Author + information
- Received August 28, 2015
- Revision received October 7, 2015
- Accepted October 8, 2015
- Published online February 9, 2016.
- Alexander Papolos, MD,
- Jagat Narula, MD, PhD,
- Chirag Bavishi, MD, MPH,
- Farooq A. Chaudhry, MD and
- Partho P. Sengupta, MD, DM∗ ()
- ↵∗Reprint requests and correspondence:
Dr. Partho P. Sengupta, Mount Sinai’s Zena and Michael A. Wiener Cardiovascular Institute, Marie-Josée and Henry R. Kravis Center for Cardiovascular Health, One Gustave L. Levy Place, New York, New York 10029.
Background Increased use of echocardiography (echo) raises questions of whether echo is an overused diagnostic procedure in the United States.
Objectives This study investigated national trends, practice patterns, and patient outcomes associated with inpatient echo use reported in the Nationwide Inpatient Sample (NIS).
Methods We identified admission diagnoses most commonly associated with echo use and performed multivariate logistic regression within each diagnosis cohort to assess whether echo use was associated with all-cause inpatient mortality. Secondary analysis was performed within our institution to validate use trends identified in the NIS database.
Results Between 2001 and 2011, the absolute volume and incidence of echo steadily increased at average annual rates of 3.41% and 3.04%, respectively. In 2010, the use of echo was associated with lower odds of inpatient mortality among hospitalizations for acute myocardial infarction (adjusted odds ratio [OR]: 0.74; 95% confidence interval [CI]: 0.63 to 0.86; p < 0.001), cardiac dysrhythmia (adjusted OR: 0.72; 95% CI: 0.55 to 0.94; p = 0.02), acute cerebrovascular disease (adjusted OR: 0.36; 95% CI: 0.31 to 0.42; p < 0.001), congestive heart failure (adjusted OR: 0.82; 95% CI: 0.72 to 0.94; p = 0.005), and sepsis (adjusted OR: 0.77; 95% CI: 0.70 to 0.85; p < 0.001). In 2010, these 5 diagnoses accounted for 3.7 million hospital admissions (9% of all hospitalizations); however, echo was reported in only 8% of cases. Secondary analysis of imaging practices at our institution confirmed underuse of echo among patients who died during hospitalization for indications identified in the NIS database.
Conclusions Despite increasing rates of performance, echo may be underused during critical cardiovascular hospitalizations.
Echocardiography (echo) is the most ubiquitous, versatile, and cost-effective cardiac imaging modality available (1–5). These attributes, in combination with an infinitesimal risk-to-benefit ratio, explain why echo has become a mainstay of the cardiovascular evaluation. Widespread reliance on echo has fostered growing concern for its overuse and has led to the formation of appropriate use criteria (1). Although several studies have demonstrated adherence to these guidelines in clinical practice (2–4), national trends in the use of echo have not been fully described. The closest approximations are from Medicare literature, which showed that the use of echo (transthoracic echocardiography [TTE] and transesophageal echocardiography [TEE]) increased by 7.7% per year from 1999 to 2004 (5), and it nearly doubled from 1999 to 2008 (6). Reports from the Veterans Healthcare Administration also demonstrated an annual increase in total echo volume of 3.9% from 2000 to 2007, but this finding was primarily driven by population expansion rather than by a change in imaging practices (7). Given the paucity of data, this study was conducted to describe trends, practice patterns, and patient outcomes associated with the use of echo through a comprehensive assessment of hospitalizations available in the Nationwide Inpatient Sample (NIS) database (8).
We first queried the NIS to quantify temporal trends in inpatient echo use between 2001 and 2011. Next we explored the 2010 database to investigate whether echo use was associated with all-cause in-hospital mortality among the diagnoses for which echo is most commonly performed.
The NIS is the largest, publicly available, all-payer inpatient database in the United States (8). Annually, the NIS is composed of discharge-level data from roughly 8 million hospitalizations and approximates a stratified sample of 20% of community hospitals in the United States. The sampling methodology of the NIS permits the application of weighting variables that allow for the calculation of national estimates, which have been validated against other U.S. hospital registries (9). Each hospitalization within the database contains clinical and resource-use information. Patients’ diagnoses are documented in parallel, as both International Classification of Disease-9th edition-Clinical Modification (ICD-9-CM) and clinically meaningful clusters of ICD-9-CM’s, termed Clinical Classification Software (CCS) codes.
Temporal trends in echo use
The Healthcare Cost and Utilization Project online NIS registry (HCUPnet) (10) was queried by the ICD-9-CM for echo (procedure code 88.72). This provided national estimates of hospitalizations in which echo was performed from 2001 through 2011. The ICD-9-CM 88.72 is used for coding of all forms of echo. Therefore the database cannot differentiate admissions in which TTE examinations or other forms of echo such as TEE were performed. However, this limitation is likely to be insignificant given that the approximate ratio of TTE use to TEE use is close to 100:1 (6), as well as the clinical insight that few patients receive TEE without first undergoing TTE.
The 2010 NIS was used to evaluate for an association between echo use and diagnosis specific all-cause hospital mortality. The dataset was first confined to admission diagnoses associated with the greatest proportion of echo use (representing one-half of all estimated inpatient echo studies in 2010) (10). This included acute myocardial infarction (AMI), cardiac dysrhythmia, acute cerebrovascular disease, congestive heart failure (CHF), coronary artery disease (CAD), sepsis, valvular disease, and nonspecific chest pain (CCS diagnoses codes: 2, 96, 100, 102, 101, 106, 108, and 109). We subsequently applied 3 criteria to the sample: 1) non–emergency department admissions were excluded to control for confounding hospital-to-hospital transfers; 2) hospitalizations lasting <24 h were excluded to counter bias created by critically ill patients in whom a fatal outcome occurred without sufficient lead time to obtain an echo examination if clinically indicated; and 3) neonatal patients were excluded. The final sample included an estimated 3,651,503 hospitalizations and contained 31% of all estimated inpatient echo examinations performed in 2010. State-specific rates of inpatient echo use were derived by dividing estimated statewide echo volume by the corresponding volume of estimated hospitalizations, as recorded in the 2010 NIS.
Patient-related and hospital characteristics
To assess for association between echo performance and diagnosis-specific hospital mortality rates, we stratified the sample by primary CCS diagnosis code (admission diagnosis). Available independent variables for analyses included the patient’s age, race, sex, insurance status, number of comorbidities, median household income by zip code, illness severity score, and admission day (weekend vs. weekday). Information on race was categorized as white, black, Hispanic, and other (Asian, Pacific Islander, Native American, and unknown). Insurance status was classified based on the expected primary payer and included Medicare, Medicaid, private insurance, and other insurance types, including patients who were uninsured. The number of comorbid conditions was a sum of all listed diagnoses. Median household income per patient’s zip code was divided into income quartiles defined as follows: 1) $1 to $40,999; 2) $41,000 to $50,999; 3) $51,000 to 66,999; and 4) ≥$67,000. Illness severity was based on All Patient Refined–Diagnosis Related Group (APR-DRG) severity score, a validated clinically coherent set of illness severity and risk of mortality subclasses that adjust for the interactions between a patient’s primary diagnosis, comorbidities, age, and procedures occurring during hospitalization, defined as: 1) minor; 2) moderate; 3) major; and 4) extreme (11).
The primary endpoint was in-hospital mortality and was analyzed using multivariate regression, as described later.
Temporal trends in the incidence of national echo use per hospitalization between 2001 and 2011 were obtained by dividing the estimated annual volume of studies performed by the estimated total number of annual hospitalizations. Use growth (absolute and per hospitalization) was analyzed by calculating the average annual percentage change from 2001 to 2011.
The 2010 sample was dichotomized into hospitalizations in which echo was or was not performed, and descriptive statistics were generated on frequencies and proportions for categorical variables (age, sex, race, insurance, income quartile by zip code, number of chronic diseases, administration of diagnostic cardiac catheterization, administration of cardiac stress test [exercise and pharmacologic], admission day, and APR-DRG severity score). Means with standard error were reported for continuously coded variables (age and number of chronic diseases). Student t tests and the chi-square tests were used to evaluate demographic features among groups. To assess for an association between echo performance and diagnosis-specific hospital mortality rates, we stratified the sample by admission diagnosis (primary CCS diagnosis code). Within each subpopulation multivariate logistic regression was conducted on weighted numbers and was adjusted for patients’ age, sex, race, insurance, income quartile by zip code, number of chronic diseases, administration of diagnostic cardiac catheterization, administration of cardiac stress test, admission day, and APR-DRG severity score. Diagnostic catheterization was adjusted for, despite no significant difference observed between the sample arms, because it could potentially portend a mortality bias.
Single-center validation analysis
To validate our analysis of the NIS we evaluated all hospitalizations that occurred during the calendar year 2014 within our home institution, a large tertiary academic medical center in New York City. The ICD-9-CM procedure code 88.72 was reported by Mount Sinai hospital for reportable inpatient TTE performed before October 1, 2015. The calendar year 2014 sample contained 57,547 hospitalizations, among which we gated on the 8 admission diagnoses of interest and applied the same inclusion criteria as described earlier. The population was dichotomized by the performance or lack thereof inpatient echo, and demographic features were compared using Student t tests and chi-square tests as appropriate. We observed that the average age in the echo group was advanced 11 years beyond that of the nonecho group; this was markedly different from the age distribution of the NIS study sample. We attempted to correct for this disparity and more closely emulate the NIS study sample by including only patients who were ≥60 years of age. We then performed both univariate analysis and multivariate logistic regression adjusting for patients’ sex, race, insurance type, number of chronic diseases, administration of diagnostic cardiac catheterization or cardiac stress test, admission day, and APR-DRG severity index among each admission-diagnosis cohort to assess for an association between echo performance and hospital mortality rates.
Results were reported as adjusted odds ratios (ORs) with 95% confidence intervals (CIs). All statistical analyses incorporated primary sampling units and clusters to obtain national estimates and were performed using Statistical Analysis Software (SAS, Cary, North Carolina) version 9.3 with a 2-sided significance level set at p < 0.05.
This study was considered to be exempt research with respect to the requirement for informed consent, as assessed by the Mount Sinai Medical Center Institutional Review Board. All authors vouch for the accuracy and completeness of the data presented.
Analysis of trends
An estimated 7,669,000 echo examinations were performed in U.S. hospitals from 2001 to 2011. During this period, echo use grew at an average annual rate of 3.41% (Central Illustration, panel A). An evaluation of resource use and patients’ outcomes found that hospitalizations in which echo was coded as the primary procedure had declining rates of hospital mortality and length of stay. However, these trends were associated with rising hospitalization charges (Central Illustration, panels B to D). These trends motivated us to undertake a detailed analysis for investigating whether underuse of echo may be related with hospital outcomes and thus serve as a marker of clinical care.
Geographic echo distribution
The highest rates of echo use per hospitalization were observed throughout the East Coast (Figure 1). On a state-by-state basis, Rhode Island, New York, and Montana were the leading inpatient echo users, whereas the lowest rates were observed in Wyoming, Alaska, and New Mexico.
2010 nationwide inpatient sample and sample demographics
Table 1 displays the 2010 NIS and the study population stratified by echo use and categorical variables. An estimated 816,500 echo examinations were performed nationally in 2010, corresponding to 2.09% of all hospital admissions for that year. As expected, the proportion of hospitalizations in which echo was performed was significantly higher within our sample (7.01%; p < 0.001). The distributions observed within categorical variables were similar between the echo and nonecho arms of our study population. However, significant differences were observed across all domains, barring use of diagnostic cardiac catheterization. In a comparison of the 2 sample arms, patients undergoing echo were younger (p < 0.001), received more cardiac stress tests (p < 0.001), resided in higher-income zip codes (p < 0.001), were at higher risk of death and morbidity (p < 0.001), and had more chronic medical problems (p < 0.001) (Table 1).
Sample echo rates
The prevalence of echo use in the admission-diagnosis cohorts was as follows: valvular disease (22%), acute cerebrovascular disease (12%), cardiac dysrhythmia (10%), CHF (7%), AMI (6%), CAD (6%), sepsis (5%), and nonspecific chest pain (4%) (Figure 2). To explore the surprising low echo rates reported in patients with AMI and CAD, we scrutinized the uses of parallel imaging modalities capable of providing a determination of ventricular function. Rates of both cardiac stress testing and cardiac magnetic resonance imaging were negligibly low, and cardiac computed tomography documentation was variable. However, as expected, a large proportion of the AMI and CAD cohorts underwent diagnostic coronary catheterization (64% and 62%, respectively) (Figure 2).
Multivariate logistic regression adjusting for key variables (age, sex, race, insurance, income quartile by zip code, number of chronic diseases, administration of diagnostic cardiac catheterization, and administration of cardiac stress testing, admission day, and APR-DRG severity score) within each admission-diagnosis cohort was performed to evaluate for association between echo use and the odds of a patient’s death. We found that the use of echo was associated with significantly lower odds of all-cause hospital mortality in AMI (adjusted OR: 0.74; 95% CI: 0.63 to 0.86; p < 0.001), cardiac dysrhythmia (adjusted OR: 0.72; 95% CI: 0.55 to 0.94; p = 0.02), acute cerebrovascular disease (adjusted OR: 0.36; 95% CI: 0.31 to 0.42; p < 0.001), CHF (OR: 0.82; 95% CI: 0.72 to 0.94; p = 0.005), and sepsis (adjusted OR: 0.77; 95% CI: 0.70 to 0.85; p < 0.001). Significance was not observed in the subsamples for CAD (adjusted OR: 0.66; 95% CI: 0.40 to 1.08; p = 0.09), valvular disease (adjusted OR: 0.71; 95% CI: 0.40 to 1.25; p = 0.24), or nonspecific chest pain (adjusted OR: 0.75; 95% CI: 0.18 to 3.15; p = 0.7) (Figure 3).
Single-center validation analysis
Between 2003 and 2014, the incidence of echo use per hospitalization at the Mount Sinai Medical Center steadily increased at an average annual rate of 4.75%. Echo was performed during 14.6% of the 57,547 hospitalizations in 2014. Of these hospitalizations, 18,401 were for 1 of the 8 admission diagnoses of interest and met the inclusion criteria previously described. Within this population we observed that the patients undergoing echo were older (p < 0.001), more likely to receive diagnostic coronary catheterization or stress testing (p < 0.001), and more critically ill (p < 0.001) (Online Table 1).
In our analysis of imaging practices in our home institution we observed underuse of echo for indications identified in the NIS database (Table 2), most notably among patients with AMI, 25% of whom were discharged without echo evaluation and 17% of whom were discharged without any form of cardiac imaging. Death occurred in 9.4% of patients admitted for AMI, and the mortality rate was marginally higher in those patients in which echo was not performed (10.1% vs. 9.1%), although this difference was not statistically significant (p = 0.14).
The association between echo performance and the odds of inpatient mortality did not reach significance in 6 of the remaining 7 admission diagnoses. Notably, however, 42% of patients who died during hospitalization with reasons related to their primary diagnosis did not undergo an echo examination before death (Table 2).
The cumulative use of parallel imaging modalities capable of providing an assessment of cardiac function other than echo (contrast ventriculogram, cardiac computed tomography, radionuclide ventriculogram, cardiac magnetic resonance imaging, pharmacological and nonpharmacological stress testing) per diagnosis cohort were similar to our findings nationally (Online Table 2). The reason for the lack of echo in these patients, however, could not be established during the analysis.
Widespread availability and familiarity with echo have led to concern for its potential overuse. Recently, investigators have reported that echo use nearly doubled between 1999 and 2008 (6), constituting approximately one-half of all cardiac imaging services among Medicare beneficiaries (5). Interestingly, however, our analysis of the largest publicly available all-payer inpatient database in the United States suggests that echo maybe underused for common and appropriate indications.
National echo use
Between 2001 and 2011, the absolute volume and incidence of echo per hospitalization steadily increased at average annual rates of 3.41% and 3.04%, respectively, less than one-half the average annual growth rate described among Medicare beneficiaries from 1999 to 2004 by Pearlman et al. (5). This discrepancy likely reflects differences in study populations because our demographic was specific to the inpatient setting and pertained to an all-payer population. This finding implies high rates of echo use among Medicare beneficiaries in the outpatient domain.
Overall, the estimated national incidence of echo performance per hospitalization was 2.09% in 2010, which is consistent with an independent assessment of the NIS database reported previously (12). The geographic distribution of echo use was heterogeneous, with the highest rates observed on the East Coast. These results are consistent with previous reports of echo use among Medicare beneficiaries in 1995 by Lucas et al (13). Although our analysis did not lead to a definitive explanation for regional variation in echo use, we speculate that this variation may reflect geoeconomic disparities in access to health care, previously described geographic variation in diagnostic practices, and elevated age-adjusted major cardiovascular death rates in the region (14,15).
Mortality rates in patients undergoing echo
In 2010, approximately one-half of all inpatient echo examinations could be identified in the NIS database during hospital admissions for AMI, cardiac dysrhythmia, acute cerebrovascular disease, CHF, CAD, sepsis, valvular disease, and nonspecific chest pain. Adjusting for key variables, we observed that the use of echo was associated with lower odds of hospital mortality among 5 of these disease processes. As the least expensive and lowest-risk cardiac imaging modality available (16), echo is a first-tier diagnostic tool that routinely leads to the initiation of various therapeutic interventions such as heart failure regimens, use of implantable devices, surgery, and revascularization, all of which have been shown to improve mortality rates in the appropriate clinical setting (17). Although this study was not designed to determine a causal link between echo and patients’ outcomes, we believe that the association observed here is the product of information provided by echo that assists physicians in management decisions and patient risk stratification. This hypothesis is supported by a study conducted in a major academic medical center, which found that 32% of inpatient echo examinations led to an active change in medical care (3). Extrapolation of this intervention rate could explain the positive association observed in our analysis.
The argument could be made that patients treated in larger medical centers equipped with comprehensive ancillary and diagnostic services would be more likely to receive an echo examination and perhaps have access to more expert physicians. This position would impose a significant bias toward echo use and improved patient outcomes. For this reason we find it important to convey that there was no difference in the distribution of echo use among hospitalizations in small, medium, or large medical centers for the admission diagnoses studied.
Diagnosis-specific echo rates
The use of echo for the initial evaluation and re-evaluation of ventricular function after acute coronary syndrome is in accordance with current standards of care and is supported by appropriate use criteria (1). However in 2010, echo was identified in only 6.2% of our AMI cohort. Within this cohort, 64% underwent diagnostic coronary catheterization, and assuming that ejection fraction was approximated by contrast ventriculography, an assessment of cardiac function could not be substantiated in roughly 30% of this group. Although low rates of echo use in the NIS database may reflect underreporting because echo may not be required for reimbursement by DRG assignments (non-DRG procedures), the data continue to highlight underuse of functional cardiac assessment, as reported in previous investigations. A population-based study of 2,317 patients hospitalized with myocardial infarction from the Mayo Clinic in Minnesota reported that the use of echo within the first 30 days after AMI increased from 14% in 1979 to 65% by 1986 (18). Another study from Olmsted County, Minnesota compiled features of 791 patients who developed heart failure after AMI and reported that left ventricular function was not assessed after the diagnosis of heart failure in 38% of patients (19). More recently, data from the Worcester Heart Attack Study reported that in 2003, echo was used in slightly more than 50% of patients hospitalized for AMI and that ejection fraction was not assessed among 27% of patients before discharge (20).
We attempted to investigate AMI-related cardiac imaging practices in our home institution. In our sample of 277 admissions to the Mount Sinai Medical Center in 2014 for AMI, we found that 75% of these patients underwent echo, and 41% underwent another mode of cardiac imaging, primarily coronary artery catheterization. We found that 17% of patients admitted for AMI were discharged without an assessment of cardiac function. Of these 277 hospitalizations, 26 were associated with the patient’s death (9.4%), and among those who died, cardiac imaging was not performed in 27%.
Similarly, the appropriate use criteria support the re-evaluation of cardiac function by echo among patients with heart failure, arrhythmia, and known valvular disease in the event of a change in clinical status without a clear precipitant (1). In our nationally representative sample, rates of echo use among patients requiring hospitalization with these admission diagnoses ranged from 7% to 22%.
We confirmed our observation that echo is underused on the national level by demonstrating the underuse of echo in a tertiary care referral center located in one of the most diagnostically advanced regions in the United States. However, multivariate mortality analysis of our single-center cohort did not reach statistical significance.
Low rates of echo performance observed nationally may in part be explained by the diagnostic stewardship of physicians or by cost bundling agreements between insurance companies and hospitals that place extra incentives to shift inpatient procedures into the post-discharge outpatient domain.
Considerable caution is required in interpreting the results of the study. Information in the NIS is generated from discharge abstracts designed to facilitate hospital payment. These abstracts customarily are prepared by trained coding specialists after reviewing inpatient medical records to capture provider care and services that warrant reimbursement. Just as in any other administrative database, the NIS data are subject to potential oversights in documentation and coding. The low use of echo in the NIS database may simply represent institutional underreporting of ICD-9-CM procedure coding in the inpatient claim data. Because the database comprises survey-level data of de-identified discharge information, coding inaccuracy, coding precision, and readmission data cannot be corrected for. Indications for echo administration are not recorded in the NIS, so admission diagnosis was used as a surrogate. The NIS does not provide information about survival beyond the inpatient period; hence, longer-term outcomes are unknown. Finally, the sample size of our single-center validation cohort was underpowered to demonstrate a significance association between echo and mortality; larger prospective studies are needed.
Echo use was associated with decreased odds of hospital mortality among 5 of the leading 6 admission diagnoses for which echo was most commonly reported in the 2010 NIS database. These 5 diagnoses account for approximately 3.7 million national hospitalizations annually; however, in 2010 the NIS database reported echo use in only 8% of cases. Because patient selection and appropriate echo use are key to cost efficiency, this study suggests that echo may be underused during critical cardiovascular hospitalizations, most notably in the treatment of AMI.
COMPETENCY IN SYSTEMS-BASED PRACTICE: Patients hospitalized with myocardial infarction, heart failure, cardiac dysrhythmia, acute cerebrovascular ischemia, or sepsis who underwent echo faced a lower likelihood of inpatient mortality than did patients without echo.
TRANSLATIONAL OUTLOOK: Further studies are required to clarify the links between access to echo and clinical outcomes among hospitalized patients in the United States.
Dr. Narula has received research grants to the institution from Philips Healthcare and GE Healthcare in the form of equipment. Dr. Sengupta is an advisor to Saffron Technology Hearts Labs; and is a consultant to Edwards Lifesciences. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose. Brian Griffin, MD, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- acute myocardial infarction
- All Patient Refined–Diagnosis Related Group Severity Score
- coronary artery disease
- Clinical Classification Software
- congestive heart failure
- International Classification of Disease-9th edition-Clinical Modification
- Nationwide Inpatient Sample
- odds ratio
- transesophageal echocardiography
- transthoracic echocardiography
- Received August 28, 2015.
- Revision received October 7, 2015.
- Accepted October 8, 2015.
- American College of Cardiology Foundation
- Douglas P.S.,
- Khandheria B.,
- Stainback R.F.,
- et al.
- Ballo P.,
- Bandini F.,
- Capecchi I.,
- et al.
- Pearlman A.S.,
- Ryan T.,
- Picard M.H.,
- Douglas P.S.
- Andrus B.W.,
- Welch H.G.
- ↵Healthcare Cost and Utilization Project. Facts and Figures: Statistics on Hospital-Based Care in the United States 2010. Available at: https://www.hcup-us.ahrq.gov/db/nation/nis/NIS_Introduction_2010.jsp. Accessed September 20, 2014.
- ↵Whalen D, Houchens R, Elixhauser A. 2004 HCUP Nationwide Inpatient Sample (NIS) Comparison Report. HCUP Methods Series Report #2007-03. Online February 2, 2007. U.S. Agency for Healthcare Research and Quality. Available at: https://www.hcup-us.ahrq.gov/db/nation/nis/reports/2004niscomparisonrpt.jsp. Accessed September 20, 2014.
- ↵Healthcare Cost and Utilization Project. Agency for Healthcare Research and Quality, Rockville, MD. 1999–2011. Available at: http://hcupnet.ahrq.gov/HCUPnet.jsp?Id=DB97390A368039A9&Form=SelLAY&JS=Y&Action=%3E%3ENext%3E%3E&_LAY=Researcher. Accessed September 20, 2014.
- ↵Averill R, Goldfield N, Hughes J, et al. 3M APR-DRG Classification System Version 26.1 (effective 10/01.2008) Methodology Overview. 3M Health Information Systems. Wallingford, CT. 2008. Available at: http://www.hcup-us.ahrq.gov/db/nation/nis/v261_aprdrg_meth_ovrview.pdf. Accessed September 20, 2014.
- ↵Pfuntner A, Wier LM, Stocks C. Most Frequent Procedures Performed in U.S. Hospitals, 2010: Statistical Brief #149. In: Healthcare Cost and Utilization Project (HCUP) Statistical Briefs [Internet]. Available at: http://www.ncbi.nlm.nih.gov/books/NBK132428/. Accessed November 8, 2015.
- Go A.,
- Mozaffarian D.,
- Roger V.,
- et al.
- Ferrari V.A.,
- Whitman B.,
- Blankenship J.C.,
- et al.
- Yancy C.,
- Jessup M.,
- Bozkurt B.,
- et al.