals with less than a high-school education) one (OR 1.31) and two or more (OR 1.49) comorbid conditions when compared to individuals with no such conditions, and obesity (OR 1.25). The main-effects model exhibits similar ORs when compared to the univariate models, with one exception: activity limitation is associated with a significant 30% increase in the likelihood of seeing an orthopedist, whereas in the univariate model, no significant relationship is observed.

Table 3.

Models Predicting Orthopedics Visits in a Single Year by OA Adults, MEPS 2002–2005 (pooled) N=9,935

	Univariate	Main-effects Only	Interaction
Characteristic	OR (95% CI)	Adjusted* OR	Adjusted* OR (95%CI)
Age			see interactions
18 – 44	0.86 (0.72,1.04)	0.91 (0.75,1.11)
45 – 64	0.99 (0.86,1.13)	0.93 (0.80,1.08)
65 and over (ref.)
Female	1.04 (0.92,1.18)	1.08 (0.94,1.24)	1.09 (0.96,1.25)
Race/ethnicity
White (nonHispanic) (ref.)
Hispanic	0.55 (0.44,0.73)	0.74 (0.58,0.94)	0.75 (0.59,0.95)
Black (nonHispanic)	0.60 (0.50,0.73)	0.68 (0.55,0.82)	0.68 (0.55,0.83)
Other (nonHispanic)	0.52 (0.39,0.69)	0.63 (0.47,0.86)	0.64 (0.47,0.86)
Marital status
Married (ref.)
Widowed, Separated, Divorced	0.73 (0.64,0.84)	0.76 (0.65,0.89)	0.76 (0.65,0.89)
Never married	0.83 (0.66,1.04)	0.95 (0.74,1.21)	0.95 (0.75,1.21)
Education
Less than high school (ref.)
High school grad	1.55 (1.31,1.84)	1.40 (1.17,1.69)	1.38 (1.15,1.66)
Some college	1.49 (1.23,1.81)	1.38 (1.12,1.69)	1.37 (1.11,1.68)
College grad	1.71 (1.38,2.12)	1.62 (1.28,2.05)	1.61 (1.27,2.03)
Graduate school	1.79 (1.40,2.29)	1.67 (1.28,2.18)	1.67 (1.28,2.18)
Geographic region
Northeast (ref.)
Midwest	0.84 (0.68,1.05)	0.84 (0.68,1.05)	0.84 (0.67,1.04)
South	0.72 (0.59,0.88)	0.78 (0.63,0.95)	0.77 (0.63,0.94)
West	0.61 (0.49,0.76)	0.65 (0.52,0.80)	0.64 (0.52,0.79)
Reside within MSA	0.98 (0.83,1.15)	—†	—†
Number of Other Comorbid Conditions
None (ref.)
One	1.31 (1.12,1.53)	1.26 (1.08,1.48)	1.25 (1.06,1.47)
Two or more	1.49 (1.28,1.73)	1.42 (1.21,1.67)	1.38 (1.17,1.62)
Obese (BMI ≥ 30 kg/m2 )	1.25 (1.11,1.40)	1.25 (1.11,1.41)	1.26 (1.12,1.42)
Activity limitation	1.14 (1.00,1.29)	1.30 (1.12,1.50)	see interactions
Perceived overall health fair/poor	1.00 (0.88,1.13)	—†	—†
Insurance Type
Private (ref.)
Public only	0.68 (0.60,0.76)	0.74 (0.64,0.86)	0.71 (0.62,0.82)
None	0.28 (0.21,0.38)	0.34 (0.25,0.47)	0.33 (0.24,0.45)
Interactions
age 18– 44 vs. 65+ among those with activity limitation			1.64 (1.03,2.24)
age 18– 44 vs. 65+ among those without activity limitation			0.72 (0.57,0.87)
age 45– 64 vs. 65+ among those with activity limitation			1.34 (1.03,1.65)
age 45– 64 vs. 65+ among those without activity limitation			0.76 (0.63,0.90)
activity limitation among age 18 – 44			2.13 (1.42,2.85)
activity limitation among age 45 – 64			1.65 (1.30,2.01)
activity limitation among age 65+			0.94 (0.75,1.13)

^*Odds ratios are adjusted for all other characteristics in the model

^†These candidate variables were not retained in the final model because their P-values did not approach 0.05.

An important interaction from the orthopedist model is identified in the interactive model-building process. Younger ages are associated with increased likelihood of seeing an orthopedist among those with activity limitation (age 18 – 44 OR 1.64 and age 45 – 64 OR 1.34 when compared to age 65 and older), but with a decreased likelihood (18 – 44 OR 0.72 and age 45 – 64 OR 0.76) among those without an activity limitation. The effect of activity limitation on utilization is likewise modified by age: while we do not see significant impact of activity limitation in the elderly group, it is associated with increased likelihood of orthopedist utilization in the youngest (age 18 – 44 OR 2.12) and middle (age 45 – 64 OR 1.65) age groups.

Seeing a physical therapist

Univariate models show 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 to privately insured subjects (Table 4). Self-identification as Hispanic (OR 0.53), nonHispanic Black (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) are also associated with decreased odds of incurring physical therapy visits. Female gender (OR 1.19), increasing levels of education (ORs ranging from 1.75 for high school graduates to 3.09 for graduate school when compared to individuals with less than a high school education) and one (OR 1.61) or two or more (OR 1.99) comorbidities are associated with an increased likelihood of seeing a physical therapist. Results from the main-effects model are similar to those of the univariate models, with two exceptions: Hispanic ethnicity is no longer a significant predictor, and activity limitation becomes a significant predictor of increased likelihood of physical therapy utilization (OR 1.28). The multivariable model uncovers an important interaction between activity limitation and age: activity limitation is associated with increased odds of seeing a physical therapist, but only among individuals 45 – 64 (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 shows the middle age group less likely than the older age group to incur physical therapy utilization among those without an activity limitation (OR 0.78).

Table 4.

Models Predicting Physical Therapy Visits in a Single Year by OA Adults, MEPS 2002–2005 (pooled) N=9,935

	Univariate	Main-effects Only	Interaction
Characteristic	OR (95% CI)	Adjusted* OR (95%CI)	Adjusted* OR (95%CI)
Age			see interactions
18 – 44	1.08 (0.85,1.36)	1.10 (0.86,1.40)
45 – 64	1.05 (0.84,1.32)	0.96 (0.76,1.21)
65 and over (ref.)
Female	1.19 (1.01,1.40)	1.30 (1.10,1.54)	1.32 (1.11,1.55)
Race/ethnicity
White (nonHispanic) (ref.)
Hispanic	0.53 (0.40,0.71)	0.73 (0.51,1.06)	0.74 (0.51,1.07)
Black (nonHispanic)	0.53 (0.40,0.71)	0.67 (0.51,0.89)	0.67 (0.51,0.88)
Other (nonHispanic)	0.54 (0.36,0.80)	0.61 (0.41,0.90)	0.61 (0.41,0.90)
Marital status
Married (ref.)
Widowed, Separated, Divorced	0.63 (0.52,0.76)	0.63 (0.51,0.77)	0.63 (0.52,0.77)
Never married	1.03 (0.78,1.37)	1.06 (0.80,1.40)	1.05 (0.79,1.39)
Education
Less than high school (ref.)
High school grad	1.75 (1.40,2.17)	1.56 (1.23,1.98)	1.54 (1.21,1.96)
Some college	1.97 (1.55,2.52)	1.77 (1.34,2.34)	1.76 (1.34,2.33)
College grad	2.48 (1.86,3.31)	2.31 (1.70,3.14)	2.28 (1.68,3.10)
Graduate school	3.09 (2.36,4.05)	2.82 (2.06,3.86)	2.84 (2.07,3.89)
Geographic region
Northeast (ref.)
Midwest	1.00 (0.80,1.25)	1.01 (0.81,1.26)	1.00 (0.80,1.25)
South	0.69 (0.56,0.84)	0.73 (0.60,0.89)	0.72 (0.60,0.88)
West	0.84 (0.64,1.10)	0.84 (0.63,1.12)	0.83 (0.63,1.11)
Reside within MSA	1.05 (0.90,1.24)	—†	—†
Number of Other Comorbid Conditions
None (ref.)
One	1.61 (1.31,1.98)	1.60 (1.29,1.97)	1.58 (1.28,1.95)
Two or more	1.99 (1.62,2.43)	2.06 (1.67,2.55)	2.02 (1.63,2.49)
Obese (BMI ≥ 30 kg/m2 )	1.03 (0.89,1.19)	—†	—†
Activity limitation	1.08 (0.93,1.27)	1.28 (1.07,1.52)	see interactions
Perceived overall health fair/poor	1.06 (0.90,1.24)	—†
Insurance Type
Private (ref.)
Public only	0.67 (0.56,0.81)	0.81 (0.66,1.00)	0.78 (0.64,0.96)
None	0.35 (0.21,0.61)	0.47 (0.27,0.81)	0.45 (0.26,0.79)
Interactions
age 18– 44 vs. 65+ among those with activity limitation			1.46 (0.77,2.16)
age 18– 44 vs. 65+ among those without activity limitation			0.95 (0.70,1.21)
age 45– 64 vs. 65+ among those with activity limitation			1.43 (0.96,1.91)
age 45– 64 vs. 65+ among those without activity limitation			0.78 (0.56,0.99)
activity limitation among age 18 – 44			1.44 (0.82,2.06)
activity limitation among age 45 – 64			1.73 (1.26,2.20)
activity limitation among age 65+			0.94 (0.69,1.19)

^*Odds ratios are adjusted for all other characteristics in the model

^†These candidate variables were not retained in the final model because their P-values did not approach 0.05

Discussion

We estimate that 22.5 million adults between 2002 and 2005 had at least one interaction with health care 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 population². Because the 26.9 million figure is considered by its authors to be conservative², and current research concerning the reliability of patient-reported OA in MEPS indicate 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 – 1975, and then applied to the 2005 U.S. Census population estimates. Whether the 1971–1975 NHANES estimates reflect the 2005 U.S. population prevalence is not known. The high rates of comorbid conditions we observed in our OA population mirror those of other U.S.-based OA studies^{15, 16, 17}, as do the proportion of females and mean age¹⁸.

Although 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 at least one OA-related office visit during that one-year period; orthopedists, rheumatologists, and physical therapists were seen by 23%, 16%, and 13% of patients for OA-related treatment, respectively¹⁹. 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 six-month period, only 6% of OA patients consulted a rheumatologist¹⁸.

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 one or more comorbid conditions, and female gender 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 good to excellent. Interestingly, obesity is associated with increased odds of orthopedics utilization, but not the other two specialties; this may be worthy of further study. Many of the factors predicting increased utilization of ambulatory OA care found here have been found in other studies^{20, 21, 22, 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 Secondly, visits to these specialists may not have been related to 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 why utilization rates to arthritis specialists are so low in the U.S. overall, nor does it explain why some groups are more or likely to see a specialist than others. A 1999 survey of adults from all four U.S. 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 six months, and rehabilitation (physical therapy, occupational therapy, or speech therapy) during the previous three 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, others reasons included the service being too expensive or that a referral could not be obtained.²⁵ This supports the hypothesis that insurance coverage, or 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 diabetes²⁶, cancer²⁷, orthopedics²⁸ and vascular disease²⁹.

There is evidence that rheumatologists achieve superior outcomes treating RA patients relative to other providers³⁰; it is unknown whether this also applies to OA patients. What is known, however, is that the supply of rheumatologists in the U.S. is increasing at a slower rate than the demand for their services³¹. It seems clear that non-rheumatologists will need to become knowledgeable about risks associated with common OA-medications such as NSAIDS 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 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 present paper describes relatively low rates of usage of OA specialists and disparities by insurance coverage and educational status associated with such usage.