Hospital Variation in the Utilization of Short-Term Nondurable Mechanical Circulatory Support in Myocardial Infarction Complicated by Cardiogenic Shock

Introduction

Cardiogenic shock (CS), the inability to maintain adequate perfusion because of failure of the pumping mechanism, is present in 3.5% of patients presenting with acute heart failure and 6.6% in those presenting with myocardial infarction (MI), and is associated with an in-hospital mortality rate as high as 70%.^1,2 The recent availability of temporary mechanical circulatory support (MCS) has led to an expansion of the therapeutic options for CS, allowing short-term circulatory support as a bridge to recovery, transplantation, or placement of a durable left ventricular assist device in selected patients.³ Despite increasing use of MCS devices overall, rises in CS hospitalization rates have outpaced growth in MCS use.⁴ Reasons for this discrepancy are unclear and may relate to changes in patient or disease characteristics or the inability of some hospitals to develop or augment capacity to place MCS. Additionally, few randomized trials exist evaluating the use of nondurable MCS in shock patients, leading to uncertainty in its true clinical benefit.

Given differences in hospital capabilities to use MCS, as well as uncertainty in the current evidence supporting its use, we hypothesized that significant variation between hospitals may exist in the propensity to utilize MCS for treatment of MI complicated by CS. We therefore sought to (1) assess variation in MCS usage between hospitals in the United States among hospitalizations for MI complicated by CS, (2) quantify the variability between hospitals in MCS use for MI complicated by shock, and (3) assess whether those hospitals with the highest rates of MCS use had different outcomes compared with those with the lowest use.

Methods

All data and materials are publicly available through the Healthcare Cost and Utilization Project (HCUP) and can be accessed at http://www.hcup-us.ahrq.gov/databases.jsp.

Study Population

All hospitalizations of adults ≥18 years old with diagnoses of CS (International Classification of Disease, Clinical Modification, version 9 [ICD-9-CM] code 785.51) with MI (ICD-9-CM 410.11–410.91) in any position in the National Inpatient Sample (NIS), January 1,2014, to December 31, 2014, were included.^5,6 Developed through the HCUP, the NIS contains patient-level claims on 35 million hospitalizations nationally, representing a 20% sample of all hospital discharges.⁷ The NIS is publicly available for researchers through the HCUP website.⁷ Only 1 year was evaluated as the hospitals in the NIS change annually.⁷ Hospitals performing <10 percutaneous coronary interventions (PCIs) per year were excluded from the primary analysis to restrict the analysis to hospitals with cardiac catheterization laboratories, as hospitals without the capacity to perform PCI are likely unable to have the capacity to place MCS devices. To confirm the results in a database with more complete sampling, analyses were repeated in the 2015 Nationwide Readmissions Database (NRD), a database containing an 85% sample of inpatient claims from 27 states, representing 36 million discharges, roughly 56.6% of all US hospitalizations.⁷ Only January 1, 2015, to September 30, 2015, were analyzed in the NRD because of the transition to the International Classification of Diseases, Tenth Revision, Clinical Modification/Procedure Coding System (ICD-10-CM) on October 1, 2015. Because of extensive quality control by HCUP,⁷ there was minimal missing data in this analysis. Studies based on the NIS and NRD are exempt from Institutional Review Board approval at Beth Israel Deaconess Medical Center.

Exposures and Outcomes

The primary exposure of interest was receipt of MCS, defined as percutaneous and nonpercutaneous devices, intra-aortic balloon pump, and extracorporeal membrane oxygenation. Percutaneous and nonpercutaneous devices (nondurable MCS) were defined using ICD-9-CM procedure codes 37.68 (percutaneous) or 37.60 and 37.65 (nonpercutaneous). These codes would typically include percutaneous devices such as TandemHeart (CardiacAssist Inc, Pittsburgh, PA) and Impella devices (Abiomed Inc, Danvers, MA) and nonpercutaneous devices, including Thoratec paracorporeal ventricular assist device (Thoratec Inc, Pleasanton, CA), AB5000 (Abiomed Inc, Danvers, MA), BVS 5000 (Abiomed Inc, Danvers, MA), and CentriMag (Thoratec Inc, Pleasanton, CA).³ Intra-aortic balloon counterpulsation (IABP) was defined using ICD-9-CM procedure code 37.61, extracorporeal membrane oxygenation (ECMO) using ICD-9-CM procedure code 39.65 (central ECMO), and percutaneous cardiopulmonary support (peripheral ECMO) using code 39.66. Permanent ventricular assist devices were excluded (ICD-9-CM codes 37.52 and 37.66). The primary outcome was each hospital’s MCS ratio, defined as the proportion of hospitalizations for MI with CS that included use of MCS.

Covariates

Patient-level covariates included demographics and 29 Elixhauser clinical and procedural variables, determined at index hospitalization (Table I in the Data Supplement).⁸ Elixhauser Comorbidity measures were designed to predict in-hospital mortality using administrative data.^8–11 Additionally, 5 hospital level covariates were available in the NIS, determined from the American Hospital Association Annual Survey of Hospitals. These included hospital bed-size, ownership, location/teaching status, region, and total discharges. A hospital was considered a teaching hospital if it had an American Medical Association approved residency program, was a member of the Council of Teaching Hospitals, or had a ratio of full-time equivalent interns and residents to beds of 0.25 or higher.⁷

Statistical Analysis

The distribution of hospital MCS ratios were plotted using a histogram. Patient-level characteristics, rates of coronary angiography, revascularization, right heart catheterization, and in-hospital mortality of hospitalizations for CS with MI were compared across quartiles of MCS ratio via 1-way ANOVA for continuous variables and χ² tests for categorical variables. Similarly, hospital characteristics were compared across quartiles of MCS usage with ANOVA for continuous variables and χ² tests for categorical variables. Given the large sample size, the normality assumptions for ANOVA were met based on the central-limit theorem and the Welch ANOVA test was used which does not assume equality of variances. A multivariable linear mixed effects model with a logit link was used to model receipt of MCS according to patient and hospital characteristics, using all variables, accounting for the complex sampling design of the NIS. Hospital was included as a random effect to account for clustering of CS hospitalizations by hospital. The variance of the random effect for hospital was used to create median odds ratios (ORs), adjusted for all patient level variables. The median OR represents the ratio of the likelihood of an identical patient receiving MCS at 1 randomly selected hospital versus another.

Rates of in-hospital mortality were compared across quartiles of MCS use with a χ² test. A multivariable mixed effects model with hospital site as a random effect was used to evaluate whether quartile of MCS was associated with in-hospital mortality, and the Wald test used to test significance. Additionally, we examined whether analyzing the effect of the MCS ratio on in-hospital mortality in the primary analysis as a continuous rather than categorical variable would alter results.

Finally, because small hospitals with few hospitalizations for MI with CS could have less stable MCS ratios, we repeated the analyses considering only hospitals performing ≥10 PCIs and having ≥10 hospitalizations for CS and MI to identify if results differed. Moreover, as several covariates could also be outcomes, we repeated the model for in-hospital mortality, restricting the analysis to covariates likely to be present on admission to identify if results differed. Additionally, we repeated the analysis excluding transfers into acute care hospitals as these patients may represent a different population. Finally, analyses were repeated in the 2015 NRD to test the sensitivity of the findings to examination in a fully sampled population. Because of the HCUP privacy rules, values <10 are suppressed. All analyses were done with SAS version 9.4 (SAS Institute Inc, Cary, NC) at a 2-tailed level of significance of P<0.05.

Results

Unadjusted Hospital Characteristics

A total of 1813 hospitals had at least 1 discharge for MI with CS, of which 1440 (79.4%) performed ≥10 PCIs per year. Of the 1440 hospitals included in the analytic cohort, 1064 (73.9%) had at least 1 record of MCS usage. The hospital median rate of MCS use for MI with CS was 33.3%; interquartile range (IQR) 0.0% to 50.0% (Figure). Overall, 41% of hospitals admitting MI with CS patients did not perform MCS. Eighty-four hospitals (5.8%) included in the analysis performed MCS in all hospitalizations for MI with CS. The median (IQR) MCS ratio by quartile was 0.0% (0.0%–0.0%) in Q1, 22.2% (16.7%–25.0%) in Q2, 42.9% (33.3%–50.0%) in Q3, and 67.7% (60.6%–100.0%) in Q4 (Table 1). Compared with hospitals in the lowest quartile, hospitals in the highest quartile of MCS use were larger (Q4 versus Q1; percentage classified as large [>45 to >350 beds depending on region, location, and teaching status]; 128 [47.1%] versus 136 [36.2%]; P<0.001) and had a greater number of discharges (Q4 versus Q1; median [IQR] annual discharges; 2900 [1779–4522] versus 2026 [1456–2930]; P<0.001). Additionally, more than half of MCS high utilizing hospitals were urban teaching hospitals compared with less than half of low utilizing hospitals (Q4 versus Q1; 171 [62.9%] versus 175 [46.5%]; P<0.001). High-utilizing MCS hospitals were less likely to be rural (Q4 versus Q1; 16 [5.9%] versus 45 [12.0%]; P<0.001) or urban nonteaching hospitals (Q4 versus Q1; 85 [31.3%] versus 156 [41.5%]; P<0.001). MCS utilization was numerically highest in the South (mean proportion of MCS utilizing hospitals: 37.8%) and least common in the Northeast (mean proportion of MCS utilizing hospitals: 14.3%).

• Download figure
• Download PowerPoint

Patient Characteristics by Quartile of Hospital MCS Use

There were a total of 10 438 hospitalizations for shock and MI of which 9129 (87.5%) were included in the primary analysis (1312 [12.6%] excluded because of hospitalization at sites performing <10 PCIs per year [N=730], resulting in transfer to an acute care hospital [N=579], or missing [N<10]). Patient-level characteristics of all hospitalizations for MI with CS by receipt of MCS are listed in Table II in the Data Supplement. Of these 9129 hospitalizations for shock and MI, only 1 MCS device was placed in 2972 (32.6%), 2 devices in 157 (1.7%), and ≥3 devices in <10. Of hospitalizations for MI with CS, none (0.0% [0/937]) received MCS in Q1, 725 (21.8% [725/3328]) received MCS in Q2, 1374 (41.8% [1374/3284]) received MCS in Q3, and 1035 (65.5% [1035/1580]) received MCS in Q4 (P<0.001; Table 2). Hospitalizations in high-utilizing hospitals were for patients who were younger (Q4 versus Q1; mean [SD] age in years; 67.9 [13.0] versus 71.5 [13.0]; P<0.001), less frequently admitted from the Emergency Department (Q4 versus Q1; 945 [59.8%] versus 741 [79.1%]; P<0.001), and more likely to be privately insured (Q4 versus Q1; 362 [22.9%] versus 149 [15.9%]; P=0.02). Hospitalizations at high utilizing hospitals were less likely to include a diagnosis of heart failure, renal failure, chronic pulmonary disease, pulmonary circulatory disorders, other neurological disorders, weight loss, and fluid/electrolyte disorders but were more likely to have coagulopathy (Q4 versus Q1: 347 [22.0%] versus 129 [13.8%]). Other comorbidities were not significantly different between quartiles of MCS use.

Hospitalizations in high-utilizing hospitals were significantly more likely to include procedures (Q4 versus Q1; coronary angiography: 1165 [73.7%] versus 440 [47.0%]; PCI: 822 [52.0%] versus 335 [35.8%]; coronary artery bypass grafting: 309 [19.6%] versus 38 [4.1%]; right heart catheterization: 93 [5.9%] versus 13 [1.4%]; P all <0.001; Table 3). While IABP was most common overall (85.1% of all MCS use) patterns of device subtype use did not differ by quartile of MCS use (Table 4). While IABP use was consistent across quartiles of PCI utilization, nonIABP MCS use was positively associated with hospital PCI volume (P<0.001; Table III in the Data Supplement)

Predictors of MCS Use

Detailed ORs, 95% CIs, and P value for patient and hospital level factors associated with MCS utilization are listed in Table IV in the Data Supplement. Age, renal failure, chronic obstructive pulmonary disease, and neurological disorders were associated with decreased receipt of MCS, whereas receipt of coronary angiography, PCI, coronary artery bypass grafting, and right heart catheterization were associated with an increased likelihood of MCS receipt. Hospital ownership was predictive of MCS use (P=0.02), with hospitalization at government institutions being less likely to involve MCS (OR 0.72 compared with private hospitals; 95% CI, 0.55–0.93; P=0.01). There was substantial variation between hospitals in MCS use after adjustment for patient factors (adjusted median OR of receiving MCS at 2 randomly selected hospitals: 1.58; 95% CI, 1.45–1.70)

In-Hospital Mortality

Among hospitals performing ≥10 PCIs per year, crude in-hospital mortality ranged from 38.6% to 46% and decreased according to quartile of MCS use (P<0.001; Table 5 and 6). After adjustment for patient, hospital, procedural variables, and device subtype, however, there was no significant relationship between hospital MCS utilization and in-hospital mortality (P=0.78; Table V in the Data Supplement). The MCS ratio as a continuous measure was not associated with adjusted in-hospital mortality (P=0.68).

Sensitivity Analysis

Among hospitals excluded from the primary analysis because of performing <10 PCIs per year, no MCS devices were used and 155 (21.1%) of hospitalizations for MI with CS resulted in transfer to an acute care facility. Hospitals excluded from the primary analysis had a low rate of procedures (coronary angiography: 63 [8.6%]; PCI: 20 [2.7%]; coronary artery bypass grafting: <10; right heart catheterization: <10). The overall in-hospital mortality for hospitalizations for CS and MI was 43.6% (N=320) among hospitals performing <10 PCIs per year.

When limiting the population to hospitals with both ≥10 PCIs per year and ≥10 cases of MI with shock per year, median MCS utilization rates remained 36.4% (IQR 25.0%–46.7%) with significant unexplained variation (adjusted median OR of receiving MCS at 2 randomly selected hospitals: 1.57; 95% CI, 1.41–1.72; Figure I in the Data Supplement). Both unadjusted and adjusted in-hospital mortality rates were not significantly different between quartiles of MCS use (adjusted P=0.78; Tables 6 and 7 in the Data Supplement). Restricting the analysis to variables likely to be present on only admission (ie, excluding heart failure, chronic pulmonary disease, coagulopathy, paralysis, other neurological disorder, and fluid and electrolyte disorders), results were again unchanged, with neither quartile of MCS ratio (P=0.85) nor MCS ratio (P=0.83) was associated with in-hospital mortality. Additionally, excluding transfers in to an acute care hospital did not change the relationship of MCS use and mortality (quartile of MCS ratio: P=0.14; MCS ratio: P=0.52). When these analyses were repeated in the NRD (85% sample), the findings were consistent: neither quartile of MCS use (P=0.33) or MCS ratio (P=0.23) were associated with adjusted in-hospital mortality in separate models. A total of 38.7% of hospitals admitting patients with CS with MI in the NRD did not use MCS, with median (IQR) use of 30.0% (17.6%–62.5%; Table VIII and Figure II in the Data Supplement).

Discussion

While recent advances in short-term nondurable MCS devices have made cardiopulmonary support more widely available to patients with MI complicated by CS, utilization of MCS nationally has lagged behind rates of CS.⁴ Our study demonstrates wide variation in MCS use CS and MI between hospitals, unexplained by patient factors. IABP use represented the vast majority of MCS devices. Overall MCS usage rates were under 50% in the majority of hospitals admitting shock patients. Although crude in-hospital mortality was lowest in high MCS utilizing hospitals, this was no longer evident after adjustment for patient and hospital factors.

There is a paucity of data on hospital variation in MCS use for patients presenting with MI and shock. Prior studies have confirmed significant hospital variation in IABP use for patients undergoing high-risk PCI and coronary artery bypass grafting,^12,13 as well as ECMO use in the setting of shock,^14–16 but information on hospital variation among all subtypes of MCS use is limited, as is literature on the extent to which the observed variation is related to patient and hospital characteristics. In a prior study of only individuals receiving PCI in the National Cardiovascular Data Registry, Sandhu et al¹⁷ reported that nonIABP MCS utilization was <5% for over half of National Cardiovascular Data Registry participating hospitals and >20% in less than one-tenth of all hospitals. As the National Cardiovascular Data Registry is limited to participating centers and those with cardiac catheterization laboratories, it is limited in its assessment of overall national trends in CS care.¹⁷ Our study found that 41% of hospitals nationally and 26.1% of PCI-capable hospitals who admitted patients with CS did not implant MCS devices. While these hospitals may admit less severe forms of CS, not requiring MCS, the high observed in-hospital mortality rate for CS and MI among these hospitals argues otherwise. These findings may also indicate the lack of capacity to provide such care, either because of lack of provider training or appropriate facilities, or uncertainty about the benefits of MCS in this setting.

At PCI-capable hospitals, one-third of hospitalizations for MI with shock received MCS, with significant variation between hospitals and regions. Despite large observed variation in MCS use, after adjustment for patient and hospital characteristics, only hospital ownership was significantly different between high and low MCS utilizing hospitals, with government hospitals being less likely to use MCS, possibly owing to differences in insurance or practice characteristics. The continued predominance of IABP use despite multiple trials not demonstrating a mortality benefit^18–20 may reflect continued uncertainty about the data or a desire to help high-risk patients with whatever means possible.

While IABP represented 85.1% of MCS utilization, there was growth in ECMO and nondurable MCS use, though currently rates remain <9% even in the highest volume MCS hospitals. This low uptake may be because of concerns about high device cost, slow technology adoption, or skepticism about the efficacy of these devices for treatment of shock. As the volume of nonIABP MCS grows, it is reasonable to evaluate variation in use and outcomes separately from other forms of MCS. While IABP use was widespread before the publication of trials demonstrating lack of mortality benefit to IABP therapy in shock,^18–20 its relative use has declined.⁴ Whether or not the displacement of IABP by newer technology will result in improved outcomes is unclear this time. In the current study, in-hospital mortality was not different between quartiles of MCS use. Given the recently expanded indications for MCS use in nonMI shock and high-risk PCI,²¹ our study highlights the need for further investigation into the efficacy of MCS for CS and for clear evidence-based guidelines for initiation of MCS therapy in MI to improve outcomes and reduce inappropriate hospital variation. Given limited data supporting the use of IABP in MI and CS,^18–20 the observed wide variation in IABP utilization suggests the need for evidence-based guidelines on MCS initiation. While the European Society of Cardiology 2014 guidelines for revascularization list routine IABP use as a class III indication,²² the absence of such a recommendation in US guidelines could contribute to the observed variation in IABP use.

There are several limitations of the current study worth note. First, as the study was retrospective and limited in the number of variables ascertained, causality cannot be inferred, and less variability may exist in MCS use after incorporation of unmeasured variables. Second, the timing of MCS relative to other diagnoses noted on discharge or the development of MI and shock cannot be discerned, given limited granularity of the database, though a sensitivity analysis restricting the model to only variables likely to be present on admission did not identify any changes to the results. Third, inaccuracies in coding may reduce precision in the interpretation of results. Similarly, inaccuracy of coding for procedures or sampling variability may affect national estimates of cardiac procedural usage. Fourth, as the hospitals sampled in the NIS differ annually, the observed findings may not fully reflect national trends in other years. Fifth, as only inpatient data are available in the NIS, only in-hospital outcomes could be evaluated. Sixth, there was insufficient hospital level variation in subtypes of MCS to evaluate in-hospital outcomes according to device subtype. Seventh, as cost data in the NIS is extrapolated from hospital charges, no cost data are presented in the current manuscript.

Conclusions

There is significant variation between hospitals in the use of MCS for treatment of MI complicated by cardiogenic shock, with IABP use still most common. In-hospital mortality is not different between quartiles of MCS use after adjustment for patient and hospital factors. More outcomes data and detailed evidence-based guidelines on the optimal clinical conditions for MCS use are needed.

Footnotes