Relationships between driving distance, rheumatoid arthritis diagnosis, and disease-modifying anti-rheumatic drug receipt

Abstract

Objective

Disease-modifying antirheumatic drugs (DMARDs) are recommended for all patients with rheumatoid arthritis (RA). Some estimate that approximately one-half of patients with RA do not receive DMARDs. We hypothesized that patients with RA living further from rheumatologists would be less likely to receive RA diagnoses and to receive DMARDs.

Methods

U.S.-based Medicare patients aged ≥65 were study eligible. We calculated driving distance from patients’ homes to the nearest rheumatologist. Using multivariable logistic regression, we assessed relationships between driving distance and RA diagnosis and between driving distance and DMARD receipt. In one set of analyses, distance was divided into quartiles: 0–2, 2.1–5, 5.1–15.9, ≥16 miles. In a second, we used pre-defined categories: 0–15, 15.1–30, 30.1–60, ≥60 miles.

Results

Among 59,426 Medicare beneficiaries, 918 had diagnosed RA. Compared to the first quartile, increased distance was associated with decreased odds of RA diagnosis: second quartile, OR=0.96 (95% CI, 0.80–1.16); third=0.88 (0.72–1.07); fourth=0.72 (0.56–0.93), p for trend=0.01. Similar results were observed using pre-defined categories. Among those with RA, increased distance was associated with increased odds of DMARD receipt across quartiles: second=1.15 (1.06–1.25); third=1.41 (1.29–1.54); fourth=1.32 (1.18–1.46), p for trend=0.001. There was no relationship between pre-defined categories and DMARD receipt, p for trend=0.45.

Conclusion

Increased driving distance to rheumatologists was associated with decreased odds of RA diagnosis. Among those with diagnosed RA, the odds of DMARD receipt rose as distance increased from <2–16 miles, but not beyond. Urban residents living closer to rheumatologists may have barriers to DMARD use besides from geographic access.

Rheumatoid arthritis (RA) is the most common systemic rheumatic disease and is associated with high morbidity and >15 billion dollars in medical care costs annually.¹ Disease-modifying antirheumatic drugs (DMARDs) have a proven role in treating RA and are recommended for all patients,² yet several community-based studies document that >50% patients with RA do not receive DMARDs.^3–5 Preliminary data suggest that patients seeing only non-rheumatologists are much less likely to receive DMARDs than patients who see rheumatologists.^4,6,7 However, this disparity in care, and the interplay amongst access to a rheumatologist, an RA diagnosis, and receipt of a DMARD are not well understood.

Based on literature from other disease settings where treatment disparities exist, many factors likely affect access to a rheumatologist, including individual factors (e.g., age, gender, race, comorbidity) and neighborhood characteristics (e.g., percentage below poverty level, median household income, percentage with <12 years of education) that can be both proxies for individual characteristics and effectors of individual behavior.^8–11 Recent studies have raised concerns about the paucity of rheumatologists.¹² There is also concern about the concentration of rheumatologists in urban centers, which may leave patients underserved if they reside in rural areas. For these reasons, proximity to a rheumatologist, as measured by driving distance, may affect both the likelihood of an RA diagnosis and the likelihood of DMARD receipt. Increased driving distance has been associated with reduced access to primary and specialty care,^13,14 which would suggest a reduced likelihood of receiving guideline-based treatments. However, numerous studies have found a “distance bias,” in which patients who live further away from specialty care are more likely to receive treatments and have better outcomes than patients who live close by.^15–18 This distance bias is thought to be driven by the severity of disease, better prognosis, greater patient resources, and/or greater patient motivation to receive care.^9,16 For example, patients with more aggressive disease may be willing to travel further distances to receive care. Moreover, if they are able to travel, they likely have greater resources with which to access care: Gillis et al. found that Medicaid patients with lupus who traveled the furthest to receive care were those with the highest levels of education.¹⁹

In this paper, we explored the relationships between driving distance and RA diagnosis and driving distance and DMARD receipt, specifically assessing whether a distance bias plays a role in the receipt of DMARDs. We hypothesized that decreased access to a rheumatologist, as measured by increased driving distance, would decrease the likelihood of RA diagnosis. We similarly hypothesized that increased driving distance would decrease the likelihood of DMARD receipt.

PATIENTS AND METHODS

The study was reviewed and approved by the Brigham and Women’s Hospital Institutional Review Board.

Eligible patients

Medicare is a U.S. government-sponsored health insurance program for persons aged ≥65 and persons <65 with selected disabilities. In this study, eligible patients included Medicare beneficiaries age ≥65 years with prescription drug coverage through either a stand-alone Part D plan or a non-Part D retiree drug plan in 2005–2008 that was administered by CVS Caremark, a pharmacy benefits management company. Diagnostic, health care utilization and demographic data from Medicare Parts A, B, and enrollment files were linked to Caremark prescription drug claims. Eligible patients had both Caremark and Medicare eligibility, ≥1 prescription drug claim, and ≥1 Medicare Part A or B claim in the 180 days prior to study entry.

Definition of rheumatoid arthritis

While it is known that the positive predictive value (PPV) of an RA diagnosis would be improved if “receipt of a DMARD” and/or “a rheumatologist visit” were used,²⁰ these are either covariates or outcomes of interest in the present study and thus were not included in the definition of RA. Instead, we employed a widely used definition of RA, despite having a PPV that was <80% in a previous study that used healthcare claims data.²¹ This primary definition comprised patients with ≥2 diagnosis codes for RA (ICD-9 714.xx) associated with inpatient or outpatient visits ≥7 days apart. As sensitivity analyses, we used 2 alternative definitions for RA diagnosis for which the PPVs are unknown. The second definition included patients with ≥3 inpatient or outpatient diagnosis codes for RA ≥7 days apart. The third definition included patients who met the primary definition and additionally had ≥1 prescription for oral glucocorticoids (see Appendix 1) during the baseline period.

Sample used to examine driving distance and rheumatoid arthritis diagnosis

To test the hypothesis that increased driving distance is associated with a decreased likelihood of RA diagnosis, defined using established procedures in administrative datasets,²⁰ we drew a 6% random sample of all eligible patients; 1% were female and ages 65–74; 1% were female and ages 75–84; 1% were female and age 85+; 1% were male and ages 65–74; 1% were male and ages 75–84; and 1% were male and age 85+. The total “sample to assess RA diagnosis” comprised 59,426 patients with and without RA (Figure 1)

Figure 1.

Sample used to examine driving distance and DMARD receipt

To test the hypothesis that increased driving distance is associated with a decreased likelihood of DMARD receipt among patients with RA, we first selected all patients with RA from all eligible patients in the overall cohort using the RA definitions described above. Patients were assessed for an RA diagnosis during a 180-day baseline period that began on the date when they met all patient eligibility criteria (Figure 1). Among all eligible patients in the initial population, the total “sample to assess DMARD receipt” comprised 26,590 patients with ≥2 diagnoses of RA (primary RA diagnosis definition; Cohort 1); 18,254 patients with ≥3 diagnoses (second definition; Cohort 2); and 17,307 with ≥2 diagnoses and ≥1 prescription for an oral glucocorticoid (third definition; Cohort 3).

Exposure: driving distance to the nearest rheumatologist

We obtained a comprehensive de-identified list of 2,192 practicing US-based rheumatologists and their corresponding street addresses in 2010 from the American College of Rheumatology. The exposure of interest was driving distance to the nearest rheumatologist, defined based on the shortest amount of travel time, in minutes, between the patient’s and the nearest rheumatologist’s geocoded street addresses, mapped using Network Analyst (ESRI, Redlands, CA). Driving distance was not normally distributed, with a positive skew of 6.11 and a kurtosis of 130.41 in the sample to assess RA diagnosis and a positive skew of 14.81 and a kurtosis of 481.39 for the Cohort 1 sample to assess DMARD receipt.. For this reason, we chose to define exposure as a categorical variable.

Because relationships between driving distance and the outcomes may be sensitive to the categories used to define exposure, we employed two strategies. In a first set of analyses, we divided the driving distance exposure variable into quartiles. In the sample for RA diagnosis, these quartiles were 0–2.6 miles; 2.61–5.7 miles; 5.71–17.5 miles; and ≥17.6 miles. In the sample for DMARD receipt, these quartiles were 0–2.0 miles; 2.1–5 miles; 5.1–15.9 miles; and ≥16 miles. For a second set of analyses in both the sample to assess RA diagnosis and the sample to assess DMARD receipt, we divided the driving distance exposure variable into 4 pre-defined categories: ≤15 miles; 15.1 – 30 miles; 30.1 – 60 miles; and >60 miles. These categories roughly translate to minutes traveled, and were intended to represent a more intuitive approach to defining driving distance.

Outcomes

In the sample to assess RA diagnosis, the outcome was a diagnosis of RA. After an initial claim for any medical condition, e.g., cardiovascular disease, in a given year, each patient was evaluated as to whether s/he met the primary, second, and/or third definitions for a diagnosis of RA in the following 365 days.

In the sample to assess DMARD receipt, the primary outcome was ≥1 prescription for any DMARD during the 365-day follow-up period. The follow-up period began on the day a patient met the criteria for a diagnosis of RA. Secondary outcomes related to DMARD receipt included: ≥1 prescription for any biologic DMARD; overlapping prescriptions for ≥2 DMARDs for ≥60 days based on filling dates and the expected days supply (“combination DMARD”); and a dichotomous outcome, “majority of time on DMARDs,” defined as having days supply of ≥1 DMARD for ≥50% of days between the index date and the end of follow-up. For the majority of time on DMARD outcome, the DMARD need not be the same agent. DMARDs included in this analysis are noted in Appendix 1.

Covariates

To mitigate potential confounding of our exposure-outcome relationships of interest by socioeconomic, demographic, and clinical factors, we considered covariates at the individual- and neighborhood-levels. Individual-level patient demographics (age, gender, race, and region of the U.S.), type of prescription drug insurance (Medicare Part D plan versus non-Medicare Part D plan), and health care utilization (number of physician visits, unique medications, hospitalizations; use of oral glucocorticoids, non-selective anti-inflammatory drugs or coxibs, and/or opioids; and history of musculoskeletal surgery) were assessed. To evaluate comorbidities, the Deyo²² adaptation of the Charlson comorbidity index was calculated and each patient was also assessed for a history of cancer, diabetes, chronic lung disease, myocardial infarction, and/or ischemic stroke including transient ischemic attack (definitions in Appendix 2).

Neighborhood sociodemographic characteristics were drawn from the American Community Survey 2005–2009²³ at the census block group level. We next calculated the Research Triangle Institute (RTI) socioeconomic status (SES) index, a validated deprivation measure that includes area-level measures of SES.²⁴ The index is comprised of a weighted summary score for each of the following measures: % unemployment, % below the poverty line, % persons aged ≥25 with less than 12^th grade education, % persons aged ≥25 with at least 4 years of college, median value of owner-occupied homes, and median household income. As recommended by RTI, the SES Index was divided into quartiles: 0–49; 50–52; 53–56; and 57–100. A higher RTI Index score indicates higher SES. Next, we determined the number of rheumatologists per capita in the area where each patient resided by linking each patient’s zip code to their corresponding hospital service area (HSA). An HSA is a collection of ZIP codes whose residents receive most of their hospitalizations from the hospitals in that area. Within each HSA, the Dartmouth Atlas has quantified the number rheumatologists, and other specialists, per capita.^25–27 Finally, urban/rural status was defined: if >50% of the people in a census block group were deemed to live in an urban area, then the census block group was considered to be urban.

Statistical analysis

In the sample to assess DMARD receipt, we calculated frequencies for baseline characteristics (Cohort 1, main RA definition) overall and by quartile of driving distance. The number and frequency of RA diagnosis and of DMARD receipt by quartiles and pre-defined categories of driving distance was assessed in the corresponding patient samples. Then, using each of the RA definitions, we used univariable and multivariable logistic regression to evaluate the relationships between driving distance and RA diagnosis and between driving distance and DMARD receipt. All analyses were run first using driving distance quartiles as the exposure and then using pre-defined categories of driving distance as the exposure. Individual-level covariates forced into the multivariable models were age (in 3 categories, 65–75; 75–84; 85+), gender, race, region of the US, number of physician visits, different medications and acute care hospitalizations, use of oral glucocorticoids, use of NSAIDs or coxibs, use of opioids, and history of musculoskeletal surgery. Remaining covariates were entered into the models if their univariate relationship with the outcome was p<0.20. Both individual and neighborhood-level characteristics were applied at the individual-level, because the effect of community-level attributes on the individual and his/her behavior have been described as valid measures of socioeconomic status in their own right.²⁸ We obtained a p for trend for each analysis by calculating the median of each driving distance category and entering these as a continuous variable in multivariable logistic regression models. All analyses were performed in SAS 9.2 (Cary, NC).

RESULTS

Demographic and clinical characteristics of the cohort are depicted in Table 1. With increasing driving distance to the nearest rheumatologist across quartiles, the proportion of white patients and those with oral glucocorticoid and/or opioid use increased, but the proportion of patients with a higher SES index score decreased.

Table 1.

Baseline characteristics of 26,590 patients with rheumatoid arthritis (RA), overall and by quartiles of driving distance

	All patients N= 26,590	Quartiles of driving distance
		0 – 2.0 miles N= 5,532	2.1 – 5.0 miles N= 7,814	5.1 – 15.9 miles N= 6,612	≥ 16 miles N= 6,632
	N(%) or mean ± standard deviation
Age	76 ± 7	77 ± 7	77 ± 7	76 ± 7	75 ± 7
Age 65–74	12,857 (48)	2,345 (42)	3,505 (45)	3,444 (52)	3,563 (54)
Age 75–84	10,308 (39)	2,336 (42)	3,179 (41)	2,432 (37)	2,361 (36)
Age 85+	3,425 (13)	851 (15)	1,130 (14)	736 (11)	708 (11)
Race
White	21,441 (81)	4,049 (73)	5,965 (76)	5,644 (85)	5,783 (87)
Black	2,678 (10)	691 (12)	912 (12)	520 (8)	555 (8)
Other	2,471 (9)	792 (14)	937 (12)	448 (7)	294 (4)
Female gender	19,900 (75)	4,209 (76)	5,926 (76)	4,962 (75)	4,803 (72)
N physician visits	6 ± 5	7 ± 6	6 ± 5	6 ± 5	5 ± 4
N different medications	10 ± 5	9 ± 5	10 ± 5	10 ± 5	10 ± 6
N hospitalizations	0.2 ± 0.7	0.3 ± 0.7	0.2 ± 0.6	0.2 ± 0.6	0.3 ± 0.7
Oral glucocorticoid use	9,124 (34)	1,529 (28)	2,524 (32)	2,447 (37)	2,624 (40)
NSAID or coxib use	8,103 (30)	1,709 (31)	2,364 (30)	1,974 (30)	2,056 (31)
Opioid use	8,492 (32)	1,516 (27)	2,252 (29)	2,217 (34)	2,507 (38)
Any musculoskeletal surgery	398 (2)	68 (1)	104 (1)	109 (2)	117 (2)
Charlson score	2 ± 2	2 ± 2	2 ± 2	2 ± 2	2 ± 2
History of cancer	4,700 (18)	1,010 (18)	1,473 (19)	1,184 (18)	1,033 (16)
History of diabetes	7,940 (30)	1,944 (35)	2,421 (31)	1,809 (27)	1,766 (27)
History of chronic lung disease	1,564 (6)	384 (7)	482 (6)	346 (5)	352 (5)
History of myocardial infarction	388 (1)	100 (2)	115 (1)	95 (1)	78 (1)
History of stroke	1,518 (6)	346 (6)	466 (6)	381 (6)	325 (5)
Percent below poverty line
0–4.9%	8,666 (33)	1,573 (28)	3,043 (39)	2,958 (45)	1,092 (16)
5–9.9%	6,511 (24)	1,271 (23)	1,750 (22)	1,775 (27)	1,715 (26)
10–19.9%	6,490 (24)	1,230 (22)	1,611 (21)	1,228 (19)	2,421 (37)
20+%	4,923 (19)	1,458 (26)	1,410 (18)	651 (10)	1,404 (21)
RTI SES Index
0–49	11,173 (42)	2,253 (41)	2,838 (36)	1,988 (30)	4,094 (62)
50–52	6,236 (23)	1,109 (20)	1,646 (21)	1,686 (26)	1,795 (27)
53–56	6,075 (23)	1,309 (24)	2,050 (26)	2,061 (31)	655 (10)
57–100	3,106 (12)	861 (16)	1,280 (16)	877 (13)	88 (1)
Percentage unemployed
0 – 9.9%	22,239 (84)	4,298 (78)	6,435 (82)	5,955 (90)	5,551 (84)
10 – 14.9%	2,577 (10)	633 (11)	832 (11)	411 (6)	701 (11)
15 – 24.9%	1,403 (5)	480 (9)	426 (5)	187 (3)	310 (5)
≥25%	371 (1)	121 (2)	121 (2)	59 (1)	70 (1)
Median household income	47,471 ± 24,226	47,880 ± 26,989	52,114 ± 27,479	52,683 ± 22,846	36,465 ± 12,721
Percentage low education
0–14.9%	11,187 (42)	2,287 (41)	3,696 (47)	3,504 (53)	1,700 (26)
15–24.9%	6,935 (26)	1,271 (23)	1,755 (22)	1,684 (25)	2,225 (34)
25–39.9%	5,408 (20)	1,060 (19)	1,291 (17)	994 (15)	2,063 (31)
40–100%	3,060 (12)	914 (17)	1,072 (14)	430 (7)	644 (10)
Percentage high education
0–14.9%	15,298 (58)	2,498 (45)	3,819 (49)	3,547 (54)	5,434 (82)
15–24.9%	6,675 (25)	1,648 (30)	2,187 (28)	1,859 (28)	981 (15)
25–39.9%	4,185 (16)	1,218 (22)	1,654 (21)	1,110 (17)	203 (3)
40–100%	432 (2)	168 (3)	154 (2)	96 (1)	14 (0)
Crowding
0–4.9%	20,986 (79)	3,845 (70)	5,827 (75)	5,646 (85)	5,668 (85)
5.0–9.9%	2,453 (9)	561 (10)	751 (10)	497 (8)	644 (10)
10–19.9%	1,657 (6)	545 (10)	636 (8)	259 (4)	217 (3)
20–100%	1,494 (6)	581 (11)	600 (8)	210 (3)	103 (2)
Urban residence	21,227 (80)	5,515 (100)	7,666 (98)	5,260 (80)	2,786 (42)
Rheumatologists per capita HSA; median [interquartile range]	1.15 [0.73, 1.57]	1.44 [1.15, 2]	1.29 [0.97, 1.73]	1.09 [0.77, 1.49]	0.65 [0.42, 1.01]

dx = diagnosis, N=number, NSAID= non-steroidal anti-inflammatory drug, DMARD=disease modifying anti-rheumatic drug; SES=socioeconomic status; HSA=hospital service area, collection of zip codes whose residents receive most of their hospitalizations from the hospitals in that area.

Among patients in the sample to assess RA diagnosis, 1.5% (N=918) had RA as indicated by the Cohort 1 definition; 672 (1.1%) by the Cohort 2 definition and 557 (0.9%) by the Cohort 3 definition (data not shown). In Cohort 1, while 1.7% of patients in the first and second quartiles of driving distance were diagnosed with RA, only 1.5% of patients in the third quartile and 1.4% in the fourth were diagnosed with RA (Table 2). Similar frequencies were observed across pre-defined categories of driving distance among Cohort 1 patients. Cohorts 2 and 3 results are also presented in Table 2.

TABLE 2.

Number and frequency, unadjusted odds, and adjusted odds of RA diagnosis in the 6% patient sample at 365 days, by 2 approaches to define driving distance categories.

	Quartiles of driving distance				Pre-defined categories of driving distance
	0–2.6 miles N= 14,858	2.61–5.7 miles N=14,857	5.71–17.5 miles N= 14,858	≥ 17.6 miles N= 14,853	0–15 miles N=43,000	15.1–30 miles N=7,366	30.1–60 miles N=6,171	≥ 60 miles N=2,889
	N (%) or odds ratio (95% confidence interval)
Cohort 1 definition (2 diagnoses)
N (%)	251 (1.7)	246 (1.7)	221 (1.5)	200 (1.4)	701 (1.6)	106 (1.4)	73 (1.2)	38 (1.3)
Unadjusted	Reference	0.98 (0.82–1.17)	0.88 (0.73–1.05)	0.79 (0.66–0.96)	Reference	0.88 (0.72–1.08)	0.72 (0.57–0.92)	0.80 (0.58–1.12)
Adjusted	Reference	0.96 (0.80–1.16)	0.88 (0.72–1.07)	0.72 (0.56–0.93)	Reference	0.79 (0.63–1.00)	0.63 (0.47–0.86)	0.72 (0.49–1.06)
p for trend	<0.01				0.02
Cohort 2 definition (3 diagnoses)
N (%)	170 (1.1)	184 (1.2)	160 (1.1)	158 (1.1)	502 (1.1)	84 (1.1)	55 (0.9)	31 (1.1)
Unadjusted	Reference	1.08 (0.88–1.34)	0.94 (0.76–1.17)	0.93 (0.75–1.16)	Reference	0.98 (0.77–1.23)	0.77 (0.58–1.01)	0.92 (0.64–1.32)
Adjusted	Reference	1.03 (0.83–1.28)	0.89 (0.71–1.13)	0.77 (0.58–1.03)	Reference	0.83 (0.64–1.10)	0.61 (0.43–0.87)	0.74 (0.48–1.15)
p for trend	<0.05				<0.05
Cohort 3 definition (2 diagnoses + steroid prescription)
N (%)	144 (1.0)	141 (1.0)	136 (0.9)	136 (0.9)	411 (1.0)	71 (1.0)	46 (0.8)	29 (1.0)
Unadjusted	Reference	0.98 (0.78–1.24)	0.94 (0.75–1.20)	0.94 (0.75–1.20)	Reference	1.01 (0.78–1.30)	0.78 (0.57–1.06)	1.05 (0.72–1.53)
Adjusted	Reference	0.94 (0.73–1.19)	0.92 (0.71–1.19)	0.88 (0.64–1.21)	Reference	0.92 (0.69–1.23)	0.71 (0.48–1.05)	0.98 (0.62–1.56)
p for trend	0.52				0.58

In multivariable logistic regression analysis, there was a trend towards a decreasing likelihood of RA diagnosis as driving distance quartile increased. As compared to patients in the first quartile of driving distance, patients in the second quartile were 0.96 times as likely to be diagnosed with RA (95% CI, 0.80–1.16); third, 0.88 (0.72–1.07); fourth, 0.72 (0.56–0.93); p for trend < 0.01. Similarly, there was a trend towards a decreasing likelihood of RA diagnosis as pre-defined categories of driving distance increased, p for trend=0.02. Similar trends were observed when the Cohort 2 definition was used. When the Cohort 3 definition (which requirement ≥1 oral glucocorticoid prescription) was used, the trends were not significant.

Among patients in the sample to assess DMARD receipt, 10,661 (40%) patients met the Cohort 1 RA definition and received any DMARD in the 365 days following diagnosis (data not shown). When driving distance exposure was divided into quartiles, patients with RA were more likely to receive a DMARD as driving distance increased: first quartile N=1,697 (31%), second N=2,826 (36%), third N=2,979 (45%), and fourth N=3,159 (48%) (Table 3). Patients with RA were also more likely to spend the majority of time on DMARDs as driving distance increased: first quartile N=1,239 (22%) versus fourth quartile N=2,408 (36%). The same patterns were observed when pre-defined categories of driving distance were used.

TABLE 3.

Number and frequency, unadjusted odds, and adjusted odds of DMARD use at 365 days in Cohort 1, by 2 approaches to define driving distance

	Quartiles of driving distance				Pre-defined categories of driving distance
	0–2.0 miles N= 5,532	2.1–5.0 miles N=7,814	5.1–15.9 miles N=6,612	≥ 16 miles N=6,632	0–15 miles N=19,698	15.1–30 miles N=3,083	30.1–60 miles N=2,579	≥ 60 miles N=1,230
	N (%) or odds ratio (95% confidence interval)
Receipt of any DMARD
N (%)	1,697 (31)	2,826 (36)	2,979 (45)	3,159 (48)	7,395 (38)	1,419 (46)	1,249 (48)	598 (49)
Unadjusted	Reference	1.28 (1.19–1.38)	1.85 (1.72–2.00)	2.06 (1.91–2.22)	Reference	1.42 (1.31–1.53)	1.56 (1.44–1.70)	1.57 (1.40–1.77)
Adjusted	Reference	1.15 (1.06–1.25)	1.41 (1.29–1.54)	1.32 (1.18–1.46)	Reference	1.09 (0.99–1.19)	1.03(0.91–1.16)	1.06(0.91–1.23)
p for trend	0.001				0.45
Receipt of a biologic DMARD
N (%)	300 (5)	465 (6)	463 (7)	429 (6)	1,208 (6)	192 (6)	172 (7)	85 (7)
Unadjusted	Reference	1.10 (0.95–1.28)	1.31 (1.13–1.53)	1.21 (1.04–1.40)	Reference	1.02 (0.87–1.19)	1.09 (0.93–1.29)	1.14 (0.91–1.43)
Adjusted	Reference	1.02 (0.87–1.19)	1.05 (0.89–1.24)	1.00 (0.81–1.22)	Reference	0.96(0.80–1.15)	1.08(0.86–1.35)	1.07(0.80–1.43)
p for trend	0.80				0.60
Receipt of combination DMARDs
N (%)	303 (5)	512 (7)	546 (8)	583 (9)	1,337 (7)	261 (8)	237 (9)	109 (9)
Unadjusted	Reference	1.21 (1.05–1.40)	1.55 (1.34–1.80)	1.66 (1.44–1.92)	Reference	1.27 (1.11–1.46)	1.39 (1.20–1.61)	1.34 (1.09–1.64)
Adjusted	Reference	1.09 (0.93–1.28)	1.17 (1.00–1.37)	1.23 (1.02–1.49)	Reference	1.11(0.95–1.31)	1.16(0.95–1.42)	1.07(0.83–1.39)
p for trend	0.08				0.04
Majority of time on DMARDs
N (%)	1,239 (22)	2,139 (27)	2,291 (35)	2,408 (36)	5,584 (28)	1,082 (35)	960 (37)	451 (37)
Unadjusted	Reference	1.31 (1.21–1.42)	1.84 (1.69–1.99)	1.98 (1.82–2.14)	Reference	1.37 (1.26–1.48)	1.50 (1.38–1.63)	1.46 (1.30–1.65)
Adjusted	Reference	1.17 (1.07–1.28)	1.41 (1.29–1.55)	1.33 (1.19–1.49)	Reference	1.07(0.97–1.18)	1.08(0.95–1.22)	1.04 (0.89–1.22)
p for trend	0.003				p for trend: 0.45

In multivariable logistic analyses, increased driving distance, modeled in quartiles, was significantly associated with an increased likelihood of receiving any DMARD. As compared to patients with driving distances in the first quartile, second quartile patients had a 15% greater odds of receiving any DMARD (95% CI, 1.06–1.25), third quartile patients a 41% greater odds (1.29–1.54) and fourth quartile patients a 32% greater odds (1.18–1.46); p for trend=0.001. Similarly, compared to first quartile patients, increased distance to a rheumatologist was associated with an increased likelihood of spending the majority of time on DMARDs; second quartile OR=1.17 (95% CI, 1.07–1.28); third=1.41 (1.29–1.55); fourth=1.33 (1.19–1.49); p for trend=0.003. There was a trend towards increased odds of combination DMARD receipt with increased driving distance. No relationship between increased distance and receipt of a biologic DMARD was observed.

In contrast, when driving distance was modeled using pre-defined categories, no relationships between increased driving distance and any of the DMARD receipt outcomes were observed. For example, as compared to patients who lived 0–15 miles from the nearest rheumatologist, patients in higher categories had no greater or lesser odds of DMARD receipt: 15.1–30 miles, OR=1.09 (95% CI, 0.99–1.19); 30.1–60 miles, 1.03 (0.91–1.16); ≥60.1 miles, 1.06 (0.91–1.23), p for trend=0.45. Analogous results to those observed in Cohort 1 were found in Cohort 2 and Cohort 3 (data not shown).

In each of the multivariable models, multiple covariates were associated with DMARD receipt. Positive associations included rural versus urban residence, OR=1.24 (1.12–1.36) and higher scores (indicating higher SES) on the RTI SES Index, OR=1.12 (1.09–1.16) per 10 point increase. Characteristics negatively associated with DMARD receipt included diagnoses of diabetes, OR=0.75 (0.70–0.81); chronic lung disease, OR=0.73 (0.64–0.83); cancer, OR=0.89 (0.82–0.97); and/or myocardial infarction/stroke, OR=0.62 (0.58–0.67); all presented results are from the pre-defined driving distance model.

DISCUSSION

Among a sample of patients aged ≥65 with comprehensive medical and prescription drug coverage through the U.S. government-sponsored Medicare program, we examined relationships between driving distance and RA diagnosis and driving distance and DMARD receipt. We observed that increased driving distance to the nearest rheumatologist was associated with decreased odds of RA diagnosis, a result that was consistent across driving distance definitions. Among patients with an RA diagnosis, the relationship between driving distance and DMARD receipt was sensitive to the exposure definition: while increased driving distance quartiles were associated with a 12–32% increased odds of receiving any DMARD, no association was found between pre-defined driving distance categories and DMARD receipt. Our observations suggest that driving distance acts as a proxy for at least some of the burden of obtaining rheumatologic care but has less importance in determining DMARD receipt.

The association we observed between increased driving distance and decreased odds of RA diagnosis is important for both healthcare policymakers and the rheumatology community as they strategize ways to maximize the impact of a limited rheumatologist workforce. These results are in agreement with other studies that have found increased driving distance to be associated with reduced access to specialty health care services. Chan et al. found that increasing rurality was associated with an 8% increase in the share of overall visits to generalist providers and a corresponding 10% decrease in the share of visits to medical specialists.¹⁴ Currently, the Health Professional Shortage Area Physician Bonus Program offers a 10% bonus to physicians, including rheumatologists, who practice in primary care HPSA and provide services to Medicare beneficiaries.²⁹ Whether the program has expanded access to rheumatology care for elderly beneficiaries is beyond the scope of this study but merits consideration.

In evaluating the relationship between driving distance and DMARD receipt among those with a diagnosis of RA, our results depended on the granularity of the category definitions. When driving distance was defined in quartiles, with finer discrimination amongst shorter distances (75% of patients lived within 16 miles), patients who lived further from the nearest rheumatologist were more likely to receive DMARDs. This paradox is known as “distance bias,” where unmeasured factors such as disease severity, patient resources and motivation may explain better treatment among patients who live greater distances from care. One potential interpretation of our quartile findings is that patients who live a greater distance from a rheumatologist and are still diagnosed with RA are the patients who possess greater resources, severity of disease and/or motivation for care that drives them to seek out the best treatment for their condition, i.e., DMARDs. Such interpretations have been described frequently in the oncologic and other literature.^15–18 In a multi-center study of 1,479 patients with multiple myeloma, patients who lived 0–9 miles from the center had a 1.34 times greater odds of death than patients who lived > 150 miles away; patients who lived 10–49 miles, OR=1.25; and patients who lived 50–149 miles, OR=1.19.¹⁷ In another study among 716 patients with lymphoma, the survival of patients who traveled long distances was better than that of patients who lived nearest to the hospital, even though patients who traveled long distances were more likely to have aggressive and advanced disease.¹⁸ Ballard et al. observed that those who traveled <10 miles for treatment at a referral center hospital had a 20% increased odds of death as compared to those who traveled ≥10 miles.¹⁵

Despite evidence of distance bias, other studies have found conflicting results. In a recent study that better controlled for the severity of patients’ disease prior to surgical intervention, no distance bias was observed across quintiles of driving distance, although these quintiles encompassed substantial distances: mean for quintile 1=15.9 miles, quintile 2=56.1, quintile 3=117.6, quintile 4=232.7, quintile 5=702.0.⁹ Strauss et al. observed that increased driving distance was associated with decreased glycemic control among patients with diabetes and postulated that their results might be due to providers’ reluctance to prescribe more aggressive medication regimens that would increase patients’ risk of hypoglycemia; the mean (standard deviation) for driving distance traveled was 12.2 (16.7) kilometers.¹³ With these examples in mind, we sought corroborating evidence for our initial results by using an alternative definition of the exposure, one which focused on differences across larger distances but did not make fine distinctions for distances less than 15 miles, by using pre-defined categories of driving distance meant to approximate travel time. In these analyses, we observed no relationship between driving distance and DMARD receipt. Once a patient received an RA diagnosis, treatment in accordance with recommended guidelines was as likely across patients. Taken together, our two approaches to defining driving distance and our corresponding results describe only some of the disparity in DMARD receipt, and then only across very short (<16 miles) as oppposed to longer distances (≥15 miles). This suggests that driving distance plays a minimal role in DMARD receipt, where >70% of our population reside, and that other factors such as socioeconomic barriers, clinical comorbidities and disease severity, and health seeking behaviors may better explain previously observed disparities in DMARD use.^8–11

With the increasing availability of geographic information systems software and comprehensive data sources that include street addresses, researchers have expanded opportunities to evaluate relationships between driving distance, access to care, and health outcomes. Our study points to the sensitivity of these relationships to the definition of driving distance and suggests that multiple definitions should be employed to test the robustness of observed results. Our results also raise questions as to whether driving distance is a good measure of access to medical care and to medical treatments and procedures. Prescription drug receipt may sometimes be dependent on diagnosis, a likely explanation for the significant trends we observed when the RA diagnosis definition included only diagnosis codes (Cohorts 1 and 2) as opposed to the non-significant trends we observed when the RA diagnosis definition required a prescription for ≥1 oral glucocorticoid (Cohort 3) In our study, diagnosis was more strongly related to distance; this may hold implications for research examining the utilization of medical interventions or prescription drug treatments based on geography. The sociodemographic characteristics of patients in specific geographies must also be considered. In the U.S., persons living in urban areas often have lower socioeconomic status than those living in suburban or even rural areas, whereas in other countries, persons living in urban areas have higher socioeconomic status as compared to those living in rural areas. This information provides further context for our U.S.-based results. A strength of our study is that we were able to adjust for multiple individual- and neighborhood-level characteristics that are associated with SES and potential disparities in personal resources that might explain the relationship between driving distance and DMARD receipt.^9,15,26,30 Of note, >70% of patients lived within 15 miles of a rheumatologist; patients who live further away from rheumatology care may be underrepresented in our data due to the potentially reduced likelihood of having prescription drug coverage. Our data source includes patients enrolled in the U.S. government-sponsored Medicare program who have comprehensive health coverage. Our population is 65 years or older, and so our results may not generalize to younger patients. Although DMARD treatment is recommended for all patients with RA,² we acknowledge that physicians’ and patients’ decision-making regarding treatment options may be substantially different depending on patients’ age, treatment goals, risk versus benefit considerations, and overall lived experience. Finally, elderly patients’ access to RA diagnosis and treatment may be different than for younger patients. While comprehensive, our data source for patients does not have information on RA disease severity or on patients’ motivation/health seeking behavior, both thought to be associated with distance bias. Because information on these factors is unavailable, it is not possible to quantify their impact on the results we observed. Data sources with this information may further elucidate the impact of these factors. Another limitation is the low PPV of our primary definition of RA. We used two alternative definitions to explore this and found analogous results.

There is strong consensus regarding the utility and importance of DMARDs in treating RA.² Disparities in DMARD receipt have been well documented and continue to merit concern.^3–5 In this study, we explored the relationships between driving distance and RA diagnosis and driving distance and DMARD receipt. Increased driving distance to a rheumatologist was associated with decreased odds of receiving an RA diagnosis. Once patients did receive the diagnosis, the relationship between driving distance and DMARD receipt depended on the distance traveled: within approximately 16 miles, patients with RA had an increased relative odds of DMARD receipt as driving distance increased. Beyond 16 miles, we did not find such distance bias, observing an essentially flat relationship between driving distance and DMARD receipt. These data points together suggest that urban residents who live closer to a rheumatologist likely have other financial or socioeconomic barriers to DMARD use. While driving distance to the nearest rheumatologist, as a proxy for access, appears to play an important role in receiving an RA diagnosis, it does not appear to explain much of the variability in DMARD receipt. Interventions to shorten the distance to rheumatologic care will likely increase the incidence of RA diagnosis but are unlikely on their own to lead to improved rates of DMARD use.

SIGNIFICANCE AND INNOVATIONS.

• In a sample of 59,426 patients aged ≥65 in the U.S.-based Medicare insurance program, increased driving distance to a rheumatologist was associated with decreased odds of RA diagnosis.

• Among those with diagnosed RA, the odds of DMARD receipt rose as distance increased from <2–16 miles, but not beyond.

• Our results underscore the importance of conducting sensitivity analyses that consider alternate exposure definitions and raise questions as to whether driving distance is a good measure of access to medical care and to medical treatments and procedures.

• Other factors such as socioeconomic barriers, clinical comorbidities, disease severity, and health seeking behaviors may better explain observed disparities in DMARD receipt.

APPENDIX 1

Oral glucocorticoids

a. Prednisone
b. Prednisolone
c. Methylprednisolone	(Medrol)

DMARDs

1. Non-biologic DMARDs

a. Methotrexate	(Rheumatrex)
b. Sulfasalazine	(Azulfidine)
c. Hydroxychloroquine	(Plaquenil)
d. Leflunomide	(Arava)
e. Cyclosporin	(Neoral)
f. Gold	(Auranofin or myochrysine or aurothioglucose solganal or ridaura)
g. D-penicillamine	(Cuprimine)
h. Azathioprine	(Imuran)

2. Biologic DMARDs

a. etanercept	(Enbrel)
b. adalimumab	(Humira)
c. infliximab	(Remicade)
d. rituximab	(Rituxan)
e. abatacept	(Orencia)
f. anakinra	(Kineret)
g. tocilizumab	(Actemra)
h. certrolizumab pegol	(Cimzia)
i. Golimumab	(Simponi)

APPENDIX 2

Covariate definitions

Covariate	Definition
Cancer	ICD-9 140.x-195.x, 196.x-198.x, 199.x, 200.x-208.x, 230.x-234.x, 235.x-238.x, 239.x
Diabetes	ICD-9 250.xx and/or ≥2 prescriptions for the following drugs: insulin, acarbose, miglitol, metformin, acetahexamide, chlorpropamide, glimepiride, glipizide, glyburide, repaglinide, tolazamide, tolbutamide, pioglitazone, rosiglitazone, troglitazone, sitagliptin, repaglinide, nateglinide, exenatide
Chronic lung disease	ICD-9 490.0x-491.9x, 494.0x-494.9x, 496.0x-496.9x
Myocardial infarction	Hospitalization episode lasting at least 3 days and no more than 180 days with one of the following diagnosis codes listed as the primary or secondary diagnosis: 410.01, 410.11, 410.21, 410.31, 410.41, 410.51, 410.61, 410.71, 410.81, or 410.91; DRG 121, 122, 123
Ischemic stroke	Inpatient hospitalization with diagnosis in any position of: 430.x, 431.x, 433.x1, 434.x (exclude 434.x0), 436.x