Study sample

We analysed 2009, 2010 and 2011 combined National Health Interview Survey (NHIS) data. The NHIS is an in-person, population-based survey that is conducted annually by the US Centers for Disease Control and Prevention’s National Center for Health Statistics and that collects information to assess “health status, health care access, and progress toward achieving national health objectives.” . It is designed to provide estimates that are representative of the US non-institutionalised civilian population. Our study sample comprised adults aged ≥18 years who participated in the NHIS adult component ( n = 87 902). The final response rates among those who were eligible for the NHIS sample adult module, after accounting for household and family non-response, were 65.4% in 2009 ( n = 27 731), 60.8% in 2010 ( n = 27 157) and 66.3% in 2011 ( n = 33 014) . Additional information on the NHIS is available at http://www.cdc.gov/nchs/nhis.htm .

Definitions

We studied three measures of access-to-care that were defined as follows:

• •
  

  Unable to afford a prescription in past year: selected ‘prescription medicines’ item in response to “DURING THE PAST 12 MONTHS, was there any time when you needed any of the following, but didn’t get it because you couldn’t afford it?”

  
  
• •
  

  Delay in any medical care in the past year because of cost: answered ‘yes’ to “DURING THE PAST 12 MONTHS, have you delayed seeking medical care because of worry about the cost?”

  
  
• •
  

  Did not get needed medical care in past year because of cost: answered ‘yes’ to “DURING THE PAST 12 MONTHS, was there any time when you needed medical care, but did not get it because you couldn’t afford it?”

Arthritis: In this study, self-reported doctor-diagnosed arthritis (herein arthritis) ( n = 21,313; 2009 = 6696; 2010 = 6436; 2011 = 8181 ) was ascertained with a ‘yes’ response to: “Have you EVER been told by a doctor or other health professional that you have some form of arthritis, rheumatoid arthritis, gout, lupus, or fibromyalgia?”

Independent variables

We examined 14 independent variables (eight socio-demographic, two medical care access and four health status characteristics) which were selected based on known associations from previous studies and/or plausibility of associations with access-to-care .

Socio-demographic characteristics . Eight characteristics were studied: age, sex, race, hispanic ethnicity, highest educational attainment, current employment status, marital status and household income-to-poverty ratio. Age was examined as a categorical variable (18–44, 45–64 and ≥65 years) in all analyses. Current employment status (i.e., in the past 2 weeks) was categorised as employed, retired or not working (temporarily disabled, unemployed, disabled, volunteering or not in the work force). The household income-to-poverty ratio is an indicator of the depth of poverty and is used to determine eligibility for US federal and state assistance programmes. A calculated household income-to-poverty ratio variable is provided in the NHIS data set and is equal to the respondent’s household income divided by the poverty threshold for the survey year, which is calculated by the US Census Bureau for households of difference sizes . We categorised this variable as: poor (<1), near poor (1 to <2), and not poor (≥2) ; we also included an unknown (did not report income) category because imputed income data for the 2011 NHIS were unavailable at the time of the analyses.

Medical care indicators . Two variables, health insurance status and total family (out-of-pocket) OOP medical expenses, were included. Health insurance status, a four-level variable (no coverage, public only, private only and public and private), was derived based on responses from a series of NHIS questions about health insurance; this classification allowed the examination of both insurance status and insurance type (i.e., public or private). During data collection, participants reported the type(s) of health insurance coverage they had and provided their health insurance card(s), if available, to reduce misclassification. We defined public insurance as coverage by Medicaid, Medicare, other government-sponsored programmes or a military health plan (TRICARE, VA or CHAMP–VA). People who had catastrophic or dental insurance only or were covered only by the Indian Health Service were classified as being uninsured .

Participants reported their total amount of OOP medical expenses for medical and dental care, excluding health insurance premiums and over-the-counter drugs. Response categories were zero, <$500, $500–1999, $2000–2999, $3000–4999 and ≥$5000. We aggregated the $2000–2999 and $3000–4999 categories because our initial analyses indicated a small number of responses for them.

Health status . Four indicators were examined: self-rated general health, co-morbidity count, (serious psychological distress) SPD and severity of functional limitations. Self-rated general health, defined as a three-level variable (excellent/very good, good and fair/poor) in this analysis, was ascertained with “Would you say your health in general is excellent, very good, good, fair, or poor?” We derived a co-morbidity count variable as follows. Participants were asked to report whether they had ever been diagnosed with any of a series of chronic conditions including heart disease (coronary heart disease, angina, myocardial infarction and other heart disease), stroke, chronic lung conditions (emphysema, current asthma and chronic bronchitis), cancer and diabetes. For each respondent, we calculated the total number of conditions (maximum of five) and categorised this as 0, 1 and ≥2.

We examined SPD with the Kessler 6 (K6), an instrument developed specifically for the NHIS to identify nonspecific SPD . Survey participants were asked to rate, using a five-point scale (0 = none of the time to 4 = all of the time), the extent to which they had experienced each of the following six feelings in the past 30 days: sad, worthless, nervous, restless, hopeless and everything was an effort. Those with a score ≥13 were defined as having SPD .

Severity of functional limitations was defined using the participant’s rating, using a six-point scale (not at all difficult [0], only a little difficult [1], somewhat difficult [2], very difficult [3], can’t do at all [4] and do not do [5]), of their ability to do each of the following nine tasks by themselves and without any special equipment: (1) lift or carry ≤10 pounds; (2) walk or climb ≤10 steps without resting; (3) use fingers to grasp or handle small objects; (4) push or pull large objects; (5) reach up over head; (6) sit for about 2 h; (7) stand or be on feet for about 2 h; (8) stoop, bend or kneel and (9) walk 1/4 mile or three city blocks. For each respondent, we calculated the mean score for the nine items and classified scores as follows: score of 0 = none; > 0 to <1.5 = mild; 1.5 to <2.5 = moderate; and ≥2.5 = severe.

Statistical analysis

Characteristics of population with and without arthritis . To describe the US population of adults aged ≥18 years with and without arthritis, we examined the distribution (weighted) of socio-demographic characteristics, medical care indicators and health status by arthritis status using proportions and 95% confidence intervals (CI).

Cost-attributable compromised access among those with and without arthritis . First, we estimated the prevalence of each of the three measures of cost-attributable compromised access by arthritis status. Next, we examined associations between each of the measures and arthritis status using (prevalence ratios) PRs and 95% CIs in unadjusted and multivariable-adjusted logistic regression models. For each of the three compromised access outcomes, we conducted a series of logistic regressions. In the first step, we estimated the unadjusted association between the outcome and arthritis status. In the second, for each of the 14 independent variables, we estimated the association between the outcome and arthritis status in a model that adjusted for the single independent variable. We identified seven independent variables for which there was a statistically significant association ( α = 0.05) with the outcome as assessed by a Wald F statistic : age, current employment status, health insurance status, self-rated general health, co-morbidity count, SPD and severity of functional limitations. In the third step, we conducted multivariable-adjusted logistic regression modelling to estimate the associations between each outcome and arthritis status, adjusting for these seven variables.

Cost-attributable compromised access among people with arthritis . To identify the people with arthritis most likely to report each of the three compromised medical care outcomes, we examined the prevalence of each outcome stratified by each of the 14 independent variables. Next, we conducted logistic regression modelling. We started by modelling the association between the combination of each outcome and each independent variable. Then, for each combination, we reviewed the magnitude and statistical significance ( α = 0.05) of the Wald F statistic (goodness-of-fit test where larger values indicate better fit). For all of the outcomes, the Wald F statistic values were distributed bimodally across independent variables; values ranged from 100 to 500 for seven variables (age, employment status, income-to-poverty ratio, health insurance status, general health status, SPD and severity of functional limitations) and were statistically significant. In contrast, values for the remaining variables ranged from 2 to 43 and were not all statistically significant. Furthermore, the independent variables with the largest Wald F statistics were also those with the largest PRs. Therefore, in the final multivariable regressions we modelled the seven variables in association with each of the compromised care outcomes.

Because one objective of our overall analysis was to identify the people with arthritis who were experiencing the highest levels of compromised care and who, potentially, had the greatest need for interventions, we emphasised the unadjusted PRs from this analysis, as these reflected the actual relationships that were observed directly in the population and, therefore, that might be most meaningful for clinical and public health practice .

Characteristics of US adults with arthritis who did not have health insurance coverage . Our analyses indicated that health insurance was the strongest correlate of reporting cost-attributable compromised access to medical care. To identify subgroups of US adults with arthritis who were uninsured and were potentially experiencing the highest levels of compromised access, we estimated the weighted number and prevalence of uninsured adults stratified by each of the independent variables.

All analyses were conducted using SUDAAN v10 . To generalise our results to the US population of civilian, non-institutionalised adults aged ≥18 years, we applied sampling weights in all analyses. Standard errors accounted for the survey’s complex design. Throughout this report, statistically significant differences in proportions were those with nonoverlapping CIs and in PRs, those that did not overlap 1.0 .

We analysed 3 years of data combined. There were at least two recent changes in health insurance coverage in the US population during these years that might affect access to medical care. First, coverage potentially increased as a result of the 2010 Patient Protection and Affordable Care Act . Second, approximately half of US health insurance coverage is employer based and, consequently, decreased employment among US working-age adults with arthritis likely resulted in decreased health insurance coverage after 2009 . To determine whether aggregation across years was appropriate, we estimated and compared the proportion of people in each of the four categories of health insurance coverage. Across years, percentages for each group were within 1–2 percentage points, and 95% CIs for these estimates overlapped, indicating estimates were similar and that aggregation was appropriate.