Mortality Rates for Medicare Beneficiaries Admitted to Critical Access and Non–Critical Access Hospitals, 2002-2010

Importance Critical access hospitals (CAHs) provide inpatient care to Americans living in rural communities. These hospitals are at high risk of falling behind with respect to quality improvement, owing to their limited resources and vulnerable patient populations. How they have fared on patient outcomes during the past decade is unknown.

Objective To evaluate trends in mortality for patients receiving care at CAHs and compare these trends with those for patients receiving care at non-CAHs.

Design, Setting, and Patients Retrospective observational study using data from Medicare fee-for-service patients admitted to US acute care hospitals with acute myocardial infarction (1 902 586 admissions), congestive heart failure (4 488 269 admissions), and pneumonia (3 891 074 admissions) between 2002 and 2010.

Main Outcome Measures Trends in risk-adjusted 30-day mortality rates for CAHs and other acute care US hospitals.

Results Accounting for differences in patient, hospital, and community characteristics, CAHs had mortality rates comparable with those of non-CAHs in 2002 (composite mortality across all 3 conditions, 12.8% vs 13.0%; difference, −0.3% [95% CI, −0.7% to 0.2%]; P = .25). Between 2002 and 2010, mortality rates increased 0.1% per year in CAHs but decreased 0.2% per year in non-CAHs, for an annual difference in change of 0.3% (95% CI, 0.2% to 0.3%; P < .001). Thus, by 2010, CAHs had higher mortality rates compared with non-CAHs (13.3% vs 11.4%; difference, 1.8% [95% CI, 1.4% to 2.2%]; P < .001). The patterns were similar when each individual condition was examined separately. Comparing CAHs with other small, rural hospitals, similar patterns were found.

Conclusions and Relevance Among Medicare beneficiaries with acute myocardial infarction, congestive heart failure, or pneumonia, 30-day mortality rates for those admitted to CAHs, compared with those admitted to other acute care hospitals, increased from 2002 to 2010. New efforts may be needed to help CAHs improve.

More than 60 million Americans live in rural areas and face challenges in accessing high-quality inpatient care. In 1997, the US Congress created the Critical Access Hospital (CAH) program¹ in response to increasing rural hospital closures. To qualify, hospitals must have no more than 25 beds and be located at least 35 miles from the nearest alternative source of inpatient care; however, states were given leeway to broaden eligibility, and only 20% of CAHs currently meet this distance requirement.² Under the program, CAHs were exempted from prospective payments and instead receive reimbursement at 101% of costs.¹ Additionally, the federal government exempted CAHs from participation in national quality improvement programs.^3-6

Since the inception of the program, hospitals electing CAH status have seen additional payments from Medicare,⁷ improvement in margins compared with nonconverting rural hospitals,⁸ and few closures.^7,9 Hundreds of hospitals have joined the program over the past decade—by 2010, nearly 1 in 4 of the nation's hospitals were CAHs.

Despite these additional financial supports from the government, CAHs still face significant challenges compared with larger, less isolated facilities. Critical access hospitals have fewer financial and human capital resources¹⁰ and care for a rapidly aging population at high risk of poverty and joblessness.¹¹ Whether current federal efforts, which create unique quality improvement programs targeted toward these hospitals, are working is unclear. Recent cross-sectional analyses found that patients at CAHs had worse outcomes than those at non-CAHs,¹⁰ although such analyses can overlook important improvements over time. Data on patient outcomes at CAHs during the past decade would be helpful in determining if the current federal strategy has been effective or if new initiatives are needed to help these vulnerable hospitals improve the care they provide.

Therefore, in this study, we focused on 3 questions. First, did mortality rates at CAHs improve over the past decade, and how do these changes compare with those at non-CAHs? Second, given that CAHs face unique challenges in providing care in rural areas, how have they fared compared with other small, rural hospitals not in the CAH program? Third, are identifiable hospital characteristics or resources associated with improvement in outcomes for CAHs?

This study was approved by the Office of Human Research Administration at the Harvard School of Public Health. The requirement for informed consent was waived because the study used deidentified data.

Hospitals. Medicare Provider Analysis and Review files from 2002 through 2010 were used to identify nonfederal hospitals providing acute care services to Medicare beneficiaries in the 50 US states or District of Columbia. The 2002 and 2010 American Hospital Association surveys were used to obtain hospital characteristics, including critical access designation, size, ownership, teaching status, clinical resources, and region. These data were linked with the 2010 Area Resource File, which contains county-level data on median household income and poverty rate. The Rural Urban Commuting Area codes, which describe population density and urbanization, were used to define hospitals' rurality,¹² because although in the original legislation only isolated rural hospitals qualified for CAH status, states subsequently allowed hospitals in suburban or urban settings to be eligible.

Patients. The study population comprised Medicare fee-for-service beneficiaries in 2002 through 2010 with a primary discharge diagnosis of acute myocardial infarction (AMI), congestive heart failure (CHF), or pneumonia (International Classification of Diseases, Ninth Revision, Clinical Modification codes 410.xx [excluding 410.x2] for AMI; codes 398.91, 404.x1, 404.x3, and 428.0 to 428.9 for CHF; and codes 480 to 486 for pneumonia). The Center for Medicare & Medicaid Services (CMS) approach for classifying “index admissions”¹³ was used, allowing patients to be included in the sample more than once. Also based on the CMS approach, all patients were assigned to the admitting hospital regardless of whether they were transferred.

Outcomes. Patient-level Medicare data were used to determine mortality within 30 days of admission. Unadjusted mortality rates were calculated, and patient-level hierarchical logistic regression models were created for each quarter, accounting for clustering of patients within hospitals. Models included a random effect for hospital and were adjusted for age, sex, and 29 comorbid medical conditions using the Medicare risk-adjustment model developed by the Agency for Healthcare Research and Quality (eTable 1).¹⁴ An adjusted mortality rate was calculated for each hospital for each quarter for each condition, and indirect standardization was used to create composite mortality rates across the 3 conditions.

Summary statistics were calculated for hospital characteristics, demographics, and patient comorbid conditions for CAHs and non-CAHs in 2002 and 2010. Weighted hospital-level longitudinal regression models were created, with risk-adjusted mortality rate as the outcome. The models were used to calculate the change in mortality for each condition over time, and an interaction term between time and CAH status was added to determine if change over time differed for CAHs vs non-CAHs. Multivariate regression models adding hospital characteristics were then built. Last, local poverty, local physician supply, and rurality (divided into urban, large town, small town, and rural categories) were added to the models. Rurality was particularly important: although rurality is highly collinear with being a CAH, it may also be correlated with other, unmeasured (or inadequately measured) factors, including travel time and availability of clinical personnel and resources.

For patients with AMI, rates of percutaneous coronary intervention (PCI) were added to the models. Given that PCI is generally not available at CAHs, we allowed PCIs to occur at any time within 30 days of the index admission, at any hospital, as has been done previously.¹⁵

The fully adjusted models are the primary analytic results presented because of the differences in hospital characteristics between CAHs and non-CAHs and the likelihood that those differences confound the relationship between CAH status and outcomes. However, models not adjusting for hospital characteristics are presented in the eTables.

To determine whether CAHs had outcomes different from those of other small, rural hospitals without the CAH designation, 2 sets of analyses were conducted. In the first, each CAH was matched to at least 1 non-CAH based on size, rurality, teaching status, and region. Further, because matched control hospitals were somewhat larger than CAHs, the number of hospital beds was added to the models. In the second analysis, to examine whether joining the CAH program was associated with a change in mortality rates over time, a pre-post analysis was conducted among hospitals that switched to CAH status during the study period. Hospital characteristics were compared between hospitals that switched and a matched group that did not. Mortality prior to the switch was then compared with mortality following the switch, dropping the transition year, and pre-post differences were compared between hospitals that switched and controls.

Last, within the CAH group, hospital characteristics were compared between CAHs that improved over the study period and those that did not, based on average annual change in risk-adjusted composite mortality between 2002 and 2010. Alternative definitions of hospitals that improved, including those in the top quartile of change over the past decade and performing better than the median non-CAH performance, yielded similar results; thus, findings based on the definition above are presented.

Several sensitivity analyses were conducted. First, because we were concerned that the mortality differences between CAHs and non-CAHs might be attributable to differential coding of comorbid conditions, we estimated the effects of varying the prevalence of patients with important comorbid conditions such as chronic kidney disease (CKD) on our findings. Second, based on data from Lindenauer et al¹⁶ suggesting that hospitals increasingly recode their sickest patients with pneumonia as having sepsis, hospitalizations with a primary diagnosis of sepsis or respiratory failure with a secondary diagnosis of pneumonia were also examined. Third, because our findings could have been affected by changes in our sample over time because many hospitals left the regular Medicare payment program and became CAHs during the study period, analyses were performed comparing only hospitals that had the CAH designation as of 2002 with those hospitals with no CAH designation at any point between 2002 and 2010. Fourth, because of concerns that changes in Medicare Advantage penetration may have been different in rural vs urban areas, analyses were reperformed controlling for the proportion of beneficiaries enrolled in Medicare Advantage in each county.

We considered P < .05 (2-sided) to be significant. Analyses were performed using SAS version 9.2 (SAS Institute Inc) and Stata 12.1 (StataCorp).

Of the 3968 US hospitals providing acute care services to Medicare beneficiaries in 2002 and reporting hospital characteristics to the American Hospital Association, 860 (22%) were designated as CAHs; by 2010, this had increased to 1264 of 4519 (28%). Between 2002 and 2010, there was an increase in the proportion of CAHs located in urban areas (0.8% in 2002 vs 1.2% in 2010), suburban areas (5.0% vs 6.0%), and large rural towns (5.6% vs 9.1%) and a decrease in the proportion in small towns or isolated rural areas (88.6% in 2002 vs 83.7% in 2010) (Table 1). The characteristics of non-CAHs were relatively unchanged between 2002 and 2010.

Between 2002 and 2010, the population of patients cared for at CAHs increased in age and in prevalence of diabetes, hypertension, CKD, and chronic pulmonary disease. The proportion of Medicaid-eligible patients increased slightly from 2002 to 2010, as did the proportion discharged to skilled nursing or rehabilitation facilities or hospice; the proportion of patients transferred decreased (Table 2). The population of patients cared for at non-CAHs underwent many of the same changes: the prevalence of comorbid conditions increased, as did discharges to skilled nursing or rehabilitation facilities and hospice; the proportion of patients transferred decreased (Table 2).

There were differences in the trends in 30-day mortality rates over time between CAHs and non-CAHs for the 10 281 929 admissions across the 3 conditions we examined (Figure).

When a composite across the 3 conditions was formally tested, adjusting for teaching status, ownership, region, rurality, poverty, and local physician supply, composite baseline mortality was similar between CAHs and non-CAHs (12.8% vs 13.0%; difference, −0.3% [95% CI, −0.7% to 0.2%]; P = .25) (Table 3). However, between 2002 and 2010, mortality rates increased at CAHs at a rate of 0.1% per year, whereas at non-CAHs they decreased 0.2% per year, for a difference in change in mortality of 0.3% per year (95% CI, 0.2% to 0.3%; P < .001). Thus, by 2010, CAHs had higher overall mortality rates (13.3% vs 11.4%; difference, 1.8% [95% CI, 1.4% to 2.2%]; P < .001). In total, CAH admissions were associated with 10.4 excess deaths per 1000 admissions during the study period.

Patterns were similar for each of the 3 conditions individually. In 2002, adjusting for hospital characteristics, CAHs had mortality rates similar to those of non-CAHs for AMI (15.4% vs 17.2%, P = .06), CHF (10.7% vs 10.6%, P = .76), and pneumonia (13.2% vs 13.6%, P = .11) (Table 3). However, between 2002 and 2010, although AMI mortality decreased to 14.9% in non-CAHs, it increased to 19.3% in CAHs, leading to a gap of 4.4 absolute percentage points by the end of the decade (95% CI, 2.7% to 6.1%; P < .001; 24.8 excess deaths per 1000 admissions). When rates of PCI were added to the mortality models, the gap was somewhat attenuated but remained significant (mortality, 19.0% vs 15.0%; difference, 4.0% [95% CI, 2.4% to 5.7%]; P < .001; 21.2 excess deaths per 1000 admissions). Findings for CHF and pneumonia were comparable, although the differences were smaller (CHF: absolute difference, 2.3% [95% CI, 1.7% to 2.9%]; P < .001; 16.0 excess deaths per 1000 admissions; pneumonia: absolute difference, 1.7% [95% CI, 1.3% to 2.2%]; P < .001; 8.3 excess deaths per 1000 admissions) (Table 3). Findings were similar for unadjusted mortality and in models adjusting for patient characteristics only (eTable 2 and eTable 3).

The C statistic for the 2010 AMI model was 0.709 (adjusted pseudo- R² = 0.12); for the CHF model, 0.673 (adjusted pseudo- R² = 0.07); and for the pneumonia model, 0.711 (adjusted pseudo- R² = 0.11).^17-19

An analysis matching hospitals on size, rurality, teaching status, and region, additionally controlling for the number of beds at each hospital, yielded similar results. Mortality was higher in this sample compared with the national average, and CAHs had slightly higher baseline mortality rates than their non-CAH matches (13.3% vs 13.0%; difference, 0.4% [95% CI, 0.1% to 0.6%]; P = .008). However, composite mortality worsened over time at CAHs but improved at non-CAHs; thus, by 2010 the gap had increased (14.0% vs 12.6%; difference, 1.5% [95% CI, 1.2% to 1.7%]; P < .001). In 2010, CAHs had higher mortality rates than matched non-CAHs for AMI (22.3% vs 18.8%; difference, 3.5% [95% CI, 2.7% to 4.4%]; P < .001), CHF (14.0% vs 12.2%; difference, 1.8% [95% CI, 1.5% to 2.1%]; P < .001), and pneumonia (13.1% vs 11.7%; difference, 1.4% [95% CI, 1.1% to 1.7%]; P < .001) (Table 4). Hospitals that switched from non-CAH to CAH status during the study period were somewhat more often small, Midwestern hospitals and were located in areas with higher generalist physician supply but lower specialist supply ( eTable 4). Comparing mortality prior to a switch to CAH status with mortality after a switch to CAH status, CAHs had higher mortality both at baseline and after the switch than their matched controls (eTable 5).

Between 2002 and 2010, 414 CAHs (48%) demonstrated improvement, whereas 443 did not; in the non-CAH group, 2114 hospitals (68%) improved over the same period. There were no significant differences between CAHs that improved and those that did not in hospital characteristics including size, region, ownership, teaching status, and rurality (Table 5). Those that improved were located in areas with similar physician supply but a slightly higher median income.

Although many major comorbid conditions were more prevalent at CAHs compared with non-CAHs, several comorbid conditions, especially CKD, were present less often at CAHs. In sensitivity analyses, we examined the effects of adjusting for this and generally found that such an adjustment led to a slight attenuation of the differences. For example, the difference in mortality for CHF in 2010 decreased from 1.7% to 1.5% after accounting for the different prevalence of CKD; we estimated that the true prevalence of CKD would have to be 35% higher in CAHs vs non-CAHs (ie, 65.5% vs 30.5%) to explain the entirety of the mortality difference we found. Patterns were similar for AMI and pneumonia.

Examining mortality rates for our combined diagnostic category of pneumonia plus sepsis or respiratory failure with a secondary diagnosis of pneumonia led to a slight attenuation of the difference between CAHs and non-CAHs, but the gap was still statistically significant (eTable 6). Limiting the hospital sample to only those CAHs that carried this designation throughout the study period did not alter the findings (eTable 7), nor did controlling for Medicare Advantage penetration (eTable 8).

Critical access hospitals had mortality rates similar to those of non-CAHs in 2002. However, by 2010, CAHs had higher mortality rates for each of the conditions examined, although the absolute difference was only 1.8%.

There are several possible explanations for why mortality rates have worsened at CAHs compared with other hospitals. First, although most hospitals in the country, including other small rural hospitals, participate in federal efforts to collect and publicly report performance data, CAHs are exempt. One challenge is that smaller sample sizes make interpreting CAHs' performance more difficult. However, simply participating in quality improvement may provide important feedback to hospital leadership about the care their hospital provides,^4,20 although evidence for this is mixed.²¹ Alternatively, it is possible that the CAH payment mechanism is associated with a lack of improvement, because cost-based reimbursement may remove incentives to pursue efficiency; prior work shows that CAHs are less efficient than non-CAH rural hospitals, a condition that worsens the longer a hospital is in the CAH program.²² Our finding that there were no hospital characteristics significantly associated with improvement within the CAH group similarly suggests that factors other than the structural characteristics are likely associated with improvements in outcomes. Features such as the degree to which hospital leadership prioritizes quality or the involvement of the hospital in local quality initiatives may explain why some CAHs improved whereas others did not.

Another possibility is that CAHs have not kept pace with non-CAHs because of the changing nature of hospital care and the inherent limitations CAHs face in keeping up with new technologies. Clinical care has become increasingly dependent on high-technology tools, especially for conditions such as AMI.²³ This has led policy makers to call for greater regionalization of care, given that optimal treatment requires access to imaging and interventional facilities not available at all hospitals. However, even for conditions less dependent on advanced technologies, such as pneumonia or CHF, mortality in CAHs worsened. This perhaps suggests that reasons beyond lack of availability of advanced technologies play a role. It is also possible that hospitals that chose to become CAHs had greater challenges at baseline than hospitals that did not elect this designation; prior research has shown that hospitals choosing to convert to CAH status have fewer beds and lower operating margins than those that do not.²⁴ However, our results suggest that current efforts have been inadequate in helping these hospitals catch up to other small, rural institutions.

Unmeasured differences in the patient population served at CAHs vs non-CAHs, such as a higher burden of social issues like poverty and unemployment, may underlie the differences in the observed outcomes. Our models explained only a small amount of the variation in the mortality rates. Constraints on care in isolated rural areas can be substantial, and our findings suggest that the supports contained in the CAH program have not been adequate to help these hospitals overcome the challenges imposed by caring for this vulnerable patient population in remote settings. Given that the hospitalized patient population has been getting sicker over time at CAHs, the finding that the proportion of patients transferred from CAHs to other facilities actually decreased during the study period was unexpected. One clinical factor that thus may be important is appropriate triage of patients who require transfer, which we were not able to assess. New interventions, such as close partnerships with larger institutions, use of technologies such as teleconsultation, or programs that help clinicians at CAHs determine which patients may need a higher level of care, may provide benefit for patients at these hospitals.

We are unaware of any previous work that has examined how clinical outcomes have changed over time at CAHs. A prior study showed that in 2009, CAHs had worse outcomes than non-CAHs, although the study did not examine whether gaps were narrowing or widening over time.¹⁰ One study of CAHs in Iowa found that performance on patient safety indicators improved after conversion to CAH status, although there was no comparison group.²⁵ Other studies of rural health care have predominantly used cross-sectional analyses and have generally but not consistently found worse outcomes.^26-29

This study has limitations. First, mortality was the only outcome examined; although this is a key, patient-centered metric, it does not capture the entirety of care delivered in hospitals. Administrative data were used for risk adjustment, which are limited in their ability to account for some patient characteristics that could affect outcomes, such as smoking status or body mass index, as well as socioeconomic characteristics, such as income or literacy. Administrative data are also unable to account for differences in other treatment factors that may affect mortality rates, such as travel times, availability of postacute care, or follow-up with an appropriate specialist. Although there have been important demographic changes in rural areas during the past decade and differences between urban and rural areas in terms of availability of care, the finding that outcomes worsened at CAHs, even compared with small, rural non-CAHs, should offer some assurance that rurality alone is unlikely to explain these results. However, these areas warrant further study.

Among Medicare beneficiaries with AMI, CHF, or pneumonia, 30-day mortality rates for those admitted to CAHs, compared with those admitted to other acute care hospitals, increased from 2002 to 2010. Given the substantial challenges that CAHs face, new policy initiatives may be needed to help these hospitals provide care for US residents living in rural areas.

Corresponding Author: Karen E. Joynt, MD, MPH, Department of Medicine, Brigham and Women's Hospital, 75 Francis St, Boston, MA 02115 (kjoynt@partners.org).

Author Contributions: Dr Joynt had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Jha, Joynt.

Acquisition of data: Jha, Joynt.

Analysis and interpretation of data: All authors.

Drafting of the manuscript: Joynt.

Critical revision of the manuscript for important intellectual content: Orav, Joynt, Jha.

Statistical analysis: Orav.

Obtained funding: Joynt.

Study supervision: Jha.

Funding/Support: Dr Joynt was supported by grant 1K23HL109177-01 from the National Heart, Lung, and Blood Institute, National Institutes of Health.

Role of Sponsor: The National Heart, Lung, and Blood Institute had no role in the design and conduct of the study; the collection, management, analysis, and interpretation of the data; or the preparation, review, or approval of the manuscript.