• Open Access

Geographic Disparities in Patient Travel for Dialysis in the United States

Authors


  • This study was sponsored by Amgen Inc. Employees of Amgen participated in the study design and interpretation of the data, and in reviewing, revising, and providing final approval of this manuscript. Stephan C. Dunning has received research and consulting fees from Amgen Inc. John J. Kochevar received consulting and research fees from Amgen Inc. J. Mark Stephens has received consulting and research fees from Amgen Inc. Samuel Brotherton received funds from Kochevar Research Associates and Prima Health Analytics. Ann C. McClellan, Larry C. Emerson, and David T. Gilbertson have received consulting fees and travel reimbursement fees from Amgen Inc. William M. McClellan has received research grant funding and consulting fees from Amgen Inc. David J. Harrison, Shaowei Wan, and Matthew Gitlin are/were employees of Amgen Inc., and own stock or stock options in Amgen Inc. The authors wish to thank Mandy Suggitt who provided editing and graphics support on behalf of Amgen Inc.

For further information, contact: J. Mark Stephens, BA, Prima Health Analytics, 49 Bald Eagle Road, Weymouth, MA 02190; e-mail: jmarkstephens@gmail.com.

Abstract

Purpose

To estimate travel distance and time for US hemodialysis patients and to compare travel of rural versus urban patients.

Methods

Dialysis patient residences were estimated from ZIP code-level patient counts as of February 2011 allocated within the ZIP code proportional to census tract-level population, obtained from the 2010 US Census. Dialysis facility addresses were obtained from Medicare public-use files. Patients were assigned to an “original” and “replacement” facility, assuming patients used the facility closest to home and would select the next closest facility as a replacement, if a replacement facility was required. Driving distances and times were calculated between patient residences and facility locations using GIS software.

Findings

The mean one-way driving distance to the original facility was 7.9 miles; for rural patients average distances were 2.5 times farther than for urban patients (15.9 vs 6.2 miles). Mean driving distance to a replacement facility was 10.6 miles, with rural patients traveling on average 4 times farther than urban patients to a replacement facility (28.8 vs 6.8 miles).

Conclusion

Rural patients travel much longer distances for dialysis than urban patients. Accessing alternative facilities, if required, would greatly increase rural patient travel, while having little impact on urban patients. Increased travel could have clinical implications as longer travel is associated with increased mortality and decreased quality of life.

Health care access for rural patients in the United States is an ongoing concern of government policy makers. A number of studies have identified the challenges and disparities in access to care for rural patients. In particular, the greater distances traveled by patients in rural areas have been widely reported as a significant barrier to access to care, affecting appropriate utilization of health services.[1-5]

The federal government continues to monitor rural health care access for Medicare beneficiaries. As one provision of the 2010 Affordable Care Act, the Medicare Payment Advisory Council (MedPAC) was mandated to evaluate and report on Medicare payment adequacy in rural areas, and to examine access to and quality of care for rural patients.[6] This mandated report, published in June 2012,[7] augments MedPAC's routine monitoring and reporting of access issues for rural patients, which typically includes analysis of the unique problems affecting patients with end-stage renal disease (ESRD) who undergo regular hemodialysis treatment.[6, 8-10]

Currently, there are nearly 5,700 ESRD providers in the United States serving approximately 400,000 dialysis patients.[11] About 90% of these patients receive dialysis at one of the ∼5,500 facilities offering in-center hemodialysis,[12] visiting their facility on average 3 times per week. As dialysis patients typically have to travel for treatment between 140 and 160 times per year, travel distance may affect access to care for these patients more than for most people, even other chronically ill patients with high health care utilization. This may be a particular issue in rural areas where typical travel distances to treatment are longer.

While the higher travel burden on rural patients for health care services has been generally well-documented,[2, 5, 13] there are no comprehensive published studies of dialysis patient travel in the United States, or analyses of dialysis patient exposure to increased travel burden when they have to find alternative treatment sites. This study estimated the road miles and driving times of US dialysis patients, and quantitatively assessed the relative burden on rural versus urban patients, and regional differences in dialysis patient travel. In a secondary analysis, the study estimated the additional travel if patients had to switch to a new facility in their area, choosing from currently existing facilities.

Methods

This was a retrospective modeling study of US hemodialysis patient travel times and distances from their homes to an “original” facility and a “replacement” facility. Original facility was defined as the facility where patients currently receive dialysis services, and replacement facility was the facility that they would transfer to if they needed to find a new facility.

Data Sources

We used publicly available data on dialysis facility locations and dialysis patient ZIP codes of residence. Facility addresses and characteristics were obtained from the Centers for Medicare and Medicaid Services’ (CMS) Dialysis Facility Compare database.[12] Patient counts by 5-digit ZIP code of residence were obtained from a custom report created by the ESRD Networks from their Central Data Repository.[11] Patient counts for each dialysis facility were also obtained from the 2009 CMS Cost Reports.[14, 15] These counts were used to check for errors in patient-facility assignments within the model.

Study Cohort

We limited the study to patients at nongovernment and nonuniversity-affiliated facilities in the 48 continental states that provided in-center hemodialysis as of March 2011 (N = 347,665 patients at 5,114 facilities). Of the 5,114 facilities, 107 had no patients assigned as an original facility. In addition, 15,548 patients were excluded due to mapping problems or missing data. Some patients were in ZIP codes that could not be assigned to a US Census Bureau ZIP Code Tabulation Area (ZCTA) or were assigned to areas within their ZCTA with missing or zero population. A few patients with extreme outlier values on travel distance were also excluded. The final patient cohort used in the distance modeling was 332,117 patients assigned to 5,007 facilities (Table 1).

Table 1. Study Sample of Patients and Facilities
Dialysis PatientsnDialysis Facilitiesn
  1. a

    National Patient Prevalence Report[11] extracted from ESRD Networks Central Repository on March 23, 2011.

  2. b

    Centers for Medicare & Medicaid Services’ Dialysis Facility Compare File,[12] as of February 3, 2011, extracted March 14, 2011.

  3. c

    A total of 5,286 patients were in ZIP codes that could not be assigned to a ZCTA, 10,234 patients could not be assigned to a populated area within their ZCTA, 28 patients were excluded due to extreme outlier values on incremental distance (less than ‒50 miles or greater than 200 miles).

  4. d

    A total of 107 facilities had no patients assigned as the closest facility.

Total US dialysis patientsa396,533Total US dialysis providersb5,510
Modalities other than in-center hemodialysis−34,712No in-center patients or hemodialysis stations−208
Outside contiguous 48 States−7,805Outside contiguous 48 States−77
Not at risk for relocation (government, university)−6,351Facilities where patients not at risk (government, university)−111
Missing datac−15,548Missing datad−107
Patients included in results332,117Facilities included in results5,007

Modeling

To estimate distances patients might have to travel for replacement dialysis services, 3 data points were estimated: (1) where the patient lived; (2) where the patient went for dialysis currently; and (3) where the patient would go if forced to choose a new facility. These 3 data points were mapped to latitude and longitude coordinates, and driving distances and times between them were calculated using geographic information systems (GIS) software. The calculations included:

  1. Distance/time from patient residence to current (original) facility.
  2. Distance/time from patient residence to replacement facility.
  3. Incremental distance/time for travel to replacement facility compared with travel to original facility.

Estimating Patient Residence From ZIP Code-Level Data

The patient address data used were limited to patient counts by ZIP code of residence. ZIP codes were developed by the US Postal Service for managing mail deliveries and do not have well-defined land boundaries. Therefore, we used ZCTAs as a proxy for the geographic land boundaries of patient ZIP codes of residence. ZCTAs are a land division (composed of census blocks) that approximate areas encompassed by ZIP codes. Most ZIP codes had a corresponding ZCTA, but 2,229 codes from the file of ZIP code-level patient counts did not match to a ZCTA. This resulted in the exclusion of 5,286 patients (∼1.5%) from our analysis. The unmatched ZIP codes were plotted on a map, and clustering or tendency toward a certain level of rurality or urbanicity in the excluded ZIP codes was not observed.

The method used to assign patients to a location within the ZCTA of residence was a population-weighted distributed allocation approach. Recent studies of the accuracy of different methods of estimating patient locations within ZIP codes confirm that approaches such as the method used in this study yield accurate results at the individual patient level and are appropriate for studies of area-level risk to patient service access.[16, 17] The method employed was as follows: first, we developed an algorithm to fill the United States with a lattice work of patient location points so that the density of the lattice was much greater in areas with a greater concentration of dialysis facilities. For example, given 2 ZIP codes of the same geographic size, one with 5 dialysis facilities and the other with only 1, the ZIP code with 5 facilities would have many more patient location points assigned, with greater granularity in patient location estimation. This improves the accuracy of the patient travel estimates, assuming the correct service location (dialysis facility) is identified for each patient. Dialysis facility locations were geocoded using street address data from the Dialysis Facility Compare[12] and Microsoft MapPoint 2010 (Microsoft, Redmond, Washington) and GPS Visualizer (Portland, Oregon) software. These locations were plotted on a US map and the map was divided into Voronoi cells—regions around each facility containing all areas on the map geographically closest to that facility (using straight-line distance). Next, the smallest rectangle that fully enclosed each Voronoi cell was constructed, and a 10 × 10 grid of equally spaced points was inserted inside the rectangle. Excluding any points that fell outside of the Voronoi cell, the result was a set of lattice points that formed a gridwork of possible patient residence locations near any given facility. An example of the Voronoi cells and the gridwork of lattice points around dialysis facilities near Gainesville, Georgia, are shown in Figure 1. In this figure, the density of lattice points is greater in areas with greater density of dialysis facilities. Unpopulated areas such as lakes were not given lattice points, since patients should not be assigned to areas without population. The lattice points from all Voronoi cells were combined together into one national file that served as the input for the next stage of the analysis.

Figure 1.

Voronoi Cells and Distribution of Lattice Points for Dialysis Facilities Near Gainesville, GA.

Next, patient weights were assigned to each lattice point. The ideal method would be to weight each lattice point by the number of in-center hemodialysis patients with homes in the point's immediate vicinity; however, only ZIP code-/ZCTA-level patient counts were available. To assign patient counts to each lattice point from the ZCTA-level counts, we weighted each point by population from census tract-level data. The population of each census tract was divided equally over all lattice points in the census tract. The population counts were then aggregated to the ZCTA level by adding up the population of all the lattice points in each ZCTA. The percentage of population assigned to each lattice point within a ZCTA was calculated, and the number of in-center hemodialysis patients in each ZIP code was distributed over all lattice points in the ZCTA corresponding to the ZIP code, weighted by the population values assigned. Lattice points with zero patient weight were discarded (ie, points located in ZIP codes with no in-center hemodialysis patients or in census tracts with zero population). Tables 2 and 3 illustrate this allocation process for 10 example lattice points. Table 2 contains all the data used to estimate patient weights for lattice points A to J, the census tract and ZCTA containing each point, the population of each census tract, and the number of in-center hemodialysis patients in each ZCTA. Table 3 shows how the census tract population was distributed to the lattice points and then used to allocate the hemodialysis patient counts from the ZCTA total to each lattice point.

Table 2. Source Data for Lattice Point-Level Patient Weight Calculations
Lattice pointZCTACensus TractCensus Tract PopulationNumber of Patients (ZCTA)
A458808923832564
B4588089238  
C4588089310975 
D12799568701,0234
E1279956870  
F331019343186096
G33101934521,290 
H3310193452  
I3310193452  
J3310193452  
Table 3. Example Lattice Point Patient Weight Calculations
Lattice PointZCTACensus TractAllocated PopulationPopulation Within ZCTA,%ZCTA Number of PatientsAllocated Patients

Note

  1. Where census tracts overlap ZCTA boundaries, census tract populations may in some cases be allocated twice.

A4588089238162.512.5648
B4588089238162.512.5 8
C4588089310975.075.0 48
D1279956870511.550.042
E1279956870511.550.0 2
F3310193431860.040.010040
G3310193452322.515.0 15
H3310193452322.515.0 15
I3310193452322.515.0 15
J3310193452322.515.0 15

In some ZIP codes, none of the lattice point locations assigned by the program for distribution of patients within the ZIP code were in census tracts for which we had population statistics. In some cases, this was because census tract population was missing from the Microsoft MapPoint software. In other cases, the actual population of the census tract was zero. Therefore, patients in some ZIP codes were not allocated to any lattice points within their ZIP code. This resulted in the exclusion of 10,234 patients from the analysis.

Assignment of Patients to an Original Dialysis Facility

Actual data on each patient's original dialysis facility were not obtained for this study. When the service location of a patient is not precisely known, it is common in area-based health care access research to assume that the service location closest to the patient is utilized. This assumption is also commonly used in studies of dialysis access.[18-20] However, many factors besides distance, such as facility capacity, availability of direct transportation, or the practice location of the patient's nephrologist, affect patient decisions about dialysis facility choice. Application of this “nearest provider” assumption may result in many facilities being over- or undersubscribed, relative to the actual number of patients served by these facilities. To address the weakness of this assumption for assigning patients to dialysis facilities, we first assigned patients to the nearest facility and then compared the number of patients assigned at each facility with reported patient census counts from CMS Cost Reports.[14, 15] Then, the assigned patient counts were corrected by randomly selecting patients from overallocated facilities and reassigning them to other nearby facilities so that the number of patients assigned to each facility in our models achieved a 90% or higher correlation with the actual patient counts reported by facilities in their Cost Reports.

The patient allocation program returned results for 5,007 out of the 5,114 facilities in the study cohort. It did not produce results for 107 facilities because of a variety of special situations. In most of these cases, no patients were allocated to the lattice points in the Voronoi cell of a facility (all patients in the surrounding ZIP codes were closer to other facilities). Most of the facilities with no patients allocated were in large cities, at US borders, and/or on the edge of large bodies of water.

Assignment of a Replacement Dialysis Facility

For assignment of a replacement facility, we assumed patients would choose the next closest facility to where they lived as the replacement facility. In cases where patients were reallocated to an original facility that was not the closest to home, the facility nearest to their home was selected as the replacement facility. Therefore, the replacement facility could in some cases be closer to the patient's home than the original facility, resulting in shorter travel to the replacement facility.

Analysis

Custom software developed in MapPoint and Wolfram Mathematica 8 (Wolfram Research, Champaign, Illinois) was used to calculate the driving distances and times from the patient's residence to the original facility and replacement facility. While the most widely used method in health care geographic access studies for calculating distance between patients and health care services is Euclidean (straight-line) distance,[4, 21] this method tends to underestimate actual travel distances by around 20% to 25% in most cases.[22] Our estimates of travel distance and travel time were based on MapPoint's road networks. Drive times were calculated using MapPoint's built-in “average driving speeds,” which depend on road type but not on construction, traffic, time of day, etc. The distance calculations from lattice points to assigned facilities resulted in a few extreme outlier values when the Voronoi cells of a few coastal facilities tended to be very large and consisted mostly of water. This issue did not appear to bias the results; incremental travel distances in coastal cities such as Los Angeles were approximately the same as in inland cities such as Atlanta. There were few apparent outliers. A total of 28 patients were excluded where incremental travel distances were less than ‒50 miles or greater than 200 miles.

Descriptive analyses of the distance data were performed and summary statistics including medians, interquartile ranges, means, standard deviations, and 95% confidence intervals were reported. For rural/urban analyses, each patient ZIP code was matched to the appropriate rural/urban code from the CMS National Breakout of Geographic Area Definitions by ZIP Code.[23]

Validation and Sensitivity Analysis

While the accuracy of patient assignments to residence locations within their ZIP code in this study will have some margin for error, the model results should be most sensitive to the assignment of patients to dialysis facilities, which will often lie outside the ZIP code of residence. Therefore, as a sensitivity analysis, additional models of patient assignment to dialysis facilities were developed and mean travel distances to original and replacement dialysis facilities were calculated for each model. Using the nearest provider model as the expected lower bound of distance estimates, we developed a range of models that would be expected to produce higher travel estimates, with a random assignment model as the upper bound case. The random model assigned patients to an original facility randomly from 1 of the 3 facilities nearest to their homes, and the replacement facility was also chosen randomly from the 3 nearest facilities (excluding the facility assigned as the original facility). Other models assigned varying percentages of patients to the nearest facility and the remaining percentage were assigned randomly.

To validate the study results, the nearest provider and random model estimates of patient travel distance and time to their original dialysis facility were compared with similar statistics found in recent literature on US patient travel for dialysis services.[6, 24, 25]

Results

The mean one-way driving distance to the original facility was 7.9 miles (Table 4). Rural patients traveled 2.5 times farther than urban patients: 15.9 miles versus 6.2 miles. Overall, mean one-way driving distance to the replacement facility was 10.6 miles, and the mean incremental one-way driving distance was 2.7 miles. Average driving distance (one-way) to a replacement facility was over 4 times farther for rural patients (28.8 miles) than for urban patients (6.8 miles). Travel to a replacement facility (one-way) would require an extra 0.6 miles for urban patients, and an additional 12.9 miles for rural patients. Based on 140 treatments per year as a conservative average for a full-year dialysis patient, the average rural patient would travel an extra 3,600 miles per year, over 8,000 miles total, to a replacement facility. Effects on urban patients appear to be smaller, averaging 1.2 miles of additional round-trip travel per treatment, or about 168 miles per year.

Table 4. Driving Distances to Dialysis Facility for Urban and Rural US Hemodialysis Patients
 N (Patients)Miles (One-Way) Mean ± SD (95% CI)
  1. CI, confidence interval; SD, standard deviation.

  2. Model estimates based on dialysis patients and facilities in 2011.

To original facility  
Urban274,0526.25 ± 5.99 (6.23‒6.27)
Rural58,06515.94 ± 14.26 (15.82‒16.06)
Total332,1177.94 ± 8.87 (7.91‒7.97)
To replacement facility  
Urban274,0526.80 ± 7.19 (6.78‒6.83)
Rural58,06528.83 ± 20.09 (28.67‒28.99)
Total332,11710.65 ± 13.53 (10.61‒10.70)
Incremental distance  
Urban274,0520.55 ± 6.20 (0.53‒0.58)
Rural58,06512.89 ± 15.58 (12.76‒13.02)
Total332,1172.71 ± 9.81 (2.68‒2.74)

Patient Distributions on Estimated Driving Times

A total of 66% of rural patients and 85% of super-rural patients (those in the 25% least densely populated rural ZIP codes) drove at least 16 minutes one-way for dialysis to their original facility. This increased to 90% of rural patients and 99% of super-rural patients if they had to choose a replacement facility. Regarding travel to a replacement facility, 59% of rural patients and 90% of super-rural patients would have to drive more than 30 minutes one-way (Figure 2).

Figure 2.

Distribution of Patients by Urbanicity and Travel Time.

*Rural includes super-rural. CMS defines super-rural areas as the 25% least densely populated rural ZIP codes in the United States. Travel time to replacement facility was shorter for some patients.

Model estimates based on dialysis patients and facilities in 2011.

Regional Differences

Travel to dialysis differed by geographic region of the country, with the lowest driving distances in the Northeast (average of 6.0 miles one-way) and the highest in the South Central states (average of 9.5 miles one-way; data not shown). Large differences were observed in estimates of patient travel times by ESRD Network region, with the percentage of patients traveling more than 30 minutes each way to a replacement dialysis facility ranging from 2% in Region 3 (New Jersey) to greater than 30% in Region 12 (Heartland) and Region 15 (Intermountain; Figure 3). Regions and Networks with the longest patient travel also had disproportionately high percentages of rural patients (data not shown).

Figure 3.

Percent of Patients Traveling >30 Minutes (One-Way), By ESRD Network Region.

Model estimates based on dialysis patients and facilities in 2011.

Validity and Reliability of Estimates

The 95% confidence intervals around the mean mileage estimates in the nearest provider model were very small (Table 4). Even for rural distances, the lower and upper bounds of the 95% confidence intervals were within ±1% of the mean. To test for external validity of our results, the estimated original driving distances for the nearest provider model and the random model were compared with figures reported by MedPAC in the 2011 Report to the Congress[6] (Table 5). MedPAC reported 2008 patient driving distances to dialysis for new Fee-for-Service beneficiaries. Median values for urban versus rural patients were also reported. While there are only 5 data points for comparison, the estimates from our models appeared consistent with the distribution reported by MedPAC. The results from the random model best approximated the MedPAC figures, with a median travel distance (one-way) of 5.9 miles versus 6.0 miles reported by MedPAC. The results from the nearest provider model provided more conservative estimates. The results of each model were also compared with 2 recently published studies on patient travel distance to dialysis services.[24, 25] Both these studies reported the distribution of dialysis patients by travel distance to their current facility. Results were consistent across all studies.

Table 5. Driving Distance Estimates (Miles, One-Way) to Dialysis Facility for US Hemodialysis Patients
 MedPAC Report[6]Nearest Provider ModelRandom Model
  1. MedPAC reported 2008 patient driving distances for dialysis for new Fee-for-Service beneficiaries.[6]

25th percentile2.92.33.1
50th percentile6.04.95.9
75th percentile12.811.111.2
Urban, median5.54.05.1
Rural, median10.410.610.8

Sensitivity Analysis

The results of the nearest provider and random models represent the range of results (lowest and highest) from the sensitivity analysis, with the nearest provider model producing the lowest estimates of all the models and the random model producing the highest estimates, for both travel to the original facility and incremental travel. Estimates of mean driving distance to the original facility across all models tested differed by no more than 10% between the lowest (7.9 miles one-way) and the highest (8.7 miles one-way) when calculated across all patients, and differed by no more than 1.2 miles (8%) for patients in rural areas (15.9 miles one-way in the nearest provider model compared with 17.1 miles one-way in the random model). Mean incremental travel distance estimates across the 4 models differed by no more than 7% between the lowest (2.7 miles one-way) and the highest (2.9 miles one-way) when calculated across all patients, and differed by no more than 1 mile (8%) for patients in rural areas (12.9 miles one-way in the nearest provider model compared with 11.9 miles one-way in the random model). The narrow range of the results from testing various realistic approaches to assigning patients to facilities illustrates that the model estimates were robust under a range of test assumptions.

Discussion

It is intuitively understood that rural patients typically have to travel farther than urban patients for health care, but the effects of this additional travel in the ESRD population are worth underscoring. For patients on hemodialysis, the travel burden of accessing care is particularly heavy due to 2 factors: the number of visits required (3 times per week, every week), and the specialized nature of dialysis services, which leads to relative scarcity of service location choices. For example, a previous study of health care access in rural states found that patients typically have to travel 1.5 times farther for dialysis than for a typical doctor's office visit, and up to 2 times farther for dialysis than for other highly specialized care such as chemotherapy.[2]

There is high variation in travel distances to dialysis care across US regions. However, we found that most of this regional variation in the relative availability of dialysis services is driven by rural versus urban factors as opposed to regional differences, per se. Dialysis centers in the United States tend to be clustered where there is high population density, and they tend to be scarce in thinly populated areas. This is true in all geographic regions of the country. This fact emphasizes the point that rurality, rather than geographic region, is the major factor affecting accessibility of dialysis care.

Although actual travel distance may be less critical to patient treatment decisions than the location of dialysis services close to other important health care and social services that the patient may need to access,[1] long travel times or increased travel may have significant implications for patient health outcomes and quality of life, especially in rural areas. For dialysis patients, studies in the United States and other countries have found that longer driving times are associated with lower treatment attendance, increased mortality, and decreased quality of life.[24-29] Shorter travel times to dialysis have been associated with increased quality of life and patient satisfaction.[27, 30, 31] Previous studies have reported that patients traveling 16 minutes or more (one-way) for dialysis have lower quality of life and higher mortality rates, and these effects trended significantly for patients traveling more than 30 minutes and more than 60 minutes one-way.[27]

Patients in rural areas may face transportation problems and fewer alternatives to in-center hemodialysis.[32-34] Paradoxically, dialysis facilities in rural areas are less likely to offer peritoneal or home-hemodialysis training than urban facilities, despite the reported advantages of home dialysis for rural patients.[34-36] Long travel times can also increase the financial burden for patients and their families, including higher transportation cost, lost productivity, and time and wage loss for family members.[37]

This study showed that rural patients would be disproportionately affected if they were forced to find a new dialysis facility. Such effects are not purely hypothetical; currently, and for the foreseeable future, facilities in rural areas may be particularly vulnerable to financial trouble that could lead to more closures or consolidations. In the latest (2011 and 2012) MedPAC reports, concern was expressed regarding the widening gap in Medicare margins (Medicare revenue minus costs) of urban versus rural facilities, with rural facilities continuing to lose money on Medicare patients.[6, 10] The 2011 report estimated that while the average Medicare margin across all dialysis facilities in 2009 was +3.1%, margins averaged +4.1% for urban facilities compared with –1.4% for rural facilities.[6] In the latest report, MedPAC stated that this gap widened further in 2010. The average Medicare margin for all dialysis facilities dropped to +2.3%, with urban facility margins of +3.4% and rural facility margins falling to –3.7%.[10] MedPAC predicted that the overall average Medicare margin would drop to 1.3% in 2011, although some rural facilities were expected to benefit from the low-volume adjustment included in the new prospective payment method. Finally, in 2012, nearly 30% of rural facilities have experienced some level of additional payment reduction under the Quality Incentive Program.[38] This continuous downward pressure on dialysis facility margins may be expected to result in more facility closures. In fact, MedPAC reported that while about 60 dialysis facilities closed in 2008,[2] 90 units closed in 2009; a 50% increase in 1 year. The 90 closures affected about 3,600 Fee-for-Service patients and were disproportionately in rural areas (31% of closures were rural facilities versus 22% of facilities that remained open in 2009 to 2010).[10]

Facility closures are not the only risk to patient access to local care. As dialysis medical margins decline and pressures increase on profit margins at the large for-profit chains, dialysis patients with private insurance may find themselves turned away if payment rates offered by their insurers are deemed too low.[39, 40] Patients in rural areas are particularly vulnerable to such decisions, as there are often no alternative dialysis facilities nearby.

This study had several limitations. We did not have detailed data on patient residence or dialysis facility affiliation. Modeling techniques were employed to estimate where patients lived within their ZIP code and which dialysis facility they attended. Issues of spatial misalignment between ZIP codes, ZCTAs, and census tracts resulted in some missing data, and travel distances calculated with GIS software included some extreme outlier values that were discarded. While we examined the missing data for obvious patterns that might skew the results, these missing data could potentially have biased the results in unknown ways. Regarding travel distances and times to dialysis, a simplifying assumption was made that patients traveled to dialysis facilities from their homes, rather than from work, shopping, or other locations,[4] and that travel was over road networks at average driving speeds. If patients traveled to dialysis from locations other than home, or via means other than roads, such as subways, for example, this may have resulted in different travel distances and times than what we estimated. Finally, we made assumptions based on standard transportation science methods about which facilities patients would select if forced to find a new service location for dialysis. We do not know of any studies of dialysis patients that can either confirm or deny the validity of these assumptions.

In conclusion, this modeling study indicated that access to dialysis care for rural patients may be considerably affected by reduced facility choice, given that longer travel to care is a barrier to access. By focusing on the potential effects of care disruption, this study can assist in informing policy makers about the potential for increasing disparities in dialysis care access for rural patients. Increased travel for these patients could have significant clinical implications, and exacerbate the existing disparities in the US dialysis system.[1, 2, 27, 30, 32] The results of this study can be linked to previously published evidence of the clinical consequences of increased travel. The results suggest that the incremental travel burden on rural patients of care disruptions caused by changing facilities would increase travel time for most of these patients to levels, which could affect quality of life and mortality. As concern about access disparities for rural dialysis patients continues to be a focal point for policy makers, there needs to be better monitoring of the effects of increased travel on the patients most affected when dialysis facilities close or refuse patients.

Ancillary