Author + information
- Received April 3, 2017
- Revision received May 15, 2017
- Accepted May 18, 2017
- Published online August 7, 2017.
- Tushar Acharya, MDa,
- Adam C. Salisbury, MD, MScb,c,
- John A. Spertus, MD, MPHb,c,
- Kevin F. Kennedy, MSb,c,
- Amarbir Bhullar, MDa,
- H. Kiran K. Reddy, MDa,
- Bipin K. Joshi, MDa and
- John A. Ambrose, MDa,∗ ()
- aUniversity of California, San Francisco, Fresno, California
- bSaint Luke’s Mid-America Heart Institute, Kansas City, Missouri
- cUniversity of Missouri-Kansas City, Kansas City, Missouri
- ↵∗Address for correspondence:
Dr. John A. Ambrose, UCSF Fresno, Division of Cardiology Academic Offices, 2335 East Kashian Lane, Suite 460, Fresno, California 93721.
Objectives This study compared risk-adjusted percutaneous coronary intervention (PCI) outcomes of safety-net hospitals (SNHs) and non-SNHs.
Background Although risk adjustment is used to compare hospitals, SNHs treat a disproportionate share of uninsured and underinsured patients, who may have unmeasured risk factors, limited health care access, and poorer outcomes than patients treated at non-SNHs.
Methods Using the National Cardiovascular Data Registry CathPCI Registry from 2009 to 2015, we analyzed 3,746,961 patients who underwent PCI at 282 SNHs (hospitals where ≥10% of PCI patients were uninsured) and 1,134 non-SNHs. The relationship between SNH status and risk-adjusted outcomes was assessed.
Results SNHs were more likely to be lower volume, rural hospitals located in the southern states. Patients treated at SNHs were younger (63 vs. 65 years), more often nonwhite (17% vs. 12%), smokers (33% vs. 26%), and more likely to be admitted through the emergency department (48% vs. 38%) and to have an ST-segment elevation myocardial infarction (20% vs. 14%) than non-SNHs (all p < 0.001). Patients undergoing PCI at SNHs had higher risk-adjusted in-hospital mortality (odds ratio: 1.23; 95% confidence interval: 1.17 to 1.32; p < 0.001), although the absolute risk difference between groups was small (0.4%). Risk-adjusted bleeding (odds ratio: 1.05; 95% confidence interval: 1.00 to 1.12; p = 0.062) and acute kidney injury rates (odds ratio: 1.01; 95% confidence interval: 0.96 to 1.07; p = 0.51) were similar.
Conclusions Despite treating a higher proportion of uninsured patients with more acute presentations, risk-adjusted PCI-related in-hospital mortality of SNHs is only marginally higher (4 additional deaths per 1,000 PCI cases) than non-SNHs, whereas risk-adjusted bleeding and acute kidney injury rates are comparable.
The complex fabric of American health care results in some hospitals, often called safety-net hospitals (SNHs), disproportionately treating a poor, more underserved group of patients than other hospitals. The Institute of Medicine defines SNHs as having 2 distinguishing characteristics: they maintain an “open door” policy, offering access to services regardless of the patient’s ability to pay; and they treat a disproportionate share of patients without health insurance and other vulnerable populations (1). As such, SNHs seek to overcome significant disparities in access to health insurance in the United States. More than 48.6 million people (16% of the U.S. population) had no health insurance in 2010 (2). Lack of health insurance is associated with less regular outpatient care, greater burden of cardiovascular risk factors, and a greater likelihood to present to the hospital with an acute medical emergency (3–7). Accordingly, studies have reported higher all-cause mortality among uninsured patients as compared with the age- and sex-matched insured patients (8,9). Given that SNHs care for this unique, and potentially higher risk population of uninsured patients in a potentially limited-resource setting, it is important to understand how outcomes of SNHs compare with non-SNHs.
Although regional studies have reported higher percutaneous coronary intervention (PCI)-associated mortality in underinsured patients (10,11), it is unclear whether these outcomes are caused by the patients’ risk factors, or are attributable to the care that they receive at SNHs. This is a complex issue to disentangle, but can have an important impact on public policy. To address the different populations treated at different hospitals, efforts to create risk-adjustment models have been undertaken by national organizations, such as the American College of Cardiology’s National Cardiovascular Data Registry (NCDR). The NCDR has created risk models for periprocedural outcomes after PCI to enable hospital comparisons. However, many of the unique features of patients treated at SNHs may not be incorporated into these models. Comparing the risk-adjusted outcomes of SNHs and non-SNHs can provide important insights into the validity of NCDR risk models in comparing different types of hospitals.
To address this gap in knowledge, we assessed key in-hospital PCI outcomes, including mortality, bleeding, and acute kidney injury (AKI), by comparing risk-adjusted outcomes for SNHs and non-SNHs using the NCDR CathPCI registry (12). Identifying substantial discordance in outcomes could indicate either substantially worse care at SNHs, or poor model performance because of important, unmeasured confounders.
Details of the NCDR CathPCI Registry have been described previously (12,13). Briefly, the registry prospectively collects data on patient characteristics, clinical features, angiographic and procedural details, and in-hospital outcomes of patients receiving diagnostic angiography or PCI at participating hospitals to facilitate comparative assessment and to affect guideline adherence and quality improvement. Participating institutions enter data locally using NCDR-certified software. Standard data definitions are used at all participating sites to ensure uniformity and comparability. The CathPCI data collection form includes information on demographics, comorbidities, medications, clinical presentation, clinical course, diagnostic testing, procedures (along with indications and multiple angiographic variables), complications, and outcomes. Forms submitted to the NCDR undergo auditing to ensure completeness and accuracy before being entered into the database (12).
A total of 1,585 U.S. hospitals reported data for 3,783,700 PCIs to the NCDR CathPCI registry between third quarter of 2009 and first quarter of 2015, the period of Version 3 of the NCDR data collection form. After excluding hospitals that reported <8 quarters of data and those performing <30 PCIs per year, 1,416 hospitals and 3,746,961 patients were included in the final analysis.
Definition of SNHS
There is no universally accepted definition for SNHs. The Agency of Healthcare Research and Quality has suggested that the uninsured may comprise more than 9% of SNH stays (14). Accordingly, for this study, we defined SNHs as those hospitals that perform 10% or more PCIs on uninsured patients. Using data from the NCDR CathPCI data collection form, which collects insurance payer status at the time of PCI (categories are Medicaid, Medicare, private health insurance, military health care, state-specific plan, Indian health service, non-U.S. insurance, or “none”), we considered SNHs as those hospitals for whom ≥10% of their patients had no health insurance.
Outcomes of interest
The NCDR has published risk models for periprocedural bleeding (15), AKI (16), and in-hospital mortality (17). For each outcome, we used the previously published definitions that were used to develop the risk models. Periprocedural bleeding was defined as any of the following events occurring within 72 h after PCI: arterial access site bleeding (either external or a hematoma >10 cm for femoral access, >5 cm for brachial access, or >2 cm for radial access); retroperitoneal, gastrointestinal, or genitourinary bleeding; intracranial hemorrhage; cardiac tamponade; post-procedure hemoglobin decrease of 3 g/dl in patients with a pre-procedure hemoglobin level ≤16 g/dl; or post-procedure nonbypass-surgery-related blood transfusion for patients with a pre-procedure hemoglobin level ≥8 g/dl. For AKI, the Acute Kidney Injury Network stage 1 or greater definition was used: a ≥0.3 mg/dl absolute or ≥1.5-fold relative increase in post-PCI creatinine or new initiation of dialysis.
Baseline hospital and patient characteristics of SNHs and non-SNHs were reported as proportions and groups were compared using chi-square test. Continuous variables were reported as mean ± SD and compared using Student t tests. Skewed data were reported as median (interquartile range) and were compared using Wilcoxon rank sum test.
Hierarchical multivariable logistic regression was used to assess the relationship between SNH status and risk-adjusted mortality, bleeding, and AKI. Published NCDR risk models for bleeding (15), AKI (16), and mortality analysis (17) were used to adjust for potential confounding. In all of these models, imputation of missing data was handled as defined by the published risk models (15–17). Strength of association was expressed by odds ratio (OR) with 95% confidence intervals (CI). Two sided p values of <0.05 were considered statistically significant. We then calculated predicted or expected outcomes from the previously mentioned regression models, and the observed outcome rate of SNHs and non-SNHs was divided by their predicted rate to obtain the observed/expected (O/E) outcome ratios (18). All analyses were performed with SAS version 9.4 (SAS Institute, Cary, North Carolina).
Because hospital volume is an important predictor of PCI-associated mortality, we performed subgroup analysis in low (<400 cases/year), intermediate (400 to 1,000 cases/year), and high (>1,000 cases/year) volume hospitals to determine if mortality difference between SNH and non-SNHs was a function of their volume status. Similarly, subgroup analysis was performed based on PCI indication (acute coronary syndrome [ACS] and non-ACS). Temporal trends were analyzed.
In the absence of a universally acceptable definition of SNHs, 2 sensitivity analyses were performed with PCI-performing hospitals stratified according to their Medicaid rate and Disproportionate Share Hospital Index. Details of these analyses are documented in the Online Appendix. Finally, we examined unadjusted in-hospital mortality rates among the uninsured versus insured patients.
A total of 282 SNHs performed 531,441 PCIs between the third quarter of 2009 and first quarter of 2015, and 3,215,520 PCIs were performed at 1,134 non-SNHs (Figure 1). SNHs were more likely to be rural hospitals and were twice as likely to be located in the southern U.S. states (66% vs. 31%). They had a significantly lower annual PCI volume (365 vs. 538 PCI cases/year) and lower operator volume (82 vs. 100 cases/year), compared with non-SNHs (Table 1). Patients treated at SNHs were younger (63 vs. 65 years) and more often nonwhite (17% vs. 12%). Their smoking rates were higher (33% vs. 26%). They were more frequently admitted for PCI through the emergency department (48% vs. 38%) rather than through outpatient referral. PCIs performed at SNHs were more likely to be emergent (24% vs. 18%) and less likely to be elective procedures (34% vs. 41%). They were also more frequently primary PCIs for ST-segment elevation myocardial infarction (20% vs. 14%) than for other indications (Table 2).
In unadjusted analyses, patients undergoing PCI at SNHs had a 0.4% higher absolute risk of mortality, 0.5% higher absolute risk of periprocedural bleeding, and 0.3% higher absolute risk of AKI related to the PCI procedure (Table 3). After adjusting for patients’ characteristics using the NCDR risk model, patients treated at SNHs had higher risk-adjusted in-hospital mortality (OR: 1.23; 95% CI: 1.17 to 1.32; p < 0.001), but similar risk-adjusted bleeding (OR: 1.05; 95% CI: 1.00 to 1.12; p = 0.62) and AKI (OR: 1.01; 95% CI: 0.96 to 1.07; p = 0.51). Similarly, patients treated at SNHs had higher O/E mortality risk (1.20 vs. 0.99; p < 0.001), but similar O/E bleeding (1.09 vs. 1.06; p = 0.157) and AKI (0.96 vs. 0.93; p = 0.259) rates (Figure 2).
Results of the subgroup analyses are detailed in Figure 3. There was modest attenuation of the in-hospital mortality difference between SNH and non-SNH when stratified by PCI volume. High-volume hospitals had a lower mortality difference (OR: 1.15; 95% CI: 0.95 to 1.41; p = 0.159) followed by intermediate-volume (OR: 1.16; 95% CI: 1.07 to 1.27; p < 0.001), and low-volume hospitals having the largest mortality difference (OR: 1.26; 95% CI: 1.16 to 1.37; p < 0.001). Both ACS (OR: 1.18; 95% CI: 1.11 to 1.25; p < 0.001) and non-ACS groups (OR: 1.26; 95% CI: 1.14 to 1.39; p < 0.001) had statistically higher PCI mortality when treated at SNHs. Subgroup analyses for bleeding and AKI are similarly shown in Figure 3.
Temporal trends comparing in-hospital mortality over the time period of this analysis are shown in Figure 4. There was no significant trend toward either harm or improvement from 2009 through 2015 and the O/E in-hospital mortality ratios of SNH and non-SNHs remained fairly unchanged.
To put the results of the primary analysis in perspective, PCI-associated in-hospital mortality for insured and uninsured patients were calculated from this data set. The unadjusted in-hospital mortality rate of the uninsured was 2.2% compared with 1.5% in the insured (p < 0.001).
In the absence of universal health care coverage, uninsured and underinsured patients in the United States depend on the medical safety-net for medical and procedural care. In this study, we found that despite treating a large proportion of uninsured patients with more acute presentations, key PCI-related outcome of in-hospital mortality was only marginally higher in SNHs (23% relative risk difference; 0.4% absolute risk difference) compared with non-SNHs, whereas periprocedural bleeding and AKI were similar. These findings were robust to sensitivity analyses, including stratification by hospitals’ PCI volume and presentation with either ACS or stable coronary artery disease.
Underinsurance is a risk factor for poor PCI-associated outcomes. Because SNHs treat more uninsured patients, it is important to evaluate how their outcomes differ from non-SNHs. In 2011, a study from the Washington Hospital Center examining a 10-year experience of PCI found that uninsured patients had a higher rate of risk-adjusted cardiac events (composite of death, Q-wave myocardial infarction, and target vessel revascularization) at 1 year when compared with patients with private insurance (10). Similarly, an analysis of a contemporary PCI cohort from 4 New York teaching hospitals showed an increased risk of in-hospital mortality among patients without insurance or with Medicaid (11). The causes of these differences in outcome are unclear. Some have suggested this may reflect differences in time to revascularization and revascularization strategies (e.g., use of drug-eluting stents) between those with and without insurance (19–23), or that these patients may be less likely to receive evidence-based medications for coronary artery disease (24). In the present study cohort, the PCI-associated in-hospital mortality for the uninsured was 2.2% compared with 1.5% in the insured (p < 0.001). We evaluated if these differences in outcomes at the patient level would translate to hospitals treating them. It has been shown that hospitals with different organizational characteristics and ownership status have different PCI outcomes (25). In the present study, we found that although there were significant differences in insured rates and key risk factors for outcomes (e.g., more acute presentations) clinical outcomes of SNHs and non-SNHs were only slightly dissimilar after standard risk adjustment.
There were important geographical and demographic differences between SNHs and non-SNHs. Consistent with prior reports that have noted large uninsured populations in southern U.S. states, we found that SNHs were more common in the south (26). We also found that SNHs treated more nonwhite patients with higher smoking rates. Despite baseline differences between groups, we found statistically higher rates of in-hospital mortality but no difference in bleeding and AKI rates. Even though the relative risk of PCI-associated mortality was 20% higher, the absolute risk difference between outcomes at SNH and non-SNHs difference was small (4 additional deaths in SNHs per 1,000 patients treated; number needed to harm = 250). Similarly, comparison of O/E ratios revealed a modest statistically higher mortality in the SNH group with no difference in bleeding and AKI.
The drivers of this small increase in mortality at SNHs may be a consequence of higher mortality in the uninsured population. We evaluated if clustering of more acutely ill patients at SNHs (11% more emergency department admissions; 6% more ST-segment elevation myocardial infarction presentations) may lead to higher mortality; but in subgroup analysis, mortality at SNHs was higher in both ACS and stable coronary artery disease groups treated at SNHs. Lower PCI volume of SNHs (365 vs. 538 PCIs per year) could also be a contributor to the excess mortality (27–37). On substratification of hospitals based on volume, we found that higher hospital volume does attenuate the mortality difference between SNHs and non-SNHs to some extent, although with larger CI (Figure 2). High-volume SNHs may be performing at par with non-SNHs.
Regardless, it is reassuring that despite multiple reasons for faring worse on important periprocedural outcomes, in-hospital mortality of patients undergoing PCI at SNHs was, in reality, only slightly worse than non-SNHs. This better-than-expected performance of SNHs may partly be explained by the fact that most patients treated at SNHs or non-SNHs were insured (Table 2). Thus, the actual contribution of uninsured patients to the overall outcome of a hospital may be small. Another possibility is that the definition used in this study (based on the lack of insurance) may not accurately represent SNHs. To test this possibility, we performed sensitivity analyses by stratifying hospitals by their share of Medicaid patients and by their Disproportionate Share Hospital Index (Figure 5). Regardless of the SNH definition used (based on uninsured rate, Medicaid rate, or Disproportionate Share Hospital Index), the outcomes did not vary in any significant manner.
Going forward, it is important to note that since the introduction of the Affordable Care Act, the percentage of uninsured patients in the United States has steadily declined (from 16.0% in 2010 to 9.2% in early 2015) (38). This changing landscape may narrow the current health care gap and could redefine the role of SNHs. However, recent events leading to the imminent repeal and/or replacement of the Affordable Care Act could add further flux to an already fluid field. In the absence of a foreseeable scenario of universal health care, it is perhaps safe to assume that SNHs will continue to play a vital role in providing much needed medical care to the underprivileged. Patients with “less desirable” insurance may still tend to cluster at SNHs, because other hospitals may be reluctant to accept them. It is also fairly commonplace that under-resourced patients treated at SNHs are less mobile and are likely to present to SNHs near their home regardless of insurance status.
Because of its observational nature, our findings are subject to unmeasured and residual confounding. As with all registries, the accuracy of documentation and submission of data is dependent on participating hospitals, although there is no evidence of variation in the quality of data capture between SNHs and non-SNHs. The possibility of a revascularization selection bias (surgery vs. PCI vs. medical therapy) between comparisons cannot be ruled out. The NCDR CathPCI registry does not collect data elements on degree of risk factor control at baseline, socioeconomic status, and access to health care. These potentially important determinants of outcome were not part of the study. Results of this study may, perhaps, not apply to a small minority of hospitals not participating in the registry. However, Cath PCI registry collects data from most, if not all of the U.S. hospitals offering PCI to patients. Because of this large representation, these results should be generalizable. Finally, this study only evaluated in-hospital outcomes and it is conceivable that long-term outcomes of patients getting PCI at SNHs may differ.
Despite treating large numbers of uninsured patients with more acute presentations, risk-adjusted PCI-related in-hospital mortality of SNHs is only marginally higher than non-SNHs, with no difference in periprocedural bleeding and AKI rates. This suggests that current risk models can serve as a foundation for comparing SNHs and non-SNHs. It is perhaps even reassuring that America’s PCI SNHs, despite treating a complex and acutely ill population, are able to achieve satisfactory if not similar outcomes as non-SNHs. Whether obtaining health insurance will lead to better health care access and translate into improvement of outcomes including procedural mortality remains unclear and requires further study.
WHAT IS KNOWN? Uninsured patients, because of unmeasured risk factors and limited health care access, have poorer PCI-associated outcomes than the insured. SNHs treat a disproportionate share of uninsured and underinsured patients. PCI-associated in-hospital outcomes of SNHs are not known.
WHAT IS NEW? Despite treating a higher proportion of uninsured patients with more acute presentations, risk-adjusted PCI-related in-hospital mortality of SNHs is only marginally higher (4 additional deaths per 1,000 PCI cases) than non-SNHs, whereas risk-adjusted bleeding and acute kidney injury rates are comparable. This difference in mortality has been stable from 2009 through 2015.
WHAT IS NEXT? Whether obtaining health insurance leads to better health care access and translates into improvement of outcomes including procedural mortality requires further study.
For a supplemental appendix, please see the online version of this article.
Dr. Salisbury has received research grant support from Boston Scientific; and speaking fees/honoraria from Abiomed. Dr. Spertus has an analytic contract from the American College of Cardiology Foundation to analyze the National Cardiovascular Data Registry. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- acute coronary syndrome
- acute kidney injury
- confidence interval
- National Cardiovascular Data Registry
- odds ratio
- percutaneous coronary intervention
- safety-net hospitals
- Received April 3, 2017.
- Revision received May 15, 2017.
- Accepted May 18, 2017.
- 2017 American College of Cardiology Foundation
- National Research Council
- ↵Cohen RA, Ward BW, Schiller JS. Health insurance coverage: early release of estimates from the National Health Interview Survey, 2010. National Center for Health Statistics. June 2011. Available at: http://www.cdc.gov/nchs/nhis.htm. Accessed February 12, 2016.
- Institute of Medicine Committee on the Consequences of Uninsurance
- Institute of Medicine Committee on the Consequences of Uninsurance
- Brindis R.G.,
- Fitzgerald S.,
- Anderson H.V.,
- Shaw R.E.,
- Weintraub W.S.,
- Williams J.F.
- Anderson H.V.,
- Shaw R.E.,
- Brindis R.G.,
- et al.
- Andrews R.M.,
- Stull D.E.,
- Fraser I.,
- Friedman B.,
- Houchens R.L.
- Rao S.V.,
- McCoy L.A.,
- Spertus J.A.,
- Krone R.J.,
- Singh M.,
- Fitzgerald S.,
- Peterson E.D.
- Tsai T.T.,
- Patel U.D.,
- Chang T.I.,
- et al.
- Peterson E.D.,
- Dai D.,
- DeLong E.R.,
- et al.
- Klein L.W.,
- Kolm P.,
- Xu X.,
- et al.
- Sada M.J.,
- French W.J.,
- Carlisle D.M.,
- Chandra N.C.,
- Gore J.M.,
- Rogers W.J.
- Gaglia M.A. Jr..,
- Torguson R.,
- Xue Z.,
- et al.
- Smolderen K.G.,
- Spertus J.A.,
- Tang F.,
- Oetgen W.,
- Borden W.B.,
- Ting H.H.,
- Chan P.S.
- US Census Bureau, current population reports
- Magid D.J.,
- Calonge B.N.,
- Rumsfeld J.S.,
- Canto J.G.,
- Frederick P.D.,
- Every N.R.,
- Barron H.V.,
- National Registry of Myocardial Infarction 2 and 3 Investigators
- Vakili B.A.,
- Kaplan R.,
- Brown D.L.
- Epstein A.J.,
- Rathore S.S.,
- Volpp K.G.,
- Krumholz H.M.
- Hannan E.L.,
- Wu C.,
- Walford G.,
- et al.
- Khattab A.A.,
- Hamm C.W.,
- Senges J.,
- et al.
- Srinivas V.S.,
- Hailpern S.M.,
- Koss E.,
- Monrad E.S.,
- Alderman M.H.
- Badheka A.O.,
- Patel N.J.,
- Grover P.,
- et al.
- ↵Health Insurance Coverage: Early Release of Estimates From the National Health Interview Survey, January–March 2015. Available at: http://www.cdc.gov/nchs/nhis.htm. Accessed February 12, 2016.