Geographic Isolation and the Risk for Chronic Obstructive Pulmonary Disease–Related Mortality

Chronic obstructive pulmonary disease (COPD) is the fourth most common reason for acute medical hospitalizations nationwide (1, 2). Mortality related to COPD is increasing, and COPD is projected to be the third leading cause of death by 2020 (3). As the burden of this chronic disease grows, efforts have begun to identify patient characteristics and organizational factors (for example, hospital volume and interventional resources) of health care delivery that potentially contribute to mortality.

Emerging evidence suggests that persons who reside in rural areas have worse health outcomes than those who reside in nonrural areas. For example, rural patients with cardiovascular disease have decreased access to care that manifests as longer wait times; lower rates of percutaneous coronary interventions when they are admitted for acute myocardial infarction; and an increased likelihood of being examined at smaller-volume hospitals, which may result in worse clinical outcomes than being examined in other settings (4–7). Rural patients may also have reduced access to medical advances that are highly dependent on physician distribution (8, 9).

Limited evidence is available to indicate that rural patients with COPD have these same disparities. However, in countries outside of the United States with large rural populations, COPD is more prevalent in rural areas than in urban areas. This disparity is probably related to such factors as increased exposure to indoor air pollutants, tobacco smoke, or agricultural exposure and suggests that the burden of COPD may also be higher in rural settings in the United States (10).

As a result of the increasing COPD-related mortality rates and emerging evidence that rural patients may experience health care disparities, we conducted this study to compare 30-day mortality in urban and rural patients admitted for COPD exacerbation. We also intended to determine the degree to which differences between urban and rural mortality rates might be influenced by the characteristics of the admitting hospital, such as volume and the composition of rural patients. We hypothesized that rural residence would be associated with increased mortality but that hospital characteristics would attenuate any increase in mortality.

Methods

Data Sources

The Veterans Affairs (VA) Patient Treatment File was used to identify all consecutive COPD admissions from October 2006 to September 2008. Admissions identified were linked by using unique identifiers to the following additional VA data sources: Decision Support Service, Outpatient Care Files, Vital Status Files, and Enrollment Files. Each of these data sources and the elements contained within have been detailed elsewhere (5).

Study Sample

The process of selecting the sample was accomplished by using codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM), as adapted from extant literature (11–14). A master file containing all consecutive admissions to acute care VA facilities during the study period was used to identify patients with a principal diagnosis of COPD exacerbation, acute or chronic bronchitis, or chronic obstruction of the airway not classified elsewhere (ICD-9-CM codes 490, 496, or 491.20 to 491.28, respectively) or acute and chronic respiratory failure (codes 518.81 to 518.84) (n = 49 493).

We randomly selected 1 record for patients with multiple admissions (n = 12 744 records). We excluded admissions with principal and secondary ICD-9-CM codes other than acute exacerbation of COPD (code 491.21) (n = 6646). We also excluded admissions for patients who had no outpatient encounters during the year before their admission to minimize selection bias and ensure that patients were receiving ongoing care in the VA system (n = 3023) (15), patients who were admitted to facilities without acute care status (n = 264), and patients who were not initially admitted to an acute care medical ward (n = 215). The final sample comprised 26 591 admissions.

Data Elements and Outcome

The primary outcome of our study was 30-day mortality from the day of admission, identified by the VA Vital Status Files (16, 17). The primary independent variable of interest was rural residence, defined by using a ZIP code–level approximation of the rural-urban commuting area (RUCA) codes (18, 19). These codes characterize census tracts on rural and urban status and relationships by using standard U.S. Census Bureau Urbanized Area and Urban Cluster definitions combined with commuter information.

The RUCA algorithm creates 30 mutually exclusive categories that represent population density and affinity to nearby urban centers. We categorized these 30 categories into 4 previously defined categories: urban areas, large rural towns, small rural towns, and isolated small towns. We then collapsed the large rural town and small rural town categories into the category of rural, which ultimately led to the creation of 3 categories of urban–rural designation: urban, rural, and isolated rural.

Hospital-Level Variables

We were interested in hospital volume and hospital rurality as variables potentially influencing the association between patient rurality and COPD mortality. We defined hospital volume as the total number of COPD-related admissions to each VA facility spanning the 3-year study period and categorized this variable into tertiles: low (35 to 235 admissions), medium (236 to 399 admissions), or high (>400 admissions). Each tertile contained 43 hospitals.

We defined hospital rurality as the mean proportion of COPD admissions from RUCA-designated rural areas and categorized this variable into tertiles: major urban (1% to 15% of admissions from rural areas), rural (16% to 38% of admissions from rural areas), or major rural (>38% of admissions from rural areas). We used these categories to characterize the urban–rural patient constituencies of hospitals across categories of hospital volume but used a continuous variable for hospital rurality in our mixed-effects models.

Patient-Level Variables

We collected information about the following patient-level variables: age; sex; race (white, black, other nonwhite, or missing); admission source (nursing home, other VA medical center, other non-VA medical center, home, or other); travel time in minutes to the nearest VA medical center using existing roads, traffic patterns, and posted speed limits (0 to 15, 16 to 30, 31 to 60, 61 to 90, or >90 minutes); comorbid conditions (based on the algorithms of Elixhauser and recently updated using ICD-9-CM diagnosis codes available on the inpatient admission record [20]); mechanical ventilation (procedure codes 96.70 to 96.72) or biphasic positive airway pressure ventilation (code 93.90) on admission as surrogate measures of severity; and laboratory values within 48 hours of admission previously shown to predict mortality in COPD (serum sodium, albumin, and blood urea nitrogen levels; hematocrit; leukocyte count; and Paco₂) (21–25).

Previous records from outpatient source files were used to fill in missing data from the inpatient source files. Data on race remained missing for 1197 admissions (4.5% of the study population). Because substantial data remained missing for all laboratory values (sodium level [38.6%], albumin level [57.4%], blood urea nitrogen level [53.7%], hematocrit [51.5%], leukocyte count [49.3%], and Paco₂ [51.2%]), we imputed laboratory values 20 times for each laboratory test, as recommended for frequent missing data (26, 27). Final multivariable models were generated by using the 20 imputed data sets and the Rubin strategy (26, 27) to combine analyses of multiple imputations and generate valid statistical inferences that appropriately reflect the uncertainty due to missing values.

Statistical Analysis

We assessed differences in patient characteristics by urban–rural status by using generalized linear models to account for clustering within hospitals and compared differences in unadjusted 30-day mortality across the tertiles of hospital volume and hospital rurality by using chi-square tests. We developed multivariable models in accordance with our previous work (5) to estimate associations between patient rurality and mortality. We assessed bivariate relationships between all patient-level variables and mortality by using t tests for dichotomous variables and analysis of variance for ordinal, or continuous, variables, and we generated logistic regression models for each laboratory test to establish categories that maximized discrimination for predicting mortality on the basis of the c-statistic.

Variables that were statistically significant (P < 0.01) and those that were clinically relevant regardless of statistical significance were then considered for inclusion in multivariable models by using a stepwise logistic regression process to identify variables independently related to mortality. Final mortality models were estimated as generalized estimating equations with binomial distributed errors and a logit link function. Models assumed an independence working correlation matrix with SEs and CIs calculated by using Huber–White robust estimates (28, 29) to account for the clustering of patients within VA medical centers.

We assessed the relationship between patient rurality and mortality by first adjusting for patient characteristics and then for hospital characteristics. We used the decomposition approach as recommended by Begg and Parides (30) to separately evaluate the influence of patient-level rurality and hospital mean rurality or hospital volume on mortality. The Appendix Table shows the variables and corresponding parameter estimates, CIs, and P values for each of the covariates included in the final models. We conducted a sensitivity analysis using a model that did not include distance because risk for death was inversely related to the travel time to the nearest VA medical center and separate sensitivity analyses to assess the extent to which omitted confounding might influence study estimates (31).

The authors had full access to and take full responsibility for the integrity of the data. All analyses were conducted by using SAS, version 9.2 (SAS Institute, Cary, North Carolina). The study was approved by the University of Iowa institutional review board and the Iowa City VA Medical Center Research and Development Committee.

Role of the Funding Source

Our study was funded by the Veterans Health Administration Office of Rural Health, Veterans Rural Health Resource Center—Central Region, and the Health Services Research and Development Service, through the Center for Comprehensive Access and Delivery Research and Evaluation, U.S. Department of Veterans Affairs. The funding sources had no role in study design; data collection, analysis, or interpretation writing of the report; or the decision to submit the report for publication.

Results

We identified 18 809 patients (70.7%) with a COPD admission from urban areas, 5671 (21.3%) from rural areas, and 1919 (7.2%) from isolated rural areas. In the overall study sample, only 192 patients (<1% of study participants) were missing the residence designation. After clustering within hospitals was accounted for, the mean age was similar across the 3 groups of patients (Appendix Table). However, patients from urban areas were more likely to be female and nonwhite than were those from rural areas.

Small differences existed in comorbid conditions and laboratory values between urban and rural patients, but most were not statistically significant after accounting for clustering within hospitals. Average drive time to the nearest VA medical center was significantly longer for patients in rural and highly rural areas (P < 0.001) (Appendix Table).

Overall unadjusted mortality was higher for patients from isolated rural areas (5.0%) and rural areas (4.0%) than for those from urban areas (3.8%) (P = 0.002) but differed according to hospital characteristics. The most prominent finding was that mortality was increased in rural hospitals with lower volumes of patients admitted for COPD but remained similar among major rural hospitals with higher volumes (Table 1). Patients from rural and isolated rural areas were more likely to be admitted to hospitals with low volumes of COPD admissions, whereas patients from urban areas were more likely to be admitted to hospitals with high volumes of COPD admissions (P < 0.001) (Table 2).

Patients from isolated rural areas but not rural areas were at higher risk for death (adjusted odds ratio [OR], 1.42 [95% CI, 1.07 to 1.89]; P = 0.016) after adjustment for clinical characteristics, the proportion of COPD hospital admissions coming from rural areas, and the volume of COPD admissions. Patients admitted to hospitals with low volumes of COPD admissions had an increase in mortality that was borderline statistically significant (OR, 1.28 [CI, 0.99 to 1.66]; P = 0.059).

Longer distances to travel to the nearest VA medical center were associated with lower 30-day mortality (Appendix Table). In a sensitivity analysis excluding the measure of travel time to the nearest VA medical center, the odds of death for isolated rural veterans were no longer statistically significant (OR, 1.20 [CI, 0.93 to 1.54]; P = 0.160). The direction, magnitude, and statistical significance of risk for mortality did not change in sensitivity analyses to assess the influence of an omitted confounder on our estimates, assuming that, first, an association exists between the omitted confounder and rural status that is equal in magnitude to that between mortality and rural status and, second, an association exists that is equal to an OR of 2.5 (30).

Discussion

In this study of the association between urban or rural residence and COPD mortality, we found that patients from isolated rural areas had higher mortality and that hospital characteristics did not account for this finding. Mortality differed by hospital volume and rurality, but patients from isolated rural areas remained at increased risk for mortality even after adjustment for those factors. Mortality was not increased in patients living in nonisolated rural areas.

Research investigating potential rural disparities among patients with COPD is sparse. Research to date has been limited to large, self-report surveys describing differences in health-related quality of life between rural and urban veterans (32, 33); differences on the use of primary, specialist, and mental health care (34); and differences in cardiovascular process measures and mortality outcomes between rural and urban hospitals (6, 35–38) or between rural and urban patients (5).

Few works in the U.S. medical literature specifically examine the effect of rurality on COPD prevalence, disease severity, or health care delivery. However, epidemiologic evidence from countries outside of the United States describes disparate outcomes for rural patients compared with those of urban patients. These studies suggest that the higher prevalence of COPD in rural settings is linked to increased indoor air exposures from the burning biomass fuels among rural persons (39–45). Further studies among rural populations in these countries have reported problems with underrecognition of disease (46), deficits in the delivery of pulmonary rehabilitation (47, 48), limited use of spirometry to document disease progression, and problems with access to medical care (49). One study also reported higher rates of COPD-related hospital admissions for rural patients compared with urban patients (50).

Identifying disparities between urban and rural mortality among hospitalized patients with COPD is complex and, at a minimum, requires modeling that incorporates variables at both the patient and hospital levels. One study tried to identify factors at the hospital or organizational level that influenced COPD-related mortality (51) and reported several characteristics related to hospital type and available resources that accounted for the large variation of COPD-related mortality, which ranged from 9% to 19%. Lower-volume, or smaller, hospitals had the highest COPD mortality rates, fewer full-time specialty board staff, and fewer noninvasive ventilatory support resources.

Associations between volume and outcome have been reported in other settings and conditions. For example, higher volume and hospital teaching status were associated with improved mortality among surgical patients admitted to VA medical centers (52). An inverse relationship between hospital volume and mortality was demonstrated for myocardial infarction, pneumonia, and heart failure with a threshold, beyond which there was no further reduction in mortality as volume increased (7). The borderline statistical significance of our finding that mortality increased with hospital volume (Table 3) in part is related to aggressive adjustment for illness severity measures (for example, laboratory values) and adjustments for travel distance to the nearest VA medical center; before these covariates were added, hospital volume was significantly associated with mortality.

In addition to hospital volume, we examined the effects of average patient composition of the hospital (for example, hospital rurality), a measure that, to our knowledge, has not been reported in models examining urban–rural disparities. We found that the average patient composition of hospitals was not itself statistically significant but modestly influenced the apparent association between patient rurality and mortality.

Our study has several limitations. The sample was derived from a mostly male population of inpatient veterans with numerous comorbid conditions. Its generalizability is therefore limited. Identification of our study sample relied on the accuracy of ICD-9-CM coding; some misspecification of rural residence may have occurred; and our models relied solely on administrative data sources and did not include important disease-specific physiologic variables, such as pulmonary function testing, that were used to adjust mortality in previous research.

However, administrative data sources can provide mortality adjustments that are similar to those derived from physiologic parameters that require more laborious chart abstraction (53), and we included laboratory data demonstrated to be adequate surrogate physiologic data in risk adjustment (5). We had no data on hospital resources associated with variations in COPD mortality (for example, pulmonologist resources) but believe that our measures of hospital volume and rurality probably serve as markers for resource availability.

Many current methods define rurality, and findings may not translate across all methods. To address this potential limitation, we included a measure of distance to the nearest VA medical center. Contrary to our expectations, longer travel distances were associated with lower 30-day mortality. A possible explanation for this effect is that veterans who live farther away from a VA medical center have to be well enough to tolerate traveling the longer distance for care.

The same explanation might account for why the apparent association between isolated rurality and mortality decreases and loses statistical significance without adjustment for travel time. Rural hospitals tended to admit patients who had to endure longer travel times and thus may also be healthier than other populations upon admission, whereas urban hospitals admitted patients who had shorter travel times. Adjustment for travel time increases the apparent risk for mortality, because patients who have similar travel times to a VA medical center and presumably have a common case mix tend to have lower mortality at urban hospitals.

In conclusion, we found that patients from isolated rural areas admitted for a COPD exacerbation have an increased risk for 30-day mortality independent of the characteristics of the admitting hospitals. Mortality for patients living in nonisolated rural areas did not differ from that for patients living in urban settings. We believe that workforce and policy leaders need to be aware of this difference as a first step toward more equitable distribution of resources shown to be associated with variations in mortality, such as board-certified pulmonary staff and ventilatory support resources (51). Perhaps more important, further research is needed to highlight which specific resources rural patients lack and to substantiate these findings by using spatial techniques to define rurality at a more granular level.